ArticlePDF Available

Social Cost of Carbon: A Closer Look at Uncertainty

Authors:
Authors
Thomas E Downing Stockholm Environment Institute, Oxford Office
David Anthoff Environmental Change Institute, Oxford
Ruth Butterfield Stockholm Environment Institute
Megan Ceronsky Environmental Change Institute, Oxford
Michael Grubb Imperial College
Jiehan Guo Environmental Change Institute, Oxford
Cameron Hepburn University of Oxford
Chris Hope University of Cambridge
Alistair Hunt Metroeconomica, Bath
Ada Li Environmental Change Institute, Oxford
Anil Markandya Metroeconomica, Bath
Scott Moss Centre for Policy Modelling, MMU
Anthony Nyong University of Jos
Richard S.J. Tol University of Hamburg
Paul Watkiss AEA Technology Environment
SEI Oxford Office
266 Banbury Road, Suite 193
Oxford
OX2 7DL
Tel: 01865 426316
Fax: 01865 421898
Tom.Downing@sei.se
Social Cost of Carbon: A Closer Look at Uncertainty
Final project report
November 2005
Final revision August 2005
Department for Environment, Food and Rural Affairs
Nobel House
17 Smith Square
London SW1P 3JR
Telephone 020 7238 6000
Website: www.defra.gov.uk
© Queen’s Printer and Controller of HMSO 2005
This document/publication is value added. If you wish to re-use, please apply
for a Click-Use Licence for value added material at
http://www.opsi.gov.uk/click-use/system/online/pLogin.asp. Alternatively
applications can be sent to:
Office of Public Sector Information
Information Policy Team
St Clements House
2-16 Colegate
Norwich
NR3 1BQ
Fax: 01603 723000
e-mail: HMSOlicensing@cabinet-office.x.gsi.gov.uk
Information about this publication and further copies are available from:
Environment Policy Economics
Defra
Ashdown House
123 Victoria Street
London SW1E 6DE
Tel: 020 7082 8593
This document is also available on the Defra website.
Published by the Department for Environment, Food and Rural Affairs
Social Cost of Carbon: A Closer Look at Uncertainty Page iii November 2005
Social Cost of Carbon: A Closer Look at Uncertainty
Uncertainty is inherent in estimates of the social cost of carbon (SCC). The UK Department for Environment,
Food and Rural Affairs (Defra) initiated this research to evaluate the sources of uncertainties, plausible ranges of
estimates of the SCC and areas for further research and assessment.
The analytical framework for the project is a risk assessment that brings together elements of uncertainty in
climate change and its impacts with uncertainties in economic valuation; both are related to the context of
decision making.
This review of uncertainty in estimates of the social cost of carbon is summarised in key messages:
Understanding of the social cost of carbon:
Our understanding of future climatic risks, spanning trends and surprises in the climate system,
exposure to impacts, and adaptive capacity, is improving, but knowledge of the costs of climate change
impacts is still poor.
The lack of adequate sectoral studies and understanding of local to regional interactions precludes
establishing a central estimate of the social cost of carbon with any confidence.
The balance of benefits and damages in the social cost of carbon shifts markedly over time, with net
damages increasing in later time periods. Estimates of the SCC are particularly sensitive to the choice
of discount rates and the temporal profile of net damages
Vulnerability and adaptation to climate change impacts are dynamic processes responding to climatic
signals, multiple stresses, and interactions among actors. Large scale impacts, such as migration, can be
triggered by relatively modest climate changes in vulnerable regions.
Uncertainty and risk:
Climate uncertainties and the climate sensitivity are key factors in larger estimates of the social cost of
carbon.
Uncertainties in coverage, sectoral assessments and regional processes are likely to be significant, but
are difficult to judge without further model development and inter-model comparison.
Decision variables such as the discount rate and equity weighting also are extremely important.
The range of estimates of the social cost of carbon:
Estimates of the social cost of carbon span at least three orders of magnitude, from 0 to over 1000 £/tC,
reflecting uncertainties in climate and impacts, coverage of sectors and extremes, and choices of decision
variables.
A lower benchmark of 35 £/tC is reasonable for a global decision context committed to reducing the
threat of dangerous climate change and includes a modest level of aversion to extreme risks, relatively
low discount rates and equity weighting.
An upper benchmark of the SCC for global policy contexts is more difficult to deduce from the present
state-of-the-art, but the risk of higher values for the social cost of carbon is significant.
Uncertainty in the social cost of
carbon: lines of evidence.
Significant improvement in
estimates of the SCC will require
well validated assessments at the
regional scale of the dynamic
processes of vulnerability and
adaptation. Partnerships among
researchers and stakeholders in
developing countries are essential.
Climate change
Economic valuation
Discount rates, Equity
wei
g
htin
g,
risk ambi
g
uit
y
Expert
knowledge
elicitation
Meta-analysis of
literature
Fund & Page
Catastrophic
changes Regional
model Decision
Framework
Social Cost of Carbon: A Closer Look at Uncertainty Page iv November 2005
Social Cost of Carbon: A Closer Look at Uncertainty Page v November 2005
Contents
1 Introduction ..................................................................................................................................... 1
2 A methodology based on risk assessment ....................................................................................... 5
3 Understanding the social cost of carbon........................................................................................ 11
4 Uncertainty and risk ...................................................................................................................... 27
5 The range of estimates................................................................................................................... 33
6 Further research and next steps ..................................................................................................... 37
7 Conclusion..................................................................................................................................... 41
8 References and bibliography ......................................................................................................... 43
Annex: Description of integrated assessment models FUND and PAGE ............................................. 53
Annex: Sensitivity of the SCC to extreme events, equity weighting, discounting, and risk and
ambiguity aversion ................................................................................................................................ 65
Annex: Understanding uncertainty through expert knowledge............................................................. 71
Annex: Regional studies of dynamic and multiple stresses .................................................................. 79
Social Cost of Carbon: A Closer Look at Uncertainty Page vi November 2005
Tables
Table 1. Objectives and achievements of the scoping uncertainty project............................................. 2
Table 2. Measures of central tendency.................................................................................................. 10
Table 3. Locating the literature in a risk assessment framework ......................................................... 13
Table 4. Distribution of the SCC from a meta-analysis of the literature, GBP2000/tC ........................ 15
Table 5. Sensitivity of FUND estimates for different climate sensitivities.......................................... 29
Table 6. Summary of the probability that the SCC exceeds a given threshold in FUND .................... 34
Table 7. Summary of FUND results, GBP2000 /tC reference, averages and standard deviation ......... 34
Table 8. Selected estimates of the SCC from experts for a range of scenarios, £/tC ........................... 34
Table 9. Key messages .......................................................................................................................... 42
Figures
Figure 1. An iterative framework for setting climate policy ................................................................... 3
Figure 2: Schematic mapping of multiple lines of evidence in understanding uncertainty in the social
cost of carbon. ......................................................................................................................................... 5
Figure 3. A risk assessment framework ................................................................................................. 6
Figure 4. Expert confidence in estimates of the social cost of carbon for three groups of experts based
on their overall expectation of the SCC ................................................................................................ 13
Figure 5. The composite cumulative density function of the marginal social cost of carbon. .............. 16
Figure 6. Probability distributions from FUND .................................................................................... 17
Figure 7. Distribution of the total climate damages ($2000) for the A2 reference scenario from the
PAGE model ......................................................................................................................................... 17
Figure 8. Comparison of distributions of estimates of the SCC........................................................... 20
Figure 9. Regional disaggregation of median estimates of the SCC in FUND .................................... 22
Figure 10. Sectoral disaggregation of median estimates of the SCC in FUND ................................... 22
Figure 11. Temporal profiles for regional estimates of the SCC in FUND........................................... 23
Figure 12. Temporal profiles for sectoral estimates of the SCC in FUND ........................................... 24
Figure 13. Expert responses to scenarios of the drivers of the SCC, £/tC............................................ 29
Figure 14. Distribution of the SCC in FUND with four climate sensitivities ...................................... 30
Figure 15. Planned developments in integrated assessment models .................................................... 37
Social Cost of Carbon: A Closer Look at Uncertainty Page vii November 2005
Annex Tables and Figures
Annex Table 1. Summary of FUND results, GBP2000 /tC quartiles................................................... 57
Annex Table 2. Summary of FUND results, GBP2000 /tC reference, averages and standard deviation
............................................................................................................................................................... 57
Annex Table 3. Summary of FUND results, GBP2000 /tC reference, trimmed and inter-quartile means
............................................................................................................................................................... 57
Annex Table 4. FUND sectors ............................................................................................................. 61
Annex Table 5. Regions in FUND ........................................................................................................ 62
Annex Table 6. Results for different climate sensitivities with Green Book discounting .................... 65
Annex Table 7. Results for different equity weighting with PRTP = 1%............................................ 66
Annex Table 8. Results for different discounting schemes.................................................................. 67
Annex Table 9. Categories and options for elicitation of uncertainty in estimates of the social cost of
carbon.................................................................................................................................................... 74
Annex Table 10. Percentage of responses greater than £35/tC for three groups of experts................. 77
Annex Table 11. Selected estimates of the SCC from experts for a range of scenarios, £/tC.............. 77
Annex Figure 1. Distribution of results from FUND, GBP2000/tC..................................................... 58
Annex Figure 2. Sectoral time profiles of the SCC from FUND for two discounting schemes, with and
without equity weighting....................................................................................................................... 59
Annex Figure 3. Regional time profiles of the SCC from FUND for two discounting schemes, with
and without equity weighting................................................................................................................ 60
Annex Figure 4. Results for different climate sensitivities with Green Book discounting .................. 66
Annex Figure 5. Challenges in expert judgment..................................................................................71
Annex Figure 6. Interface for the knowledge elicitation....................................................................... 74
Annex Figure 7. Range of estimates of the SCC from 14 experts........................................................ 75
Annex Figure 8. Expert confidence in estimates of the SCC for three groups of experts based on their
overall expectation of the SCC.............................................................................................................. 75
Annex Figure 9. Estimates of the SCC for each of 14 respondents ..................................................... 75
Annex Figure 10. Expert responses to scenarios of the drivers of the SCC, £/tC................................ 76
Annex Figure 11. Schematic representation of classification trees from scenarios of SCC estimates by
14 experts grouped according to low responses (left) and high responses (right)................................. 78
Annex Figure 12. Observed annual rainfall anomalies for three African regions, 1900-1998, and
model-simulated anomalies for 2000-2099........................................................................................... 81
Annex Figure 13. Representation of resources, stresses and migration developed by the authors ....... 85
Annex Figure 14. Farmers’ response options to ensure food security vis-à-vis alterations in land use
practice and land use patterns in the Sudan-Sahel region ..................................................................... 85
Annex Figure 15. Rural household decision-making, migration and the rural environment ............... 86
Social Cost of Carbon: A Closer Look at Uncertainty Page viii November 2005
Social Cost of Carbon: A Closer Look at Uncertainty Page 1 November 2005
Social Cost of Carbon: A Closer Look at Uncertainty
1 Introduction
The social cost of carbon (SCC) is the estimate of the cost of climate change damages—the net effects
of impacts on economies and societies of long term trends in climate conditions, including extreme
events, related to anthropogenic emission of greenhouse gases.1 Such estimates have been compiled
in order to aid consideration of greenhouse gas emission policies and to prioritise adaptation strategies
according to their potential effectiveness.
The UK Department for Environment, Food and Rural Affairs (Defra) initiated a project to evaluate
the range of uncertainties in estimates of the SCC; this report records the results of the project.
The scope of the assessment is described below and the methodology, based on a risk assessment
framework, is introduced in Chapter 2. The key messages are grouped according to:
o Understanding the SCC—the scientific basis for a risk assessment
o Uncertainties and risks—the interpretation of current estimates of the SCC
o The range of estimates—our conclusion regarding a robust range of estimates for
national and global policy on GHG mitigation
o Further research and next steps—pathways for further research
The conclusions reflect on the project and its importance in national and international climate policy.
Appendices provide more technical material on each component of the project.
The project team brought together a diverse group of experts and analysts:
Thomas E Downing, Stockholm Environment Institute, Oxford Office (Team Leader),
Geographer working on climate change impacts, vulnerability and adaptation, developed an
early social cost of carbon model (Open Framework) as part of the EC ExternE assessment
David Anthoff, MSc Environmental Change Institute, University of Oxford, analysed equity
weighting in FUND in his MSc, work extensively on converting FUND to Delphi and
analysing uncertainty
Ruth Butterfield, Agricultural Meteorology, Stockholm Environment Institute, Oxford,
reviewed regional syndromes of socially contingent effects of climate change
Megan Ceronsky, MSc, Environmental Change Institute, University of Oxford, analysed large
scale anomalies in FUND
Michael Grubb, Economist, Imperial College, contributed to literature review and evaluation of
the SCC in decision making
Jiehan Guo, MSc Environmental Change Institute, University of Oxford, analysis of
discounting schemes
Cameron Hepburn, Economist, University of Oxford, coordinated four MSc theses and
contributed to analysis of uncertainty in the SCC
Chris Hope, Policy Analyst, University of Cambridge, author of PAGE, assisted in design of
the expert elicitation and reviewed results from the project
Alistair Hunt, Economist, Metroeconomica, Bath, review of research programmes on the social
cost of carbon and economic valuation of health
Ada Li, MSc Environmental Change Institute, University of Oxford, investigation of risk and
ambiguity aversion
Anil Markandya, Economist, Metroeconomica, Bath, review of social cost of carbon estimates
1 The definition of the SCC is elaborated further in chapter 2.
Social Cost of Carbon: A Closer Look at Uncertainty Page 2 November 2005
Scott Moss, Economist, Centre for Policy Modelling, Manchester Metropolitan University,
developed model of regional syndrome of climate vulnerability in Sahel
Anthony Nyong, Geographer, Jos University, Nigeria, contributed to design and data inputs in
regional assessment of migration and climate change in the Sahel
Richard Tol, Economist, University of Hamburg, author of FUND, assisted MSc theses and
new analyses of FUND
Paul Watkiss, Environmental Policy group, AEA Technology Environment, reviewed literature
on SCC and use of estimates in decision making
Scope of the assessment
Defra commissioned two projects in 2004. A report by the team led by Paul Watkiss of AEA
Technology and Environment addresses the use of estimates of the social cost of carbon in decision
making (Watkiss et al. 2005). It draws upon the estimates of the SCC reported in the following
chapters, relating them to different decision frameworks and contexts in which the SCC might be or
should be considered.
The terms of reference of the ‘scoping uncertainty’ project focus on a review of the literature and
nature of uncertainty in estimates of the SCC (Table 1). The overarching aim was to consider whether
a consensus exists on the SCC, rather than produce an entirely new assessment. However, the project
also produced new estimates of the SCC, most notably through upgrading the FUND model to allow
full testing of parameter uncertainties and the analyses carried out in four MSc theses.
