Endogenous Growth, Convexity of Damage and Climate Risk: How Nordhaus' Framework Supports Deep Cuts in Carbon Emissions

Download as pdf or txt
Download as pdf or txt
You are on page 1of 47

The Economic Journal, 125 (March), 574–620. Doi: 10.1111/ecoj.12188 © 2015 The Authors.

The Economic Journal published by John Wiley & Sons Ltd on behalf of
Royal Economic Society. Published by John Wiley & Sons, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

ENDOGENOUS GROWTH, CONVEXITY OF DAMAGE AND


CLIMATE RISK: HOW NORDHAUS’ FRAMEWORK SUPPORTS
DEEP CUTS IN CARBON EMISSIONS*

Simon Dietz and Nicholas Stern

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


‘To slow or not to slow’ (Nordhaus, 1991) was the first economic appraisal of greenhouse gas
emissions abatement and founded a large literature on a topic of worldwide importance