Table 1. Objectives and achievements of the scoping uncertainty project
Objectives Project achievements
Scoping and research design phase
To explore ways of improving the coverage of
SCC estimates, including new sectors, dynamic
processes of adaptation and low probability
catastrophic events
Annotated bibliography and review of SCC literature;
Locate existing studies and model results in the risk
matrix to evaluate coverage; Review existing estimates
from FUND and PAGE
To identify ongoing research programmes and
approaches for improving estimates of the SCC
Inventory research programmes; Document pathways for
further development of SCC estimates (see also the
project progress report at the completion of the scoping
phase)
Modelling and assessment phase
To explore the feasibility of empirical
improvements in the coverage of SCC estimate
Updated FUND with new work on health, tourism,
catastrophic events (collapse of the thermohaline
circulation, high climate sensitivities, large methane
releases); Reported new work from FUND on extreme
events; Prototype multi-agent model on Sahel
To incorporate the time varying discount rate
recommended by the Green Book
Incorporated a range of discounting methods in FUND
(MSc thesis) (also available in PAGE)
To explore how SCC estimates might vary over
time based on the above modeling work
Reviewed FUND results for regions and sectors (see
Annex); Data base of FUND results
To carry out a probabilistic sensitivity analysis on
key model parameters
Implemented full sensitivity testing with upgraded
FUND; Sensitivity analysis in PAGE; Knowledge
elicitation using formal methods (see Annex);
Exploration of non-linear responses using a multi-agent
model framework for the Sahel
To recommend a range of possible values for the
SCC based on the above analysis
Synthesis of estimates from FUND and PAGE, the
expert knowledge elicitation, and the exploratory multi-
agent modelling
Social Cost of Carbon: A Closer Look at Uncertainty Page 3 November 2005
Uncertainty is inherent in estimating the social cost of carbon. Experts disagree regarding the
appropriateness of cost benefit aggregations, the nature of quantifiable damages and the range of
resulting estimates of the SCC. The relevance of the decision context is noted in Ekins (1995), Grubb
(2003) and Pearce (2003), among many other commentaries on climate policy. The two extreme
views are (i) that the SCC should be part of a cost-benefit analysis with assumptions consistent across
current public policy and (ii) that climate policy is an issue of social justice and sustainability that
precludes calculation of a robust SCC estimate (and obviates the need to do so). The ‘economic’
argument follows a weak sustainability paradigm where net welfare is the measure of potential
damages, discounted to a net present value, with ‘winners’ and ‘losers’ aggregated at a global level.
The ‘social justice’ view is essentially a strong sustainability approach that cautions against trade-offs
between winners and losers in different impacts sectors, regions or societies.
A salient difference between the extremes is their view of decision making. An optimising, cost-
benefit analysis gathers the available information, recognising uncertainties, and makes a decision to
set policies for the foreseeable future. A more cautious approach suggests that policies in areas of
deep uncertainty such as climate change should be incremental—act based on the present
understanding of risks, learn about the consequences of those actions, then act again (Figure 1). An
incremental, iterative approach updates the expected SCC at each decision node. However, a low
estimate of the SCC might lead to mitigation targets that result in higher impacts, and subsequent
decisions will need to account for much higher estimates of the SCC. Or, setting targets for large
GHG reductions now, based on a ‘business as usual’ estimate of considerable climate impacts, would
stimulate technological innovations that would reduce the cost of achieving effective climate
stabilisation in the future (see Grubb et al. 2002). This sense of estimates of the SCC as part of a
decision-outcome feedback loop over the course of a century timescale may preclude calculating
robust estimates of the SCC at any single time step.
Figure 1. An iterative framework for setting climate policy
Decision nodes are represented by diamonds, with contextual considerations related to economic growth,
technology, governance, demographic change and scientific information. The consequences of the decision are
represented by outcomes (ovals) affecting greenhouse gas emissions, climate impacts and development status.
Source: After R. Richels, personal communication.
Social Cost of Carbon: A Closer Look at Uncertainty Page 4 November 2005
Acknowledgements
Funding for this work was from Defra. Some of the work reported here was carried out for the
European Commission Methodex project, an extension of the ExternE projects that produced early
estimates of the social cost of carbon associated with various fossil fuel power station emissions.
Draft results from the Defra project were reviewed at a workshop in September 2004 (see
www.socialcostofcarbon.aeat.com) and draft reports were reviewed by experts commissioned by
Defra. The project steering committee also provided helpful advice and comments. The authors are
grateful for the many insightful comments from these reviewers. We acknowledge that many of the
comments are inadequately addressed in the final report. Indeed, the wider debate regarding the SCC
is reflected in the disagreement of the reviewers with elements of this report. The authors remain
solely responsible for any oversights or errors in the report.
Social Cost of Carbon: A Closer Look at Uncertainty Page 5 November 2005
2 A methodology based on risk assessment
This chapter describes the risk assessment framework that the project developed and applied. The
first section explains the risk matrix of uncertainty in climate forcing and economic valuation of
impacts. Issues related to defining the social cost of carbon and uncertainty are described in the
second and third sections, including treatment of measures of the distribution of results and central
tendency in a data set. A note on units and conversion factors is included.
The methodology can be represented graphically as evaluating three sources of uncertainty, related to
climate change, valuation of impacts and parameters in the decision framework (Figure 2).
Uncertainty in climate forecasts are well documented, and the potential for catastrophic effects
appears to be increasing as scientists learn more about the climate system (see the conclusion of the
Exeter workshop on stabilising climate change; see Tirpak et al. 2005, www.stabilisation2005.com, ).
Economic valuation becomes increasingly speculative as the impacts move beyond market
commodities to non-market sectors, including effects on societies and economies. Decision
frameworks bound any policy assessment, implying different considerations for equity among present
populations and for future generations. For instance, a global decision maker might adopt a strong
sustainability framework based on social justice, while a local authority planner may have a more
constrained view based in strategic environmental assessment and best practice in land use planning.
The time profiles of climate change, its impacts and decision making are important. This report does
not focus on the temporal dimensions other than through the parameters of a decision framework
(most commonly the choice of a discount rate). The policy analysis report (Watkiss et al. 2005) looks
at time profiles of SCC values.
No single method, model or tool adequately captures all of these uncertainties. The complexity of
coupled socio-ecological system (climate change is driven in part by its impacts over time) and the
range of decision frameworks that might be employed in using the SCC imply that estimates of the
SCC will remain diverse and contentious. That is, there is little consensus regarding the central value
that should be adopted and relatively little confidence in the reliability of the evidence available upon
which SCC estimates can be made. In the language of decision sciences, the uncertainty is in the
realm of speculative estimates, often reflecting competing explanations.
Multiple lines of evidence are reviewed and employed in reaching our conclusions regarding
estimates of the SCC. The majority of the literature and models focus on trends in climate change
(e.g., in regional temperature) and market and non-market economic impacts. Such models as FUND
and PAGE also handle choices in the decision framework, such as discount rates and equity
weighting—the subject of four MSc theses in this project. Greater economic and climatic
uncertainties are covered through three exploratory methods: (i) a formal elicitation of expert
knowledge, (ii) an MSc thesis on catastrophic changes, and (iii) a prototype model of a region where
current trends in climate change could have widespread social and economic consequences, in this
case the Sahel region of West Africa.
Figure 2: Schematic
mapping of multiple lines of
evidence in understanding
uncertainty in the social cost
of carbon
The arrows on the three axes
imply increasing uncertainty,
although not necessarily larger
estimates of the SCC.
Climate change
Economic valuation
Discount rates, Equity
wei
g
htin
g,
risk ambi
g
uit
y
Expert
knowledge
elicitation
Meta-analysis of
literature
Fund & Page
Catastrophic
changes Regional
model Decision
Framework
Social Cost of Carbon: A Closer Look at Uncertainty Page 6 November 2005
Interpreting the SCC in a risk matrix
Framing of estimates of the SCC is organised as a matrix of confidence in projections of future
climate change and understanding of economic valuation (Figure 3). The climate axis ranges from
projections of global and regional temperature, to bounded scenarios of changes in precipitation and
risk of storms, to systemic, large scale changes such as collapse of the West Antarctic Ice Sheet, shift
in ocean circulations, or reversal of the biosphere carbon sink.
The corresponding economic axis begins with market sectors, with uncertainty expanding to the
valuation of non-market sectors such as coral reefs, and socially contingent feedbacks, such as
conflict over water, that exacerbate sectoral impacts or present non-marginal impacts at the local to
regional level. Note that socially contingent effects are a class of non-market impacts, where B might
be considered micro-economic effects and C includes macro-economic effects.
The gradient across the matrix, from top-left to bottom-right suggests an increase in uncertainty. The
larger scale climate changes are still speculative and often described as surprises outside the realm of
current global model predictions. The relative lack of studies of non-market and socially contingent
effects increases uncertainty in estimates of the SCC.
The gradient also reflects different timings of impacts—systemic changes in the global climate are
posited on a century time scale (e.g., collapse of the West Antarctic Ice Sheet); collapse of regional
societies and economies is not forecast in the next few decades (if at all). Some of the largest
uncertainties—such as release of methane hydrates—are events that are not fixed to a particular time
frame. On the other hand, impacts on market-based resources related to projections of temperature
and sea level rise may follow a relatively smooth profile over the next few decades (time profiles from
FUND are shown Chapter 3.
At present, the most commonly held assertion is that the net non-market and socially contingent costs
will be adverse (rather than benefits). However, there is insufficient evidence to suggest that the
gradient from upper-left to lower-right is necessarily a substantial increase in the total social cost of
carbon.
The axes and cells are described in qualitative terms below.
We use this framework to gauge progress in understanding the social cost of carbon in this report.
Note, that the matrix is not intended to be a sampling frame or to weight independent estimates. That
is, we do not attempt to derive probabilistic estimates of the SCC for each cell in the matrix and to
produce a final estimate based on the aggregation of such values.
Uncertainty in valuation
A. Market B. Non-market C. Socially
contingent
1. Projection
2. Bounded risks
Uncertainty in
Climate Change
3. System change and
surprise
Figure 3. A risk assessment framework
This is a simpler version of Figure 2, without the overlays related to decision choices.
Social Cost of Carbon: A Closer Look at Uncertainty Page 7 November 2005
The typical situations with each cell may help to illustrate both the range of issues inherent in
estimating the SCC as well as the role of the risk matrix.
For the column of impacts related to markets (A):
A1: Global and regional temperatures are projected to increase with relatively high
confidence. To the extent that warmer conditions would expand the area suitable for
agriculture, leading to climate impacts (in this case benefits) that are readily valued through
market exchanges (such as the price of major commodities, value of agricultural land, net
profit to producers or net benefit to consumers). Sea level rise is the other major climatic
element with high confidence, leading to impacts on coastal communities, loss of dryland and
wetland, forced migration, and the costs of coastal protection.
A2: Most climate elements are uncertain at the regional level, but current climate models
project changes within a reasonable range. Such bounded risks include increases or decreases
in precipitation, intensity and tracks of storms, and the frequency and magnitude of other
climatic extreme events (e.g., floods, droughts, lightning). The market impacts, for example
of drought on agriculture, can be estimated in principle although it is difficult to differentiate
between the effect of climate change and other stresses and responses that shape economic
outcomes. Current scenarios of climate change may underestimate drought risks, leading to a
possible bias toward short-term benefits of climate change for agriculture.
A3: System change and surprises are plausible climate outcomes that are not readily evaluated
in a probabilistic framework, such as a weakening of the thermohaline circulation, changes to
the phases of the major ocean-atmosphere modes (such as ENSO), the more extreme
scenarios of collapse of the West Antarctic Ice Sheet, large releases of methane hydrates or
reversals of the terrestrial carbon uptake. While the market effects can be described, the
impacts over large areas and time scales are not linear and therefore difficult to value in a
micro-economic framework. For example, what would be the (net) value of displacement of
all of the major world coastal cities due to a 3-5m sea level rise? (for example, see the results
of the Atlantis project, Lonsdale et al., 2005, Nicholls et al. 2005, Tol et al. 2005).
Effects on non-market sectors (B) are more difficult to value in that there are little empirical data on
how people in different countries and economic classes value amenities, species, landscapes and other
qualities of livelihoods. Contingent valuation based on willingness to pay or willingness to accept
principles give some guidance, but such values are often contentious and may not scale up from local
issues to the widespread effects of climate change. Examples of the sectors and issues in this column
are:
B1: Warmer temperatures and higher humidity—both projected to increase with some
confidence—will alter the amenity value of climates. In northern Europe, for instance, longer
and warmer summers will encourage more people to enjoy the outdoors and visit local tourist
destinations. On the other hand, a greater incidence of heat waves in southern Europe may be
problematic and losses in boreal and mountain ecosystems and winter tourism are likely.
B2: The bounded risks of changes in major cyclones, for instance, would affect coastal
ecosystems and agricultural land subject to increasing frequency and severity of coastal
flooding and salt water inundation. The value of species lost in local environments is difficult
to estimate.
B3: Catastrophic effects that lead to global losses of species are even more difficult to value,
not least because the impacts of climate change on global ecosystems and species biodiversity
is not well understood.
The socially contingent column (C) captures the secondary effects and multiple stresses of climate
change across a range of sectors. For instance, it is possible that reasonably small changes in climate
change could lead to significant impacts through multipliers (such as the effect of water shortages on
agriculture), high vulnerabilities (such as migration triggered by increased cyclone frequencies) and
behavioural responses to the risk (such as disinvestment from commercial agriculture in some regions
Social Cost of Carbon: A Closer Look at Uncertainty Page 8 November 2005
due to a perceived increase in drought risks). Such socially contingent effects are a sub-set of non-
market impacts. The mechanisms of such responses may not be readily captured in either micro-
economic valuations or macro-economic models. The range of potential values is likely to be
influenced by the decision framework—for instance whether potential liability for regional damages
is a motivation for a precautionary approach. Examples include:
C1: Projected changes in mean temperatures and sea level rise, at least over the next few
decades, are unlikely, on their own, to trigger significant socially contingent effects. The
exception may be snow melt and glacial lake outburst floods, significant in some regions.
C2: Changes in water cycles, along with drought and flood risks, are potential drivers of
regional migration, loss of an agricultural economy and crises for mega cities without reliable
water supplies. The extent of the world where such effects are most likely has not been
rigorously evaluated, but the Sahel and coastal deltas such as Bangladesh are frequently
mentioned. Regions of existing and exacerbated water scarcity could be subject to conflict.
C3: The displacement of entire cities due to extreme sea level rise is a good example of a
socially contingent effect with high uncertainty—in both the risks of climate change and in
the means to value such impacts. A case study of the potential impacts of and adaptation to a
5 meter sea level rise illustrates the issues (see Lonsdale et al., 2005, Nichols et al. 2005, and
Tol et al. 2005).
The risk matrix is a guide to understanding uncertainties in the social cost of carbon (taken up in the
next chapter). The risk matrix does not show explicitly three additional factors affecting uncertainty.
Two are mentioned above: (i) the role of decision frameworks and choice and (ii) the time profile of
climate change and its impacts.
The third factor (iii) concerns the method for aggregating estimates of the SCC in each cell to an
overall value. It is not immediately apparent that decision makers would simply add up net values for
each cell in the matrix. They may wish to account for those who suffer losses differently from those
who gain. Such a concern might arise from awareness of political responsibilities, assessment of the
risk of disruption associated with losses, or recognition of the non-substitutability of some
environmental systems and cultural inheritance. Or, they may chose to weight some values
differently than others—for instance market values might not be equity weighted while a high equity
weight might be applied to the socially contingent values.
The risk matrix is a frame of reference, but does not imply specific values for the SCC for the less
certain impacts and valuations (that is, for row 3 and columns B and C). Further studies and estimates
of all of the cells are required to judge the extent to which sampling across all of the cells is required
to produce a robust estimate of the SCC. However, the Intergovernmental Panel on Climate Change
(IPCC) suggests that the larger impacts will become more likely as global temperatures rise
particularly beyond the middle range of 2-3 °C (see IPCC 2001a, and the Summary for Policymakers,
www.ipcc.ch/pub/spm22-01.pdf). The cascade of impacts across sectors and regions becomes an
increasing concern if global warming exceeds 5 °C or so. However, this conclusion is in the nature of
expert judgement, since there are few detailed studies presently available in the literature.
Defining the social cost of carbon
The term, social cost of carbon (SCC), generally refers to the marginal cost of climate change
impacts. The SCC is usually estimated as the net present value of the impact over the next 100 years
(or longer) of one additional ton of carbon emitted to the atmosphere today. This should not be
confused with the total impact of climate change or the average impact (the total divided by the total
emissions of carbon). The SCC is expressed as the economic value (in US$, € or GB£) per ton of
carbon (tC). In this assessment, the baseline is the year 2000 for the emissions as well as for the net
present value. In some literature, but not in this report, marginal damages are related to 1 ton of
carbon dioxide. 1t C = 3.664t CO2. So, a value of £100/tC would be equivalent to £ 27/t CO2.
Social Cost of Carbon: A Closer Look at Uncertainty Page 9 November 2005
The sensitivity of the SCC to the timing of the additional emission of 1tC can be evaluated in models
such as FUND. Emission of the additional carbon in 2020 would occur against a reference scenario
of larger impacts (assuming the climate system has not stabilised by that time) and generally results in
a larger value of the SCC at that time. The temporal profile of SCC estimates is taken up in the report
on policy implications (Watkiss et al. 2005).
Uncertainty and measures of the distribution of estimates
Estimates of the SCC are often distributions of results from a wide range of assumptions and plausible
values for uncertain parameters. This raises the question of which measures to use to portray the
range of results as well as the central tendency.
For example, there are a considerable number of extreme values in the full suite of results from
FUND. Some are likely to be anomalies in the model and are considered outliers—these have been
filtered from the results presented here. Extreme values that remain are possibly conditions in which
the impacts of an additional ton of carbon on regional climates affect the projected economy. The
marginal SCC refers to the effect of 1 additional (marginal) ton of carbon released to the atmosphere.
The implied assumption is that the marginal greenhouse gas emission leads to impacts that are only
slightly different from a reference scenario of greenhouse gas emissions. However, if the climate
change crosses a threshold of sensitivity, the impacts may be quite large, indicating a non-linear
response. In effect, the FUND model results may be drawn from more than one population—those
scenarios that conform to the model’s expectation of marginal impacts and those scenarios that
indicate non-linear changes in regional economies.
In such situations, it is not possible to a priori define the best measure of the central tendency of the
data set. Table 2 shows three approaches to measuring the central tendency. If the data conforms to a
normal distribution (or a homogeneous population with few real outliers), the average and standard
deviations are unbiased estimates. However, the arithmetic mean is sensitive to outliers that may not
be representative of the underlying probability distribution.
Measures based on a cumulative probability function include the quartiles and median. The
distribution of the data is captured in the median and quartiles: The minimum, maximum, and three
quartiles (lower 25%, median or 50% and upper 25%) are derived from the ordered data set. The
median is the value for which 50% of the data are larger. The median is less sensitive to outliers, but
is biased towards low values when the probability distribution has a long, high-value tail.
An alternative to the arithmetic average is to trim the data to remove some outliers and then calculate
a trimmed mean. Trimming more outliers, from a data set with large positive anomalies, reduces the
trimmed average. Thus, a trimmed mean with 20% of the outliers removed (10% from each tail) is
lower than a trimmed mean with 10% of the outliers removed.
In this report, we use several measures of the central tendency. For example, where the data sets are
available we report a trimmed mean with 10% or 1% of the values removed (5% or 0.5% from each
tail): this is represented as SCC(T)10 or SCC(T)1. The range of plausible estimates of the SCC is
designated SCCmin to SCCmax. These values are not referenced to a specific use or decision framework.
We also suggest a value for the SCC for setting global targets for mitigation: the range of plausible
values is labelled as SCCglow to SCCghigh.
We also report other measures where they are commonly cited in the literature or model results. The
FUND annex shows the results for the various measures of central tendency.
Social Cost of Carbon: A Closer Look at Uncertainty Page 10 November 2005
Table 2. Measures of central tendency
Normal distribution
Absolute
minimum
-2 Standard
deviations
-1 Standard
deviation
Average +1 Standard
deviation
+ 2 Standard
deviations
Absolute
Maximum
Quartiles
Q0 Q1 Q2 Q3 Q4
Absolute
minimum
Lower 25% of values Median, 50%
of values
Upper 25% of values Absolute
maximum
Trimmed mean
Tmean(20) < Tmean(10) < Tmean(5) < Tmean(1)
SCC
SCCmin SCCglow SCC(T)10 SCCghigh SCCmax
These measures are derived from model results (or collections of studies in the case of the meta-analysis). Of
course, the normal distribution does not contain an absolute minimum or maximum.
A note on units and conversion factors
This assessment necessarily involves technical detail on units and conversion factors.
The FUND model uses USD1995 as the benchmark, while PAGE reports damages in USD2000.
Where possible, we have inflated the FUND results to USD2000 by using the average U.K. Retail
Price Index over the period from 1995 to 2000, an increase of 22.5%.
We have converted both PAGE and FUND results from USD2000 to GBP2000 ($1.42 = 1.00) and
Euro2000 ($1 = €1.01) using purchasing power parity exchange rates from 2000. Thus, the
conversion from FUND USD1995 to GBP2000 is a multiplier of 0.863.
Results from the literature are cited in the units and time periods reported. For example, the meta-
analysis reported below reflects the range of base years used in each study. To provide a consistent
analysis, the modelling and olicy studies assumed the estimates in the literature used USD1995.
These estimates are updated to GDP2000 using the multiplier of 0.863. For example, an estimate of
the SCC of USD1995 100/tC = USD2000 122.5/tC = GBP2000 86.3/tC.
Social Cost of Carbon: A Closer Look at Uncertainty Page 11 November 2005
3 Understanding the social cost of carbon
Our understanding of the social cost of carbon depends on a cascade of steps, each with inherent
uncertainties:
(i) Reference socio-economic scenarios (baseline) !
(ii) Climate change projection/scenario !
(iii) Local to regional impact modelling !
(iv) Projected baseline and impacts over time !
(v) Valuation of local to regional impacts including adaptation !
(v) Aggregation to global values
Uncertainties and choices in the evaluation influence the final social cost of carbon. For instance,
notable sources of uncertainties are:
A reference baseline of high economic growth is often assumed to lead to less vulnerability to
climatic risks, at least in the loss of life, as wealthier societies can afford a wider range of
adaptation strategies.
Climate change projections of temperature are commonly included, but quantified impacts of
changes in multi-year drought are not.
Impacts at the scale of livelihood are not easily scaled up in a regional model based on GDP per
capita or the share of agriculture in the economic accounts.
Adaptation over time attenuates impacts in many sectors (for instance as farmers adjust to new
climatic conditions); while behavioural responses to climate outlooks could accelerate effective
adaptation or induce maladaptation and higher costs that may not be warranted.
Relatively small climate impacts may cross a threshold of vulnerability that leads to positive
feedbacks and non-linear effects (e.g., the socially contingent column in the risk matrix).
The choice of valuation methods, for instance willingness to pay or willingness to avoid damages,
is at least as important as the reference scenario.
Equity weighting seeks to compensate for the different value of marginal impacts to poor people
compared to rich people; but it does not account for rights or differential vulnerability per se.
A precautionary approach seeks to avoid damages without offsetting consideration of benefits.
Discounting procedures and the time profile of rising GHG emissions and climate impacts are
well documented factors (see Watkiss 2005 for time profiles and Guo 2004 for comparison of
discounting schemes).
Our assessment adopts a conventional view of the SCC as a representation of future impacts that are
given in the assumptions of a certain reference scenario. Of course, the reality is that present
estimates of the SCC might (or should, depending on one’s view of economic policy) influence our
efforts to stabilise greenhouse gas concentrations (and reduce emissions). If we underestimate the
future risks of adverse impacts, and do not stabilise the climate system, we increase the likelihood that
future impacts will be much greater than we currently estimate. This recursive nature of setting
targets based on a cost-benefit analysis is central to the concerns addressed in the companion
assessment led by Paul Watkiss (2005).
This chapter illustrates such concerns from the lines of evidence developed in the project. Our key
messages are contained in the section headings. Note that we focus on the nature of the uncertainties
behind the range of SCC estimates; as such we use several measures of central tendency and
cumulative probabilities.
Social Cost of Carbon: A Closer Look at Uncertainty Page 12 November 2005
3.1 Our understanding of future climatic risks, spanning trends and surprises in the climate
system, exposure to impacts, and adaptive capacity, is improving, but knowledge of the cost
of climate change impacts is still poor.
The project did not seek to revise the chain of assessment underlying global estimates of climate
change damages (essentially the steps (i) to (v) above. We began the project with a review of the
published literature, related to the risk matrix presented in Chapter 2. We also elicited estimates of
the social cost of carbon from experts, using a prescribed set of scenarios. Results from these two
lines of evidence support this key message.
Richard Tol reviewed existing estimates of the social cost of carbon (Tol 2004), which includes a
fairly complete list of references to original studies (see the annex).2 Five substantive conclusions
emerge from the review of the literature.
First, the Defra seminar in 2003, including Tol’s meta analysis, the Defra background paper (Pittini et
al. 2003, 2004) and Pearce’s review (2003), were reasonably complete in terms of the published
literature, but not in terms of the full coverage of potential impacts. However, we have not uncovered
a substantial body of new estimates. This reflects the relatively restricted character of the field—it is
unlikely that a major project or result would escape our collective notice—and the relative lack of new
work in this area.
Second, the coverage of existing studies is almost exclusively in the upper left quadrant of our risk
matrix (see Table 3). Most of the studies, and relatively greater confidence, is in the market-projected
climate change cell. For instance, FUND is benchmarked to changes in temperature (and sea level
rise), with only an indirect connection to changes in precipitation (included in the middle row).
Third, the range of uncertainties mentioned in the literature includes the familiar concerns. Most
studies include some regional and sectoral breakdown and discounting over time. Other sources of
uncertainty, such as equity weighting, cross-sectoral interactions, and a full range of future economic
scenarios are mentioned in some studies. (See the following boxes for detail on discounting and
equity weighting.)
Fourth, few of the published studies provide sufficient detail (of either the model or results) to
decompose the uncertainties and their relative importance. Thus, a formal meta-analysis of all of the
sources of uncertainty in the prevailing literature is not possible (without getting additional
information and model results from each study).
Fifth, some uncertainties have been ignored. Regional impact assessments (such as the plethora of
country studies) are not captured in the global estimates—which are based on global integrated
assessment models at a coarse spatial and socio-economic scale. Valuation issues such as aggregating
social preference functions, risk aversion and socially contingent factors have not been explored in the
published quantitative estimates.
The conclusions in Table 3 also apply to FUND and PAGE. In a strict sense, both models only use
global projections of mean temperatures (the first row) although the regional impacts are based on
studies that include some estimates of the bounded risks (the second row). Both FUND and PAGE
include market and non-market sectors, although neither would claim that the sectoral coverage is
complete. Neither provides robust results of system changes and socially contingent effects, although
both have explored these areas to some extent (see the Annex for FUND results undertaken by this
project).
2 An Endnote library of references and abstracts on the social cost of carbon has been prepared as well.
Social Cost of Carbon: A Closer Look at Uncertainty Page 13 November 2005
Table 3. Locating the literature in a risk assessment framework
Uncertainty in valuation
A. Market B. Non-market C. Socially
contingent
1. Projection Over 95% of the studies are in this category; with
a bias toward market costs.
2. Bounded risks Some models have explicit scenarios but most
are tied to benchmark 2xCO2 scenarios and do
not cover local changes in weather.
Plausible effects
have been posed but
not adequately
valued nor included
in the marginal SCC
Uncertainty in
Climate Change
3. System change and
surprise A few exploratory studies*, but not sufficient to
provide robust estimates of the marginal SCC No credible studies
* Nordhaus and Boyer (2000), Ceronsky (2005)
A range of experts were asked to provide estimates of the SCC for some 30 prescribed scenarios of
climate change, coverage of impacts sectors, and decision choices (such as discount rate). The experts
included well-known advocates for a high SCC as well as for low SCC values. Each expert was asked
to rate their confidence in each response. This question was not benchmarked in any way—the
interviewer did not prompt the respondent to anchor the range or the mid-point. (This could be done,
for instance to mention the scores given by the research team, or to suggest a scale relative to the
average of the experts.) The confidence ratings are subjective. The experts received the results with
an opportunity to comment on the conclusions—but this was not an iterative exercise nor was it
designed to achieve a consensus among the experts.
Of the nearly 450 scenario-responses, fully 70% had a confidence rating of very low or low. None of
the scenarios were judged a confidence of very high and only 3% had a high confidence.
The respondents were grouped into three categories (4 to 5 in each group) based on their overall view
of the SCC (Figure 4). For those respondents who held low or high values for the SCC, their
confidence was very low for at least some scenarios. Only those who held low values had high
confidence in at least one estimate (generally one of the scenarios based on low climate change and
only market sectors). None of the respondents in the medium to high group had more than a medium
level of confidence in their estimates.
Low Medium High
Respondents
Confidence
Very
low
Very
high
Figure 4. Expert confidence in estimates of the social cost of carbon for three groups of experts
based on their overall expectation of the SCC
Confidence was judged on a 5-point scale, from very low to very high. The central values are the average for all
scenarios and respondents in the group, bracketed by the minimum and maximum confidence rating. Note that
the confidence ratings are given by each respondent for each scenario.
Social Cost of Carbon: A Closer Look at Uncertainty Page 14 November 2005
3.2 The lack of adequate sectoral studies and understanding of local to regional interactions
precludes establishing a central estimate of the social cost of carbon with any confidence.
In a field where the parameters that drive the range of estimates are well understood, a probabilistic
assessment can be constructed that produces a central estimate. Examination of the cumulative
probability distributions for different reference scenarios and compared for different models then
would give a sense of whether the central estimate is robust (i.e., estimates of the central value fall
within an acceptable range).
A robust estimate of the central value for the SCC is not possible at present. Our primary reason for
this conclusion, noted in the previous section, is that the full range of risks and exposures has not been
included in present models. Therefore, sensitivity testing of model parameters is not necessarily
based on the full range of the drivers of uncertainty.
This section explores some of the evidence and reasoning behind this conclusion. We begin by
presenting three probability distributions—from the analysis of results published in the literature,
from the revised FUND model, and from the more reduced-form depiction of impacts from PAGE.
We then examine some of the output from FUND to illustrate issues related to sectoral and regional
scale. In the following section we look at temporal and dynamic uncertainties.
As noted above, Tol (2004) reviewed published estimates of the SCC. The 103 estimates from 28
published studies are used to calculate the distribution of the SCC. These published studies
concentrate on market sectors and global or regional projections of temperature. While many include
some non-market sectors few included regional precipitation or extreme events in a rigorous fashion.
Tol filtered the estimates using four schemes:3
1. The simple average of all of the 103 estimates results in a mean value of the marginal
damages of $114/tC (£80/tC), with a standard deviation of $249/tC (£175/tC) (and a median
of $17/tC (£12/tC)).
2. For studies that report more than one estimate, the authors generally provide a weight (or
probability range) for the various estimates. This weighting increases the mean to $158/tC
with a standard deviation of $392/tC (or £111/tC and £276/tC) (and a median of $20/tC
(£14/tC)). This is partly due to some authors assigning very low weights to low estimates (for
instance, those proposed by Nordhaus, 1994).
3. Tol calculated his own weights for each study based on six criteria: whether they had been
peer reviewed, were independent impact assessments, had included dynamic climate change
scenarios rather than equilibrium responses, used economic reference scenarios, calculated the
marginal damage costs, and the year of publication. This weighting scheme results in a mean
estimate of $105/tC, still with a high standard of deviation, of $305/tC (£74/tC and £215/tC).
4. If only the peer reviewed studies are included, with Tol’s weights applied, the estimates are
much lower, with a mean of $61/tC and standard deviation of $102/tC (£43/tC and £72/tC)
(and a median of $17/tC (£12/tC)).
The results are shown in Figure 5 as cumulative density functions for three of the weighting schemes
(#2 – 4 above). The published, but not peer reviewed literature, accounts for a substantial degree of
the uncertainty at the high end of estimates. For instance, the 90 percentile marginal damage cost is
$245/tC (£173/tC) in the peer reviewed literature, increasing to $350/tC (£246/tC) if all literature is
included (but with Tol’s weighting for the quality of the assessment (#3 above).
In summary, the distribution of results from the meta-analysis of the literature is shown in Table 4.
3 The estimates cited in the literature are taken to be US$1995. In the following text, these are inflated to
US$2000 using the average UK Retail Price Index, an increase of 22.5%. The inflated results in US$2000 are
converted to GBP2000 based on $1.42 = £1.00 (as for the FUND and PAGE model results).
Social Cost of Carbon: A Closer Look at Uncertainty Page 15 November 2005
Table 4. Distribution of the SCC from a meta-analysis of the literature, GBP2000/tC
All literature Peer reviewed
literature
No weights Weighted by
study authors Weighted by R
Tol
5% -9 -10 -8
Mean 80 111 43
95% 300 550 210
Based on Tol (2004).
Tol emphasises the discount rate (see box) and aggregation across countries (equity weighting
according to per capita income) as the two most significant factors explaining the range of results.
However, it is likely that these are only two of the most salient differences recorded in the studies, and
further uncertainties may be important.
The shape of the curves shows the difficulty in defining a robust central estimate. The mean values in
the four filters applied to the data are in the range of £35 to £91/tC. The probability distribution is
right-skewed, with a steep increase up to about $25/tC and a long tail of much higher values. Values
in the region of £5-10/tC, as suggested by Pearce (2003), are in the region of high uncertainty in the
S-shape of the curve. A small change in the value results in a large change in probability. For
example, moving from $0 to $25/tC is a jump from 10% to 50% probability in these plots. Beyond
about $20/tC the three curves start to diverge. Thus, the probabilities of higher estimates are strongly
influenced by the assumptions made in applying different filters to the literature.
Discounting
Social time preference is the value society attaches to present consumption. The Social Rate of Time Preference
(SRTP) is used to discount future benefits and costs. The Green Book recommends that the SRTP be used as the
standard real discount rate.
The rate at which individuals discount future consumption, on the assumption of an unchanging level of
consumption per capita over time, is called the Pure Rate of Time Preference (PRTP). The Green Book
suggests a PRTP of around 1.5 per cent a year for the near future. If per capita consumption is expected to grow
over time, future consumption will be plentiful relative to the present and thus have lower marginal utility. This
effect is represented by the product of the annual growth in per capita consumption (g) and the elasticity of
marginal utility of consumption (µ). The Green Book indicates the annual rate of g is 2 per cent per year, and
the elasticity of the marginal utility of consumption (µ) is around 1. SRTP is the sum of these two components:
SRTP = PRTP + µ * g
With a pure time preference rate of 1.5%, and values of 2% of g and 1 for µ, the resulting discount rate is 3.5%.
Note that the Green Book allows other declining discount rates to be used in more cautious risk assessments.
Source: Green Book, HM Treasury
Social Cost of Carbon: A Closer Look at Uncertainty Page 16 November 2005
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-25 0 25 50 75 100 125
dollar per tonne of carbon
probability
Figure 5. The composite cumulative density function of the marginal social cost of carbon.
The curves represent weights assigned by the study authors (top, grey), Tol’s quality weights (middle, black),
and Tol’s quality weights including only peer-reviewed studies (bottom, light grey). The values shown are for
US$1995 (from the original studies). Source: Tol (2004).
FUND also produces probability distributions. The current model (November 2004) tests the full
uncertainty with parameters set to values that are sampled from distributions for each parameter. (A
reference mode using ‘best guess’ estimates for the parameters is also possible. This mode is not used
in this report although results are presented in the Annex. Two of the main drivers of the SCC
estimates are the choice of discounting scheme and equity weighting (see boxes for more detail).
Figure 6 shows probability distributions for the full set of parameters, for several discounting
schemes. The 0% Pure Rate of Time Preference (PRTP) without equity weighting is unrealistic, but
provides a standard basis of comparison of impacts. The UK Treasury Green Book discounting
scheme starts with a discount rate of 3.5% for the first 30 years, then 3% for 45 years, and then a
declining rate to 1%.4 FUND produces a range of estimates that includes a substantial proportion of
net benefits (i.e., estimates less than 0). But it also has a strong skew toward higher numbers, well
beyond £150/tC. The distribution of FUND results for the lower discounting schemes (Green Book
and PRTP=1%) suggests the upper and lower quartiles are in the range of £-10/tC and £100/tC after
rounding off. Trimmed means for two discounting schemes are (£/tC):
Trimmed mean (10%) Trimmed mean (1%)
Green Book, with EW 38 57
PRTP=0%, no EW 98 157
4 The Green Book discounting scheme results in estimates close to the 1% PRTP scenario, rather than the 3%
PRTP scenario. This is because the Green Book rates are consumption discount rates, which already include the
growth component. The PRTP scenarios have this added on. Thus the 3% PRTP scenario represents
considerably higher discounting than Green Book discounting.
Social Cost of Carbon: A Closer Look at Uncertainty Page 17 November 2005
0
50
100
150
200
250
300
-16000
-4000
-1000
-200-50 050
200
1000
4000
16000
GBP2000
Frequency
Green Book EQ
PRTP=0, EQ
PRTP=0
PRTP=3, EQ
PRTP=3
Figure 6. Probability distributions from FUND
The heavy green line is the Green Book with equity weighting. Other runs are for PRTP=0% and 3%, with
equity weighting (solid lines) and without equity weighting (dashed lines). The uncertainty is for the full set of
parameters, but with some of the extreme values (positive and negative) eliminated as implausible.
Values are in GBP2000, converted from the FUND USD1995 by multiplying by 0.863. The PRTP values are
discounted by the growth rate in addition to the Pure Rate of Time Preference. Note that the x-axis scale is not
linear in order to highlight the range of values in the tails of the distributions. See the Annex for further details
and results.
0 50 100 150 200
Probability
$trillion
Figure 7. Distribution of the total climate damages ($2000) for the A2 reference scenario from
the PAGE model
The A2 scenario is from the IPCC Special Report on Emission Scenarios. It has a high projected population with
mixed regional economies reliant on fossil fuels. Carbon emissions are quite high, 28.9 GtC/yr in 2100. See
Nakicenovic and Swart, 2000, www.grida.no/climate/ipcc/emission.
Social Cost of Carbon: A Closer Look at Uncertainty Page 18 November 2005
Equity Weighting
With a utilitarian social welfare
function, each person’s utility
counts equally. It is generally
accepted that each additional
unit of consumption provides
diminishing marginal utility.
That is, giving £1 to a rich
person produces less utility
(welfare or happiness may
substitute as rough equivalents)
than giving £1 to a poor person.
Social Cost of Carbon: A Closer Look at Uncertainty Page 19 November 2005
The impacts/damages routines in PAGE are much simpler than FUND. Damages are disaggregated
into 8 world regions but only two sectors (market and non-market). It assumes all impacts will be
adverse, scaled to a 2xCO2 benchmark for damages (as in FUND. A strength of PAGE is the relative
ease of altering assumptions and the rapid calculation of probability distributions. The baseline
scenario for the project (including the policy report) is the IPCC SRES A2 reference scenario of GHG
emissions, purchasing power parity exchange rates, the Green Book SRTP, and an equity weight of 1
(using PAGE2002 V1.4e green book).
Under the A2 scenario, the mean impacts of climate change are GBP 51 trillion ($73 trillion) for a
time horizon of 2200 and discounted to a net present value. The PAGE model uses a range of
parameters, including for discount rate and equity weighting. The range of results is shown in Figure
7. A small number of runs that gave impacts above $200 trillion are not shown on the graph, but are
included in the mean impacts of $73 trillion.
The resulting distribution of the SCC (£/tC) for emissions in 2000 is:
5% mean 95%
9 46 130
This assessment draws upon four formal lines of evidence for the estimate of the social cost of
carbon—the published literature, new results from the FUND and PAGE models, and the elicitation
of estimates from experts. The latter was not designed to produce a probabilistic range of estimates;
the values are reported here only for comparison.
Surprisingly, the four lines of evidence show some consistency in the central estimates. Means range
from £40 to £60/tC for the peer reviewed literature (converted to GBP2000) and in the PAGE and
FUND model results undertaken for this assessment. The corresponding median estimates are on the
order of £10 to £40/tC.5
The range of estimates is still quite wide, from £-50/tC to well over £200/tC. The 5% and 95% range
in FUND is the widest span, and the models and experts have a wider range than the means and
medians reported in the analysis of the literature (as expected) (see Figure 8).
The PAGE results are all positive, but with less indication of the very large costs from FUND. This is
not surprising, since PAGE is benchmarked to this literature and is not an independent valuation of
the sectoral and regional impacts. The PAGE mean estimate is £47/tC).
Similarly, the knowledge elicitation reveals a range of estimates with the median very similar to the
literature, FUND and PAGE. However, the elicitation from the experts was not designed to reach
consensus around a central value, nor was it intended to set bounds to the range of plausible estimates
(see the Annex for more details).
5 The median has a 50% probability of being exceeded and is less than the mean due to the right-skewed
distribution of the SCC estimates.
Social Cost of Carbon: A Closer Look at Uncertainty Page 20 November 2005
-100
-50
0
50
100
150
200
250
300
350
5% 25% 50% 75% 95%
£/tC
Meta-analysis (peer reviewed): average
Meta-analysis (peer reviewed)
FUND: Green Book, EW: T(1)
Fund: 5%, 25%, 50%, 75%, 95%
Page: 5%, mean, 95%
Experts: 25%, 50%, 75%
Figure 8. Comparison of distributions of estimates of the SCC
The markers indicate data points from four sources. The meta-analysis of the literature, using only the peer
reviewed literature and Tol’s weights, shows values from £-8/tC (5%) to £211/tC, with an average of £43/tC.
The average from FUND (with 1% of the outliers trimmed) for the Green Book discounting and equity
weighting is £57/tC, with a range from £-54/tC (5%) to £310/tC (95%).
What do we know about the decomposition of the estimates to the sectoral and regional level? To
address this question, we rely primarily on the results from FUND since neither the literature, PAGE,
nor the knowledge elicitation provide sufficient disaggregation to analyse our confidence in the SCC
estimates. It is precisely because FUND attempts to build up consistent global estimates from
regional and sectoral analyses that an examination of the robustness of those components is possible.
Five concerns are apparent in the analysis of the disaggregated results: (i) regional and sectoral
balance of impacts, (ii) regional validation, (iii) independence of the sectoral damages, (iv)
aggregating damages and distribution of winners and losers, and (v) other constraints on impacts.
(i) Figure 9 and Figure 10 present the regional and sectoral breakdown of results from FUND. For the
regional breakdown, the richer regions (such as Japan/Korea, USA, China and Western Europe)
account for a large fraction of the costs, compared to the smaller economies of the small island states
or Eastern Europe for example. In this example, Japan/Korea, China and Western Europe account for
30% of the total benefits and costs of climate change, for the case of no discounting and no equity
weighting. With equity weighting and the Green Book discounting, Africa has the highest impacts.
Similarly, the sectoral disaggregation from FUND is shown in Figure 10. The dominance of the
results by agriculture and energy costs for space heating and cooling is notable. For the Green Book
scheme with equity weighting, these three sectors account for some 75% of the total costs estimated in
FUND (i.e., adding benefits to costs).
(ii) FUND applies the same sectoral damage functions, with different parameters, to each region and
the results are not validated by high resolution national or regional assessments. In fact, few such
assessments exist, so this remains an enduring challenge (see the conclusion regarding further work,
section 6.2). An informal discussion of the FUND results with an impacts specialist from China (Lin
Erda, personal communication, 2004) suggested that there would be considerable differences of
opinion regarding the underlying impact models, in addition to the well-known debates over
economic valuation. Where the model estimates are dominated by a few regions, the case for regional
validation would be even stronger. Ideally, validation should be done at the regional and sectoral
level, and over time, to ensure a robust analysis.
Social Cost of Carbon: A Closer Look at Uncertainty Page 21 November 2005
(iii) The sectoral impacts are considered in isolation, with the assumption that climate change drives
the sectoral impacts independently of other effects.6 Yet, the multiple stresses of climate change may
lead to an acceleration of the impacts. Conversely, adaptive capacity might be built up across sectors
resulting in reduced damages (or increased benefits). A classic example of the former is the case
where water shortages will prevent irrigated agriculture from reaping its full benefits. This appears to
be a significant constraint in China—the ability to take advantage of longer growing season and
increased radiation would depend on an irrigation infrastructure and water resources that are not likely
to be available in the near future.
(iv) The distribution of the SCC values within a region may differ significantly from the regional
total. This is the well known issue of spatial resolution. Regions with net costs near 0 (in these
results), including the Middle East, small island states, and Canada, are likely to have significant
impacts in some sectors. Some justification can be made for treating a country or integrated regional
economy as one ‘exposure unit’, to the extent that trade-offs between winners and losers within an
economy can be addressed by specific policies. Even so, the balance of effects between one sector
and another may be difficult to accommodate. For example, reduced heating costs will benefit
northern Europe while increased cost of air conditioning and cooling will be significant in southern
Europe. And, reduced heating costs might not compensate for loss of land and species due to sea
level rise. The sectors where damages are significant (i.e., not the damages net of benefits) may be a
primary concern for decision makers.
(v) Sectoral damages are often related to first-order impact variables without other social or economic
constraints. It is by no means clear that the early benefits of climate to agricultural potential (as noted
for China) will be realised in the face of the enduring agricultural surpluses, constraints on trade, and
costly producer price supports.
A conclusion from this analysis is that regional-sectoral estimates of the SCC are not well validated,
and those produced from global models should not be taken as reliable regional or sectoral estimates.
This does not necessarily lead to the conclusion that existing global estimates of the SCC are
unrealistic. The model results presented here are based on multiple runs (1000 in the case of FUND)
using a range of input parameters. To some extent, the uncertainty at the region-sector level should be
reduced in aggregating to the global level—a specific region-sector may have a low estimate in one
run and a high estimate in another run. This reinforces the need to understand the sources of
uncertainty in the SCC and to evaluate the estimates in an explicit risk framework. The regional-
sectoral validation remains a high priority if better estimates of the SCC are to be developed and used.
6 Note that the coastal zone sectors are related through a simple model of the least cost of coastal protection or
retreat (the value of abandoning dryland and wetland).
Social Cost of Carbon: A Closer Look at Uncertainty Page 22 November 2005
-£8
-£6
-£4
-£2
£0
£2
£4
£6
£8
£10
£12
JPK
SEA
ANZ
MDE
SIS
CAN
CAM
EEU
LAM
SAS
SSA
FSU
MAF
USA
CHI
WEU
PRTP=0%
Green Book, EQ
Figure 9. Regional disaggregation of median estimates of the SCC in FUND
The bars show the total damages with a Pure Rate of Time Preference = 0%), sorted from those that benefit from
climate change (with a total benefit of £7/tC) to those that suffer the greatest losses (total losses are £55/tC).
The regional values for the Green Book discounting scheme with equity weighting is shown for comparison—
with some notable differences in the distribution of winners and losers. Note that the regional breakdown of
FUND results is not intended to imply estimates of the SCC at the regional level. The uncertainty in the regional
values (not shown) is likely to be greater than the global uncertainty (already considerable).
Key JPK = Japan and Korea
SEA = Southeast Asia
ANZ = Australia and New Zealand
MDE = Middle East
SIS = Small island states
CAN = Canada
CAM = Central America
EEU = Eastern Europe
LAM = Latin America
SAS = South Asia
SSA = Sub-Saharan Africa
FSU = Former Soviet Union
MAF = North Africa
USA = United States of America
CHI = China
WEU = Western Europe
-£30
-£20
-£10
£0
£10
£20
£30
£40
£50
£60
Agriculture
Water
Forests
Seaprotection
Dryland
Wetland
Species
Imigration
Emigration
Heating
Cooling
Death
Morbidity
PRTP=0%
Green Book, EQ
Figure 10. Sectoral disaggregation of median estimates of the SCC in FUND
The bars show the total damages with a Pure Rate of Time Preference = 0%), grouped according to impacts on
the rural economy, coastal zone, energy costs, and health. The total of the sectors that benefit (for PRTP=0%) is
£26/tC, compared to total losses of £74/tC. The sectoral values for the Green Book discounting scheme with
equity weighting is shown for comparison—with a similar ranking of sectors. Note that the sectoral breakdown
is not intended to imply estimates of the SCC for single sectors.
TOTAL WINNERS: £26/tC
TOTAL LOSERS: £74/tC
TOTAL LOSERS: £55/tC
TOTAL WINNERS: £7/tC
Social Cost of Carbon: A Closer Look at Uncertainty Page 23 November 2005
3.3 The balance of benefits and damages in the social cost of carbon shifts markedly over time,
with net damages increasing in later time periods. Estimates of the SCC are particularly
sensitive to the choice of discount rates and the temporal profile of net damages.
Clearly there are some benefits to climate change. Agriculture can benefit from higher carbon dioxide
concentrations and longer growing seasons. Rainfall may increase in some regions, sufficient to
balance or exceed increased evapotranspiration due to warmer temperatures. Reduced costs of
heating are likely to be widespread.
In FUND, net benefits are apparent for the first few decades of the 21st Century. This can be seen in
(Figure 11) which represents the time profile of annual damages in different regions.7 In these results,
FUND shows net benefits for China in both scenarios. Regional benefits for other regions differ
largely due to the equity weighting. What is less clear, is the extent of the benefits (noted in section
3.2 for agriculture and China) and the length of time for which benefits might exceed damages. If the
benefits are large and last for say 30-40 years, then higher discount rates will tend to tilt the balance in
favour of net benefits in calculating the SCC as a net present value. Conversely, with smaller net
benefits for a shorter period of time, lower discount rates will tend to favour the longer term exposure
to increasing net damages. Thus, temporal uncertainty interacts with the discount scheme (at least in
FUND). The issue does not arise in PAGE, which assumes the SCC is a net damage from the baseline
time period.
SCC, GBP2000/tC, GB EW
-£4
-£3
-£2
-£1
£0
£1
£2
2000
2010
2020
2030
2040
2050
2060
2070
2080
2090
2100
2110
2120
2130
2140
2150
2160
2170
2180
2190
2200
2210
2220
2230
2240
2250
2260
2270
2280
2290
£/tC
Europe & FSU
Africa & MEast
South Asia
SE Asia & China
Pacific
Latin America
USA & Canada
SCC, GBP2000/tC, GB
-£2
-£2
-£1
-£1
£0
£1
£1
2000
2010
2020
2030
2040
2050
2060
2070
2080
2090
2100
2110
2120
2130
2140
2150
2160
2170
2180
2190
2200
2210
2220
2230
2240
2250
2260
2270
2280
2290
£/tC
Europe & FSU
Africa & MEast
South Asia
SE Asia & China
Pacific
Latin America
USA & Canada
SCC, GBP2000/tC, PRTP=0% EW
-£25
-£20
-£15
-£10
-£5
£0
£5
£10
£15
£20
2000
2010
2020
2030
2040
2050
2060
2070
2080
2090
2100
2110
2120
2130
2140
2150
2160
2170
2180
2190
2200
2210
2220
2230
2240
2250
2260
2270
2280
2290
£/tC
Europe & FSU
Africa & MEast
South Asia
SE Asia & China
Pacific
Latin America
USA & Canada
SCC, GBP2000/tC, PRTP=0%
-£2
-£2
-£1
-£1
£0
£1
£1
2000
2010
2020
2030
2040
2050
2060
2070
2080
2090
2100
2110
2120
2130
2140
2150
2160
2170
2180
2190
2200
2210
2220
2230
2240
2250
2260
2270
2280
2290
£/tC
Europe & FSU
Africa & MEast
South Asia
SE Asia & China
Pacific
Latin America
USA & Canada
Figure 11. Temporal profiles for regional estimates of the SCC in FUND
Results are for the Green Book (top) and PRTP=0% (bottom) discounting schemes, with equity weighting (left)
and without equity weighting (right). Note the units are GBP2000/tC. Convergence to zero in the long term is
because of discounting rather than reduction of absolute valuations (which may increase over time).
7 It is these annual damages that are discounted to a net present value
Social Cost of Carbon: A Closer Look at Uncertainty Page 24 November 2005
The corresponding temporal profiles for groups of sectors are shown in Figure 12. As already noted,
agriculture and heating have early benefits, in both discounting schemes. The same observations
regarding balancing benefits and damages over time apply.
In this assessment, we have not explored the sensitivity of our outcomes to very different temporal
profiles. It would be possible, for instance, to weight losers more than winners, following a concern
of policy makers to prevent dangerous climate change rather than to optimise the balance of
mitigation and impacts.
SCC, GBP2000/tC, Greenbook EW
-£10
-£8
-£6
-£4
-£2
£0
£2
£4
2000
2020
2040
2060
2080
2100
2120
2140
2160
2180
2200
2220
2240
2260
2280
£/tC
Agriculture & Forests
Water
Species
Coastal
Health
Heating and cooling
SCC, GBP2000/tC Greenbook
-£6
-£5
-£4
-£3
-£2
-£1
£0
£1
£2
2000
2020
2040
2060
2080
2100
2120
2140
2160
2180
2200
2220
2240
2260
2280
£/tC
Agriculture & Forests
Water
Species
Coastal
Health
Heating and cooling
SCC, GBP2000/tC, PRTP=0%, EW
-£60
-£50
-£40
-£30
-£20
-£10
£0
£10
£20
£30
2000
2020
2040
2060
2080
2100
2120
2140
2160
2180
2200
2220
2240
2260
2280
£/tC
Agriculture & Forests
Water
Species
Coastal
Health
Heating and cooling
SCC, GBP2000/tC, PRTP=0%
-£6
-£5
-£4
-£3
-£2
-£1
£0
£1
£2
£3
2000
2020
2040
2060
2080
2100
2120
2140
2160
2180
2200
2220
2240
2260
2280
£/tC
Agriculture & Forests
Water
Species
Coastal
Health
Heating and cooling
Figure 12. Temporal profiles for sectoral estimates of the SCC in FUND
Results are for the Green Book (top) and PRTP=0% (bottom) discounting schemes, with equity weighting (left)
and without equity weighting (right). Note the units are GBP2000/tC. Convergence to zero in the long term is
because of discounting rather than reduction of absolute valuations (which may increase over time).
Social Cost of Carbon: A Closer Look at Uncertainty Page 25 November 2005
3.4 Vulnerability and adaptation to climate change impacts are dynamic processes responding
to climatic signals, multiple stresses, and interactions among actors. Large scale impacts,
such as migration, can be triggered by relatively modest climate changes in vulnerable
regions.
The usual estimates of the SCC are based on average conditions (e.g., increments of warmer
temperatures) in generalised damage functions with few feedbacks over time, between regions or
among actors. However, assessments of the impacts of climatic variations, and the ability to adapt to
them, are based on increasingly sophisticated, dynamic models of decision making, multiple stresses,
and socio-institutional conditions. For example, syndromes of poverty and environmental degradation
have been developed and tested with scenarios of different climatic risks. At present, it is not possible
to scale up such local to regional models of dynamic responses to global estimates of the social cost of
carbon.8
This assessment of the SCC included an exploratory evaluation of the kinds of vulnerability hot spots
that might lead to large scale impacts of climate change. We noted three conditions would
immediately qualify as hot spots:
Coastal deltas where dense populations are subject to increased coastal erosion, recurrent
storm surges and cyclone risk. Migration out of the high hazard area is constrained by the
lack of space, social and cultural factors and poverty. Bangladesh is the archetypical
example, but mega-cities in coastal zones could also become increasingly hazardous.
Semi-arid regions at the boundary of agricultural and pastoral production systems where
climatic episodes, primarily of drought, already create stresses and may tip the system into an
increasingly instable state. Migration is constrained by ethnic conflicts, as well as economic
constraints. The Sahel is an archetypical example, and indeed has been subject to decreasing
rainfall since the 1960s.
Small island states where sea level rise, and possibly increased cyclone risks, threaten the
physical resources, literally inundating an entire country. Migration is the only feasible
solution, often to a foreign country (Nicholls et al. 2005).
A key question for global estimates of the SCC is the extent of area and number of people subject to
such conditions. A related concern is how to value such non-linear impacts and the multipliers
associated with migration, loss of an economic sector and socio-political stress and conflict. These
are the sorts of socially contingent impacts that are represented by the right hand column of our risk
matrix, and specifically the cells C-1 and C-2.
This assessment did not attempt to quantify these risks in terms of £/tC or to compile an inventory of
indicators of impacts (a related EC project is beginning to do some of this). Rather, we identified the
issue in general terms, focussed on a representative example (migration in the Sahel in West Africa)
and developed a pilot model of the potential impacts of climate change. These results are presented in
some detail as an annex to this report.
To the extent that vulnerability and adaptation are dynamic processes, with significant changes over
time, they should both be understood in estimates of the long-run impacts of climate change. The
literature on vulnerability and adaptation science (e.g., Downing 2003) is rapidly growing,
recognising the many climate and non-climate factors that influence the levels of risks that threaten
specific social, economic and environmental conditions. The approach to adaptation that prevails in
the SCC literature, in contrast, is quite limited. Adaptation is generally seen as a reduction in
potential impacts related to a few macro-level variables, such as GDP per capita. Or, the case for
adaptation is made based on an equilibrium comparison of climate sensitivity in other regions,
8 The lack of regional validation of global estimates of the SCC has two implications: (i) regional breakdowns
from global models should not be taken as accurate assessments of regional damages (see section 3.2) and (ii)
regional experts and climate policy negotiators are less likely to have confidence in global SCC estimates.
Social Cost of Carbon: A Closer Look at Uncertainty Page 26 November 2005
assuming the adaptation capacity is easily translated to new conditions and other places. In reality,
adaptation is a process that will have its own costs, including the costs of failed investments and
maladaptation.
Social Cost of Carbon: A Closer Look at Uncertainty Page 27 November 2005
4 Uncertainty and risk
Different integrated assessment models generate different estimates partly because they adopt
different assumptions about (uncertain) future states of nature (such as global warming), society (e.g.,
population growth) and economies (e.g. GDP), and of the sensitivity of impact sectors to both the
exogenous driving forces and climate change. However, estimates of the SCC are arguably driven
even more by different assumptions about preferences (e.g., ethical choices and decision frameworks)
and future policy responses (the role of the decision framework and time in Figure 1).
It should be clear that uncertainty regarding the climate system, impacts and their valuation is only
one driver of the estimates of the social cost of carbon. Three categories of drivers are commonly
noted:
1. Exogenous uncertainty. Over a time horizon of two centuries, the underlying rate of
innovation and economic growth are uncertain and generally taken as exogenous to the
estimate of the SCC. However, a climate policy model integrating the SCC and greenhouse
gas mitigation should include economic growth and technology as endogenous properties.
Equally, much of the underlying climate science has to account for exogenous uncertainty, for
instance in future solar radiation and volcanic eruptions.
2. Policy uncertainty. Different assumptions about public and private responses to climate
change generate different estimates of the marginal social cost of carbon.
3. Ethical judgments. Ethical judgements about equity, time weighting and our aversion to risk
(and ambiguity in risks) have a significant impact upon the marginal social cost of carbon.
In the following lists, uncertainties are classified:
* topic of an MSc thesis completed for the project
^ included in the scenarios used in the expert knowledge elicitation
The Clarkson and Deyes (2002) paper identifies four main ‘economic valuation uncertainties’:
1. The range of sectors included: market and non-market (^)
2. Assumptions about how valuations of climate impacts will change over time
3. Assumptions about equity weighting (*,^)
4. Assumptions about discounting (*,^)
These are certainly important. The background paper for the Defra International Seminar on the
Social Cost of Carbon (Pittini and Rahman 2003, 2004) adds a further three drivers of variability:
5. The valuation of low-probability catastrophic effects (*)
6. The valuation of ‘socially contingent’ effects (^)
7. Differences between valuation based upon willingness to pay and willingness to accept loss
In addition to these seven items, two further items which have so far escaped much attention by policy
makers are also important:
8. The degree to which preference heterogeneity is accounted for (*). Replacing the standard
(but incorrect) assumption of identical preferences with the assumption that people have
different (heterogeneous) preferences can produce surprising results. For instance, Gollier
and Zeckhauser (2003) show that if people use constant heterogeneous discount rates, their
aggregate behaviour will reveal a declining discount rate.9 Hence allowing for heterogeneity
in time preference means that more weight is placed on the future, generating a higher
estimate of the social cost of carbon. Similar results may obtain for heterogeneous
preferences on risk or equity weighting.
9 The logic is that as time advances, more weight is placed upon the preferences of people with lower discount
rates.
Social Cost of Carbon: A Closer Look at Uncertainty Page 28 November 2005
9. Assumptions about risk and ambiguity aversion (*). As dealing with climate change
represents an exercise in risk management, our risk preferences are critically important.
Moreover, because distributions over outcomes are generally not properly defined — they are
‘ambiguous’, in that the probability distribution itself is uncertain — our preferences over
ambiguity aversion are also critical. Integrated assessment models that do not take risk and
ambiguity aversion into account are implicitly assuming risk and ambiguity neutrality. The
existence of the insurance industry clearly suggests that this is an erroneous assumption.
Of these nine items, the sensitivity of estimates of the social cost of carbon to the five starred items
(*) were evaluated in four MSc theses (supervised by Cameron Hepburn, see Anthoff 2004, Ceronsky
2004, Guo 2004, and Li 2004). These five items — equity weighting, discounting, catastrophic
impacts, preference heterogeneity and risk and ambiguity aversion — were selected because they are
large drivers of variability. We do not examine growth and development scenarios, speed and
effectiveness of adaptation, or innovations in CO2 abatement technology. Abstracts of the four theses
on these topics are found in the annex.
The project also conducted a knowledge elicitation of 14 experts. The methodology posed scenarios
of key drivers of uncertainty and variations in the SCC and asked the experts to provide an estimate of
the SCC as well as their confidence in the estimate. In addition to the economic factors noted by a
carot (^) above, the scenarios included the range of projected temperature and sea level rise, whether
precipitation and storm risks were included, whether adaptation was included, and a choice of local,
world average or EU decision perspective (primarily relating the values assigned to impacts such as
loss of species and human life). Further details are included in an annex to this report.
This chapter discusses the results grouped according to three key messages.
4.1 Climate uncertainties and the climate sensitivity are key factors in larger estimates of the
SCC.
Clear drivers of the SCC are the assumptions regarding climate change itself. If climate change is
expected to be in the lower range of the IPCC, less than 2°C by 2100, then lower estimates of the SCC
are expected. On the other hand, if climate change is in the upper range, above 5°C by 2100, then it is
difficult to escape a higher estimate of the SCC.10
For instance, in Figure 13 expert responses to the scenarios of relatively low climate change (less than
2°C, labels beginning with L) averaged less than £10/tC. In contrast, scenarios of higher climate
change (over 5°C, labels beginning with X) were in excess of £50/tC, and often above £100/tC.
Ceronsky (2004) tested FUND for different climate sensitivities—the equilibrium warming expected
with a 2xCO2 scenario. The current ‘best guess’ is 2.5°C, with a range in the IPCC extending to
4.5°C. The SCC is 5-6 times higher at 4.5° than at 2.5° (Table 5, Figure 14). And, if the climate
sensitivity is more extreme, the SCC increases by a further factor of 3 to 7 (comparing 9.3° to 4.5°).
10 Estimates of higher temperature changes have emerged recently, see the results of the Exeter conference and
Stainforth et al. (2005).
Social Cost of Carbon: A Closer Look at Uncertainty Page 29 November 2005
-50
0
50
100
150
200
250
300
350
400
450
500
L>>M>5>>>
L>>N>5>A>
L>>N>3>A>
A_MIN
L>>M>3>>>
L>>N>5>>>
L>>M>>>>>
L>>N>3>>>
L>>N>3P>>
MD>M>3P>W
MD>N>3PAW
MD>N>5L>W
MS>N>3P>W
MS>NS3PAW
MD>NS3P>W
MD>N>3P>W
MS>NS3P>W
XS>N>3P>W
XSSM>1P>W
X>>>>>>>
XD>>>>>>
XSS>>>>>
XSSN>>>>
XSSNS1L>L
XSSNS1LAW
XSSNS1>>>
XSSNS1L>W
XSSNS1P>>
Z_MAX
XSSNSI1L>>
XSSNS1L>E
XSSNS1L>E
SCC £/tC
Min=-11 Median=22 Average=49 Max = 501; StDev = 66
Figure 13. Expert responses to scenarios of the drivers of the SCC, £/tC
The range is the minimum, average and maximum for 14 experts. The scenarios are coded on the 9 characters
of labels:
Temperature and sea level rise: Low, Medium, high (X)
Drought and regional declines in precipitation or increased Storms
Climate System surprise
Market or market and Non-market sectors
Socially contingent impacts included, along with market and non-market sectors
Discount rate: 1%, Green Book (3) or 5% (Pure Rate of Time Preference)
Per capita income equity weighting or weight Losers greater than winners
Adaptation included
Local, World average or European decision perspective
> indicates no information on this factor is included in the scenario
Table 5. Sensitivity of FUND estimates for different climate sensitivities
2.5˚C 4.5˚C 7.7˚C 9.3˚C
PRTP=0% 57 321 1446 2321
Green Book 18 100 268 357
PRTP=1% 11 88 357 571
PRTP=3% -2 17 73 116
Notes: The values produced by FUND (in the July04 version) were converted to GBP2000 and then indexed to
the 4.5°C Green Book value (=100, in bold). Thus, the value for 2.5°C and PRTP=0% is about half (57%) of
the 4.5°C Green Book value. Conversely, the 9.3°C , PRTP=3% value is 16% (index = 116) higher than the
Green Book 4.5°C value. The July04 version of FUND was subsequently updated, however the relative range
of results should be similar. Source: Ceronsky (2004).
Social Cost of Carbon: A Closer Look at Uncertainty Page 30 November 2005
Figure 14. Distribution of the SCC in FUND with four climate sensitivities
Results are for Green Book discounting. Note that these runs of FUND are for the July2004 version; more
recent results are presented elsewhere in this report. Extreme results from FUND have been dropped in these
probability distributions. Source: Ceronsky (2004).
This high degree of sensitivity to the underlying climate projection, might be reflected in policies that
seek to limit the extent of economic exposure to the higher range of damages. Further, the sensitivity
of estimates of the SCC to climate projections reveals a methodological issue of some importance.
The usual method of calculating the (marginal) SCC is to project a reference scenario (comprising at
least economic growth and climate change), add a pulse of carbon at the start of the run and calculate
the difference between the reference run and the scenario of added climate change resulting from the
marginal increase in a greenhouse gas. In reality, the climate-impacts-policy system is more
recursive. A cogent argument runs as follows:
A low estimate of the social cost of carbon, if used to set policy now, will lead to low targets
for stabilisation of carbon in the atmosphere. This is likely to lead to rapid climate change,
unless our understanding of the climate system is fundamentally wrong. So, over time the
SCC is likely to increase, partly due to the delay between releasing carbon into the
atmosphere and experiencing the impacts. Conversely, a high estimate of the SCC now,
would lead to high targets and lower climate changes, which hopefully would have lower
costs in the future.
Thus, in reality the SCC cannot be estimated without considering the feedbacks between policy,
emissions and impacts. We know of no model that makes this policy feedback explicit, although
some (including FUND) have a weak link between experienced impacts and economic growth, which
also affects GHG emissions.
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
-50 50 150 250 350 450 550
MD $/tC (Green Book discounting)
probability
Best Guess (2.5˚C)
C1 (4.5˚C)
C2 (7.7˚C)
C3 (9.3˚C)
Social Cost of Carbon: A Closer Look at Uncertainty Page 31 November 2005
4.2 Uncertainties in coverage, sectoral assessments and regional processes are likely to be
significant, but are difficult to judge without further model development and inter-model
comparison.
Chapter 3 documented issues in the regional and sectoral coverage of existing estimates of the SCC,
with the conclusion that a lack of regional/sectoral validation hampers confidence that the full range
of potential impacts have been reflected in current global estimates. The lack of independent
estimates and adequate data sets for validation preclude definitive statements about the range of
uncertainty that might be expected in each cell of our risk matrix. A priori, there is little evidence to
indicate that each new sector or better regional representation will lead to higher or lower global
estimates. Indeed, there is a sense that many of the key drivers of uncertainty are known and further
refinement might lead to compensating effects, at least at the global level.
We propose below to pursue a systematic bounding exercise where experts attempt to define the range
of potential impacts for each cell of the risk matrix (see section 6.1 below). As a starting point, we
would expect that further research on the market-projections quadrant (the upper-left cells) are
unlikely to lead to large effects on global estimates. On the other hand, decision values and
frameworks are certain to dominate the socially contingent-system change quadrant (the lower-right
cells) and it may not be possible to bound these estimates in a useful way. In between, uncertainties
might be several orders of magnitude but should be amenable to further refinement.
4.3 Decision variables such as the discount rate and equity weighting also are extremely
important.
A recurrent thread in the literature on the SCC is the importance of the discount rate and equity
weighting. This is confirmed in our assessment, with new investigations led by Cameron Hepburn
(Anthoff 2004, Guo 2004, Li 2004; see the Annex for a synopsis of these theses).
In FUND, as an example of the general effects, the higher discount rate of 1% PRTP produces
estimates that are on the order of 1/5th of the SCC calculated with a PRTP of 0%. The Green Book
scheme produces similar or somewhat lower estimates as a PRTP of 1% in FUND.
Equity weighting also has a potentially significant effect as well. With a PRTP of 0%, equity
weighting could increase the SCC by a factor of 5 to 15, depending on the way equity weights are
calculated. Even with higher levels of discounting, the increase might be a factor or 3 to 12. As noted
above, the results are sensitive to the shape of temporal profile and may be somewhat different in
other models.
Complex discounting schemes have been proposed and were tested in FUND. Their effect appears to
be sensitive to compounding factors such as economic growth rates. The Annex introduces this
literature.
Social Cost of Carbon: A Closer Look at Uncertainty Page 32 November 2005
Social Cost of Carbon: A Closer Look at Uncertainty Page 33 November 2005
5 The range of estimates
Despite uncertainties in the SCC, it is possible to draw some conclusions from the current estimates.
Although the range is quite large, there seems to be a reasonable consensus regarding a lower bound.
5.1 Estimates of the social cost of carbon span at least three orders of magnitude, from 0 to over
1000 £/tC, reflecting uncertainties in climate and impacts, coverage of sectors and extremes,
and choices of decision variables.
The minimum expectation of the SCC is £0/tC, or even a net benefit on the order of £5-10/tC. The
lower range of climate scenarios produced estimates of this order from the experts (Figure 13). At the
same time, the upper range of climate scenarios produced estimates as high as £500/tC. The
distribution of results from FUND are from £0 to over £1000/tC for a PRTP=0% and from £-100 to
over £500/tC for the Green Book discounting scheme (Figure 6). Similarly, the PAGE damage
estimates range from £0 to over £400 /tC.
Of course, the extreme tails of these estimates depend as much on decision values (such as
discounting and equity weighting) as on the climate forcing and uncertainty in the underlying impact
models. However, even reducing the range by half, say from £10/tC to £500/tC, still produces a wide
range of estimates.11 The high valuations cannot simply be dismissed as outliers.
5.2 A lower benchmark of 35 £/tC is reasonable for a global decision context committed to
reducing the threat of dangerous climate change and includes a modest level of aversion to
extreme risks, relatively low discount rates and equity weighting.
The Defra paper (Clarkson and Deyes, 2002) recommended £35/tC as the lower estimate for policy
evaluation. Pearce (2003) reviewed this estimate, concluding that a central value was in the region of
£10-20 /tC. Since the Defra paper, we have updated FUND and PAGE to produce new estimates,
evaluated the expert judgments and explored the role of the drivers of uncertainty more thoroughly in
a risk framework. The 2003 and 2004 Defra workshops reviewed progress in understanding the
uncertainties in estimates of the SCC, but was not designed to reach a new consensus.
Clearly, there continues to be enormous debate regarding the range of estimates of the SCC, whether a
central value makes sense and whether a minimum threshold for policy evaluation could be supported
from the current estimates. This debate is likely to continue for some years; certainly the absence of
new models and regional/sectoral studies leaves the validity of global estimates in some doubt.
We have not attempted to define a plausible, robust minimum value for all contexts. That would
require substantial new work on the difficult uncertainties (for instance, the non-market and socially
contingent effects that might arise from increases in climatic hazards).
Rather, we have evaluated whether the lower benchmark in the Defra paper is credible. Note that this
estimate is specifically related to a global decision context that has already agreed to the UNFCCC
commitment to prevent dangerous climate change. The global context also implies at least a modest
aversion to large scale risks, a long term view that is often associated with relatively low discount
rates, and concern for global welfare that implies at least a modest level of equity weighting.
We observe that £35/tC (using GBP2000 values) is a reasonable lower benchmark for the SCC in this
context. To be clear, we refer to this benchmark as: SCCglow. It is our judgment (not a model estimate
per se) of the lower benchmark (not statistically defined) that reflects a global context (g).
11 The full range might be expressed as £-102 /tC to £103 /tC, or even £104 /tC for some of the outliers in FUND.
This would be a range of five to six orders of magnitude.
Social Cost of Carbon: A Closer Look at Uncertainty Page 34 November 2005
This conclusion draws upon two lines of evidence. First, the model results show that this benchmark
has a significant likelihood of being exceeded. In FUND, with the Green Book discounting scheme
and equity weighting, there is about a 40% chance that the SCC exceeds £35/tC (Table 6). Table 7
shows several measures of the central tendency in FUND results—the Green Book trimmed mean
estimate with equity weighting is £38/tC. Similarly, the median value from PAGE is £46/tC.
Second, a number of scenarios judged by the experts give rise to values near or above £35/tC. The
respondents were grouped according to their overall perception—tending to low estimates of the SCC,
medium estimates, or high estimates. Among those with generally low estimates, values exceeding
£35/tC were given, but only for high climate scenarios and usually with other decision factors such as
European values (Table 8). In the middle group, values around £35/tC occurred for all climate
scenarios, and were quite common for middle to high scenarios. As expected, those with higher
scores regularly produced estimates greater than £35/tC.
Table 6. Summary of the probability that the SCC exceeds a given threshold in FUND
Green Book, EW PRTP=0% PRTP=0%, EW PRTP=3%
£35/tC 40% 52% 78% 8%
£50/tC 33% 47% 77% 5%
£140/tC 12% 27% 73% 2%
EW = Equity weighting.
Table 7. Summary of FUND results, GBP2000 /tC reference, averages and standard deviation
Reference Average Standard deviation Trimmean(10%) IQMean
EW No EW EW No EW EW No EW EW No EW EW No EW
Green Book £20 £19 £63 £24 £314 £165 £38 £20 £23 £11
PRTP=0% £728 £56 £815 £171 £1,375 £671 £785 £98 £601 £54
PRTP=1% £174 £11 £429 £43 £1,221 £240 £294 £24 £182 £10
PRTP=3% -£1 -£2 £40 -£1 £434 £165 £30 £0 £5 £5
EW = Equity weighting based on per capita income; No EW = no equity weighting
Median is the 50% value. Reference is the single run with the ‘best guess’ of FUND parameters. Average is the
arithmetic average (or median). The trimmean discards the first and last 5% of the values. The IQMean is the
average of the values between the lower and upper quartiles. Results are shown for four discounting schemes.
Table 8. Selected estimates of the SCC from experts for a range of scenarios, £/tC
Climate
Low Medium High
Respondent SCC range SCC range SCC range
L1 1 8 8 17 11 66
L2 -3 7 6 29 12 111
L3 -2 1 4 21 19 190
M1 8 33 17 50 17 78
X1 8 22 33 83 56 145
X2 0 22 0 36 11 167
Social Cost of Carbon: A Closer Look at Uncertainty Page 35 November 2005
5.3 An upper benchmark of the SCC for global policy contexts is more difficult to deduce from
the present state-of-the-art, but the risk of higher values for the social cost of carbon is
significant.
We did not reach a consensus for an upper benchmark for the SCC, SCCghigh in our notation. The
above tables suggest that, under pessimistic scenarios of climate change, it is not implausible to
consider estimates of the illustrative value proposed by Clarkson and Deyes (2002) of £140 /tC or
even higher (see also Pittini and Rahman, 2003, 2004). For instance, the analysis of the literature
showed approximately a 10% probability of the SCC exceeding £100/tC for all of the literature and no
weights and a 5% change that the SCC exceeded £210/tC using the peer reviewed literature and Tol’s
weights).
Examination of the risk matrix would suggest that a decision maker with some aversion to large-scale,
high-consequence risks would extend the SCC estimates to at least some socially contingent effects,
with some preference to reduce the risk of system changes and large scale consequences. While these
are not quantified, they are likely to be additional to the central estimates often cited for the SCC
based on market sectors, some non-market sectors and scenarios of climate change based on
temperature (with less clear links to changes in precipitation). The addition of equity weighting or
European values, perhaps related to concerns for a liability regime, would push the estimate of the
SCC still higher.
The experiments forcing FUND with releases of methane hydrates, high climate sensitivities and a
reduction in the thermohaline circulation show that large numbers are possible, but difficult to verify
(Ceronsky 2004, see Annex).
The deep uncertainty in the underlying accounts of the SCC precludes assigning a confidence interval
to the upper benchmark.
Social Cost of Carbon: A Closer Look at Uncertainty Page 36 November 2005
Social Cost of Carbon: A Closer Look at Uncertainty Page 37 November 2005
6 Further research and next steps
In this chapter we offer insights for further research on the social cost of carbon.
6.1 Substantive improvement in estimates of the social cost of carbon requires well validated
assessments at the regional scale that value the dynamic processes of vulnerability and
adaptation.
A comprehensive plan for improving estimates of the social cost of carbon has not been formulated,
although there are regular meetings on integrated assessment and the benefits of climate mitigation.
The IPCC fourth assessment report includes a chapter in working group II on the integration of
adaptation and mitigation, and possibly this will lead to a greater sense of how the various
uncertainties might be addressed in the future.
We propose four broad pathways of development. The first is the continued development of global
integrated models, such as FUND. The second is to complement the global assessments with detailed
sectoral studies and regional integrations. The third pathway is to explore the dynamics of socially
contingent processes including adaptation. The fourth development is to systematically address the
uncertainties through meta analyses and expert knowledge elicitation. These development strategies
are described in more detail here.
Perhaps ten significant developments are ongoing or contemplated in the integrated assessment
community. These developments can be charted in our risk matrix (Figure 15). Reducing uncertainty
in the geophysical drivers of climate change include (i) improving the scale of assessment and
understanding aggregation and disaggregation issues at the regional-sectoral scale, (ii) linking damage
functions to probabilistic scenarios of climate change; (iii) understanding cross-sectoral and multi-
stressor effects and (iv) refining estimates of potentially catastrophic impacts. Reducing uncertainty
in economic valuation includes: (1) adding new sectors to the damage functions; (2) broadening the
range of economic techniques (such as the premium attached to risk aversion); (3) including
additional metrics that policy makers may wish to take into account; (4) bounding exercises to provide
a first-order estimate of the range of potential damages (see section 6.1 below); (5) understanding the
dynamic aspects of vulnerability and adaptive capacity and their relationship to damages over time;
and (6) exploring the effects of alternative value systems, particularly in the loss of non-market
resources and non-marginal, socially contingent effects. Of course, these developments are
contingent upon improvements in climate and impacts science, which we note below.
Improving estimates related to the larger uncertainties—the lower and right-hand cells—requires
several improvements, and these may be the more difficult developments, constrained by the lack of
data, the choice of analytical tools and the framing of climate policy decisions.
Uncertainty in valuation
A. Market B. Non-market C. Socially
contingent
1. Projection
1, 2
i
1, 2, 4, 5, 6
i
1, 2, 4, 5, 6
i
2. Bounded risks
1, 2
i, ii, iii
1, 3, 4, 5, 6
ii, iii
3, 4, 5, 6
ii, iii
Uncertainty in
Climate Change
3. System change and
surprise
2
ii, iii, iv
3, 4, 5, 6
ii, iii, iv
3, 5, 6
ii, iii, iv
Figure 15. Planned developments in integrated assessment models
I – ix relate primarily to uncertainty in climate change; 1 – 6 relate primarily to economic valuation.
Social Cost of Carbon: A Closer Look at Uncertainty Page 38 November 2005
Valuations of the social cost of carbon are benchmarked on global sectoral impacts studies. Usually
this involves an analyst collecting the literature and deciphering an equation that fits the results to the
model inputs available (usually regional temperature change and economic conditions). Further
development of such global impacts models is essential. However, they are unlikely to provide the
critical development path for reducing uncertainty in the social cost of carbon. Rarely do the global
models include robust valuation methods. Their scale is too coarse to pick up the local conditions of
vulnerability. Processes of adaptation are usually ignored, which means a neglect of the socially
contingent damages. Increasingly, such efforts are adopting the SRES (Nakicenovic and Swart, 2000)
as the only framing for vulnerability, thus ignoring scenarios of greater future baseline vulnerability
(since all countries have significantly higher GDP per capita in the SRES).
The higher priority is to conduct robust regional studies that can focus on multiple stresses and
socially contingent effects. By regional we mean studies from the level of a district to country and
perhaps larger, depending on the availability of information, participation of experts and stakeholders,
and socio-economic integration.
For global/sectoral and regional/multi-stressor studies a range of metrics is desirable (bio-
geographical, human impact, economic values). A full understanding of climate change (not just
temperature and a couple of scenarios) are essential, including existing trends and large ensemble or
probabilistic forecasts with climate change. By convention, the studies should have a reference time
period of 2100, but a focus on the next 50 years is desirable (which has a larger effect on discounted
values) and a sense of the long term commitment beyond 2100 is helpful. The scale should match the
scale of exposure of the actors, rather than be defined in purely Cartesian terms.
The OECD Working Party on Global and Structural Policies has developed a work plan for a second
phase of research on the benefits of climate policies (OECD 2004). This follows from a set of
background and contributed papers that review the state-of-the-art (available on their web site, in the
process of publication; see http://www.oecd.org/document/).
Many potential impacts of climate change have yet to be included into the models used for estimating
the marginal damage costs of carbon dioxide. These include recreation, tourism, amenity, urban
infrastructure, many diseases, river floods, and storms. The reason for exclusion is that too little is
known about these impacts to come up with a credible, global and regionally specific estimates of
impacts. (In some cases, this knowledge is now emerging, and the reason for exclusion is the time-lag
between primary impact study and comprehensive economic impact assessment.)
Even if too little is known about the impact for inclusion in marginal damage cost estimates,
knowledge may suffice for a “bounding exercise”. In a bounding exercise, one makes rough
calculations of the minimum and maximum damages, and compares these to the estimated damages to
arrive at a rough estimate of the error introduced. Minimum damage estimates are as necessary as
maximum damage estimates, because some impacts may be beneficial.
As a simple illustration of a bounding exercise, consider the following. Dorland et al. (1999) report
that a 6% increase in the average wind speed, not inconceivable in Northwest Europe in 2050, would
increase average wind storm damage by 300 to 500%. In the Netherlands, the hundred-year storm
(Daria in 1990) did damage of about 0.5% of GDP. The average damage is then about 0.01% of GDP,
accounting for other than the one-in-a-100-year storm. If we multiply this with a factor 5, then the
damage done by climate change is 0.05% of GDP. If we assume that the Netherlands is representative
for temperate countries, then we find that excluding extra tropical storms from the impact analysis
leads to a small underestimate of total damages; if we use a total damage impact of 1%, then the error
is 5%. In fact, the Netherlands is not representative, because it is on the coast and densely populated;
real damages are smaller, and the underestimate due to excluding storm damages is therefore also
likely to be smaller than 5%.
This is only an illustration of how a ‘thought experiment’ could be used to test a range of input
assumptions and their potential effect on economic valuations of the impacts of climate change.
Social Cost of Carbon: A Closer Look at Uncertainty Page 39 November 2005
Bounding exercises should be a standard part of regional and sectoral studies, conducted by the
experts involved in the assessment. This at least provides a simple approach that complements full
economic valuations and can be readily scaled up to the global level to underpin the global estimates
of the social cost of carbon.
6.2 Revisiting the SCC, and using avoided damages in global negotiations to set policy targets,
will require substantial research in partnership with scientists and policy makers in
developing countries.
Expanding research on the social cost of carbon will require international collaboration, not least to
validate regional and sectoral results at a higher resolution than captured by global models. Previous
estimates of the SCC have met with great scepticism by many scientists; further refinement of models
in Europe (or other developed countries) is unlikely to gain acceptance for policy making without the
hard work of validating estimates at the regional and sectoral level with specialists and stakeholders in
each region.
Social Cost of Carbon: A Closer Look at Uncertainty Page 40 November 2005
Social Cost of Carbon: A Closer Look at Uncertainty Page 41 November 2005
7 Conclusion
This report has reviewed the current range of estimates of the social cost of carbon against a risk
matrix of the uncertainty in climate change and confidence in economic valuation. Table 9 repeats the
key messages from the preceding chapters.
Drawing upon these messages, five conclusions are salient. First, the risk matrix is a useful reference
framework to ensure that estimates of the SCC are complete. The categories correspond to common
typologies in climate science and economic valuation. The cells in the 3x3 matrix help organise and
explain current methodologies and estimates. However, it is not clear whether the matrix could or
should be used as a sampling frame for producing a new central estimate (and this has not been
achieved in this report).
Second, the overwhelming concentration of estimates is in the first two cells—benchmarked to global
projections of temperature increases (with the corollary of sea level rise) and market and non-market
sectors. Most of the integrated assessment models, such as FUND and PAGE, are located in this cell.
Therefore the range of estimates currently offered as the state-of-the-art is incomplete. However, it is
not clear the extent to which this sampling bias necessarily leads to a systematic under-estimate of the
SCC.
Third, there is strong evidence, based on peer reviewed literature, expert judgement and results from
two global models, that the SCC is likely to exceed a policy-relevant benchmark of £35/tC. The
underlying uncertainties preclude a consensus among researchers about a ‘best guess’ or ‘upper
bound’.
Fourth, planned developments will expand the ‘frontier’ of methods and estimates to include more
sectors (particularly non-market sectors), incorporate additional aspects of climate change (bounded
risks) and sensitivity to some economic assumptions (such as discount rates and risk aversion). Over
the next five years or so, a greater pool of knowledge on the first 2x2 cells in the risk matrix should be
apparent.
Fifth, addressing the greater uncertainties of systemic climate change and socially contingent impacts
(as well as realistic adaptation) will not be developed quickly. A much wider range of methods is
necessary, and bridging local and global scales is imperative. Based on the complete risk matrix, a
full understanding of the social costs of carbon is unlikely in the next 10 years. Indeed, it is not clear
that this scale of effort is in place at present.
Integration of estimates of the SCC into stakeholder decision frameworks offers opportunities to
interpret the boundaries of the SCC according to the values of different stakeholders and decision
contexts. However, further research on the utility of approaches is required. The methodological
implications of a hierarchy of estimates, corresponding to the scale of decision making, and similarly
the use of multiple indicators of concern, should be identified at an early stage.
Some ideas of a research strategy have begun to emerge. A dual track approach is warranted: with a
balance between furthering the robustness of economic estimates in the left half of the risk matrix and
exploration of complementary methods for understanding the socially significant risks of the lower
right half of the matrix. Underpinning both is a consistent approach to expert elicitation and
grounding of estimates in regional assessments of dynamic vulnerabilities and adaptive capacity.
Social Cost of Carbon: A Closer Look at Uncertainty Page 42 November 2005
Table 9. Key messages
Our understanding of future climatic risks, spanning trends and surprises in the climate system, exposure to
impacts, and adaptive capacity, is improving, but knowledge of the cost of climate change impacts is still poor.
The lack of adequate sectoral studies and understanding of local to regional interactions precludes establishing
a central estimate of the social cost of carbon with any confidence.
The balance of benefits and damages in the social cost of carbon shifts markedly over time, with net damages
increasing in later time periods. Estimates of the SCC are particularly sensitive to the choice of discount rates
and the temporal profile of net damages.
Vulnerability and adaptation to climate change impacts are dynamic processes responding to climatic signals,
multiple stresses, and interactions among actors. Large scale impacts, such as migration, can be triggered by
relatively modest climate changes in vulnerable regions.
Climate uncertainties and the climate sensitivity are key factors in larger estimates of the SCC.
Uncertainties in coverage, sectoral assessments and regional processes are likely to be significant, but are
difficult to judge without further model development and inter-model comparison.
Decision variables such as the discount rate and equity weighting also are extremely important.
Estimates of the social cost of carbon span at least three orders of magnitude, from 0 to over 1000 £/tC,
reflecting uncertainties in climate and impacts, coverage of sectors and extremes, and choices of decision
variables.
A lower benchmark of 35 £/tC is reasonable for a global decision context committed to reducing the threat of
dangerous climate change and includes a modest level of aversion to extreme risks, relatively low discount
rates and equity weighting.
An upper benchmark of the SCC for global policy contexts is more difficult to deduce from the present state-
of-the-art, but the risk of higher values for the social cost of carbon is significant.
Substantive improvement in estimates of the social cost of carbon require well validated assessments at the
regional scale that value the dynamic processes of vulnerability and adaptation.
Revisiting the SCC, and using avoided damages in global negotiations to set policy targets, will require
substantial research in partnership with scientists and policy makers in developing countries.
Social Cost of Carbon: A Closer Look at Uncertainty Page 43 November 2005
8 References and bibliography
Aaheim, A. and A. Schjolden (2004) An approach to utilise climate change impacts studies in national
assessments. Global Environmental Change 14: 147-160.
AEA Technology (2002) Quantification and Valuation of Environmental Externalities: The Case of Global
Warming and Climate Change. Culham (Oxon): AEA Technology.
Alcamo, J., Krol, M. and Leemans, R. (1995) Stabilizing Greenhouse Gases: Global and Regional
Consequences -- Results from the IMAGE 2.0 Model. Bilthoven: RIVM.
Alley, R.B.; Marotzke, J.; Nordhaus, W.D.; Overpeck, J.T.; Peteet, D.M.; Pielke, R.A.; Pierrehumbert, R.T.;
Rhines, P.B.; Stocker, T.F.; Talley, L.D. and Wallace, J.M. (2003) Abrupt climate change. Science
299: 2005-2010.
Anthoff, D. (2004) Social cost of carbon: Equity weighting. Oxford University, Masters Thesis.
Arnell, N.W. (1999) Climate change and global water resources. Global Environmental Change 9, S31-S50.
Arrow, K.J., Cline, W.R., Maeler, K.-G., Munasinghe, M., Squitieri, R. and Stiglitz, J.E. (1996) Intertemporal
Equity, Discounting, and Economic Efficiency. In: Bruce, J.P., Lee, H. and Haites, E.F., (Eds.) Climate
Change 1995: Economic and Social Dimensions -- Contribution of Working Group III to the Second
Assessment Report of the Intergovernmental Panel on Climate Change, pp. 125-144. Cambridge:
Cambridge University Press.
Ayres, R.U. and Walter, J. (1991) The greenhouse effect: Damages, costs and abatement. Environmental and
Resource Economics 1, 237-270.
Azar, C. (1999) Weight factors in cost-benefit analysis of climate change. Environmental and Resource
Economics 13 249-268.
Azar, C. and Sterner, T. (1996) Discounting and distributional considerations in the context of global warming.
Ecological Economics 19 169-184.
Batjes, J.J. and Goldewijk, C.G.M. (1994) The IMAGE 2 Hundred Year (1890-1990) Database of the Global
Environment (HYDE). 410100082, Bilthoven: RIVM.
Bilsborrow RE (1992) Population-growth, internal migration, and environmental degradation in rural-areas of
developing-countries. European Journal of Population-Revue Europeenne de Demographie 8 (2): 125-
148 1992.
Bilsborrow, R.E. (2002) Migration, population change and the rural environment. Environmental Change and
Security Project Report, Issue no 8. Washington: The Woodrow Wilson Center.
Casse-Gintz, M.., Ludeke, M.K.B., Petchel-Held, G., Reusswig, F.Plochl, M.Lammel, G and Schellhuber, H.-J.
(1997) Fuzzy logic based global assessment of the marginality of agricultural land use. Climate
Research 8 135-150.
Ceronsky, M., (2004) Checking the price tag on catastrophe: An investigation of the sensitivity of FUND's
Social Cost of Carbon estimates to the incorporation of non-linear climate responses. Oxford
University, Masters Thesis.
Clarkson, R. & Deyes, K. 2002, Estimating the Social Cost of Carbon Emissions. Working Paper 140.
Government Economic Service Working Paper 140. London: HM Treasury and DEFRA. (www.hm-
treasury.gov.uk)
Cline, W.R. (1992) Optimal Carbon Emissions over Time: Experiments with the Nordhaus DICE Model.
Washington, D.C: Institute for International Economics (mimeo).
Cowell, F., and Gardiner, K. (1999) Welfare weights. Report to the UK Office of Fair Trading, London.
www.oft.gov.uk/NR/rdonlyres/.
Dietz, A.j., Verhagen, A. and Ruben, R. (2001) Impact of climate change on drylands with a focus on West
Africa. Dutch National Research Programme on Global Air pollution and climate change. ICCD
Research, The Netherlands.
Social Cost of Carbon: A Closer Look at Uncertainty Page 44 November 2005
Dorland, C., Tol, R.S.J. and Palutikof, J.P. (1999) Vulnerability of The Netherlands and northwest Europe to
storm damage under climate change: A model approach based on storm damage in The Netherlands.
Climatic Change 43: 513–535.
Downing, T. and P. Watkiss (2003). The social costs of carbon in policy making: Applications, uncertainty and
a possible risk based approach. Paper presented at the Defra International Seminar on the Social Cost
of Carbon, 7th July 2003, London, UK.
Downing, T.E. (2003) Lessons from famine early warning and food security for understanding adaptation to
climate change: Toward a vulnerability/adaptation science? In Climate Change, Adaptive Capacity and
Development. Smith, J.B., R.J.T. Klein and S. Huq (eds.). London: Imperial College Press.
Downing, T.E., Eyre, N., Greener, R. and Blackwell, D. (1996) Projected Costs of Climate Change for Two
Reference Scenarios and Fossil Fuel Cycles. Oxford: Environmental Change Unit.
Downing, TE, RE Butterfield, S Cohen, S Huq, R Moss, A Rahman, Y Sokona and L Stephen (2001) Climate
Change Vulnerability Toward a Framework for Understanding Adaptability to Climate Change
Impacts. Nairobi: UN Environment Programme and Oxford: Environmental Change Institute.
Ekins, P. (1995) Rethinking the costs related to global warming: A survey of the issues. Environmental and
Resource Economics 6:231-277.
Evans, E., Ashley, R., Hall, J., Penning-Rowsell, E., Saul, A., Sayers, P., Thorne, C. and Watkinson, A. (2004)
Foresight, Future flooding, Scientific summary: Volume I, Future risks and their drivers. London:
Office of Science and Technology.
Eyre, N., Downing, T., Rennings, K. and R.S.J. Tol (1998). Global Warming Damages. Externalities of Energy,
Volume 8, September.Brussels: European Commission,
Eyre, N., Downing, T.E., Hoekstra, R. and Rennings, K. (1999) Externalities of Energy, Vol 8: Global
Warming. Luxembourg: Office for Official Publications of the European Communities, pp.1-50.
Fankhauser, S. (1994) The social costs of greenhouse gas emissions: An expected value approach. Energy
Journal 15 (2), 157-184.
Fankhauser, S., Tol, R.S.J. and Pearce, D.W. (1997) The aggregation of climate change damages: A welfare
theoretic approach. Environmental and Resource Economics 10 249-266.
Fankhauser, S., Tol, R.S.J. and Pearce, D.W. (1998) Extensions and alternatives to climate change impact
valuation: On the critique of IPCC Working Group III's impact estimates. Environment and
Development Economics 3 59-81.
Ganopolski, A., Petoukhov, V., Rahmstorf, S., Brovkin, V., Claussen, M., Eliseev, A. & Kubatzki, C. (2001):
CLIMBER-2: a climate system model of intermediate complexity. Part II: model sensitivity. Climate
Dynamics 17: 735-751.
Glasby, G.P. (2003) Potential impact on climate of the exploitation of methane hydrate deposits offshore.
Marine and Petroleum Geology 20(2): 163-175.
Gollier, C. (2002a) Discounting an uncertain future. Journal of Public Economics 85, 149-166.
Gollier, C. (2002b) Time horizon and the discount rate. Journal of Economic Theory 107, 463-473.
Gollier, C. and Zeckhauser, R. (2003) Collective investment decision making with heterogeneous time
preferences. NBER Working Paper 9629, Cambridge, MA.
Grubb, M. (2003) The economics of the Kyoto Protocol. World Economics 4 (3).
Grubb, M., Köhler, J., and Anderson, D. (2002) Induced technical change in energy and environmental
modeling: Analytic approaches and policy implications. Annual Review of Energy and the Environment
27 (3): 79-114.
Guo, J. (2004) Discounting and the social cost of carbon. Oxford University, Masters Thesis.
H.M Treasury (2003) The Green Book Appraisal and Evaluation in Central Government. London: H.M.
Treasury, see http://greenbook.treasury.gov.uk.
Hammitt, J.K., Lempert, R.J. and Schlesinger, M.E. (1992) A sequential decision strategy for abating climate
change. Nature 357: 315-318.
Social Cost of Carbon: A Closer Look at Uncertainty Page 45 November 2005
Higgins, P., Mastrandrea, M. and Schneider, S. (2002) Dynamics of climate and ecosystem coupling: Abrupt
changes and multiple equilibria. Philosophical Transactions of the Royal Society 357(1421): 647-655.
Hohmeyer, O. (1996) Social costs of climate change -- strong sustainability and social costs. In Hohmeyer, O.,
Ottinger, R.L. and Rennings, K., (Eds.) (eds.) Social Costs and Sustainability -- Valuation and
Implementation in the Energy and Transport Sector, pp. 61-83. Springer, Berlin.
Hohmeyer, O. and Gaertner, M. (1992) The Costs of Climate Change - A Rough Estimate of Orders of
Magnitude. Karlsruhe: Fraunhofer-Institut fur Systemtechnik und Innovationsforschung.
Hoozemans, F.M.J., Marchand, M. and Pennekamp, H.A. (1993) A Global Vulnerability Analysis: Vulnerability
Assessment for Population, Coastal Wetlands and Rice Production and a Global Scale (second, revised
edition). Delft: Delft Hydraulics.
Hope C, (2004) The marginal impact of CO2 from PAGE2002: An integrated assessment model incorporating
the IPCC’s five reasons for concern. Integrated Assessment (submitted).
Hope, C. (2003) The marginal impacts of CO2, CH4 and SF6 emissions . University of Cambridge, Judge
Institute of Management Working Paper WP 10/2003, Cambridge.
Hope, C. and Maul, P. (1996) Valuing the Impact of CO2 Emissions. Energy Policy 24 (3), 211-219.
Intergovernmental Panel on Climate Change (IPCC) (2001) Summary for Policymakers. Geneva, IPCC.
IPCC. (2001a) Climate change 2001, Impacts, Valuation and Vulnerability, Contribution of Working Group II
to the Third Assessment Report of the Intergovernmental panel on Climate Change, Cambridge
University Press.
IPCC. (2001b) Climate change 2001, The Scientific Basis, Contribution of Working Group I to the Third
Assessment Report of the Intergovernmental panel on Climate Change, Cambridge University Press.
Jorgensen, D., Goettle, R., Hurd, H., Smith, J. et al (2004) US market consequences of global climate change.
Arlington, VA: Pew Center on Global Climate Change. www.pewclimate.org.
Kates, R.W. and Heimann, M. (1992) Where the poor live: are the assumptions correct? Environment 34, 4-11,
25-28.
Kattenberg, A., Giorgi, F., Grassl, H., Meehl, G.A., Mitchell, J.F.B., Stouffer, R.J., Tokioka, T., Weaver, A.J.
and Wigley, T.M.L. (1996) Climate models – Projections of future climate. In: Climate Change 1995:
The Science of Climate Change – Contribution of Working Group I to the Second Assessment Report of
the Intergovernmental Panel on Climate Change, eds. J.T. Houghton, L.G. Meiro Filho, B.A.
Callander, N. Harris, A. Kattenberg and K. Maskell. Cambridge: Cambridge University Press, pp. 285–
357.
Keller, K., Tan, K., Morel, F. M. M., and Bradford, D. F. (2000) Preserving the ocean circulation: Implications
for climate policy. Climatic Change, 47, 17-43.
Leggett, J., W.J. Pepper, and R.J. Swart (1992) Emissions scenarios for IPCC: An update. In Climate Change
1992. Supplementary Report to the IPCC Scientific Assessment, J.T. Houghton, B.A. Callander,
S.K.Varney (eds.). Cambridge: Cambridge University Press, pp. 69-95.
Leonard, H.J., Ed. (1989) Environment and the Poor: Development Strategies for a Common Agenda. New
Brunswick: Transaction Books.
Li, A. (2004) Investigating the effect of risk and ambiguity aversion on the social cost of carbon. Oxford
University, Masters Thesis.
Li, H., Berrens, R.P., Bohara, A.K, Jenkins-Smith, H.C., Silva, C.L. and Weimer, D.L. (2004) Would
developing country commitments affect US households: Support for a modified Kyoto Protocol?
Ecological Economics 48, 329-343.
Link, P.M. and Tol, R.S.J. (2004) Possible economic impacts of a shutdown of the thermohaline circulation: An
application of FUND. Working Paper FNU-19. Hamburg: Hamburg University.
Lonsdale, K.G., Downing, T.E., Nicholls, R.J., Vafeidis, N., Dawson, R., and Hall, J. (2005) A dialogue on
responses to an extreme sea level rise scenario in the Thames Region, England. Climatic Change
(submitted).
Social Cost of Carbon: A Closer Look at Uncertainty Page 46 November 2005
Ludeke, M.K.B. , Moldenhauer, O. and Petschel-Held, G. (1999) Rural poverty driven soil degradation under
climate change: the sensitivity of the disposition towards the Sahel Syndrome with respect to climate.
Environmental Modelling and Assessment 4: 315-326.
Maddison, D. J. (2003) The amenity value of the climate: the household production function approach. Resource
and Energy Economi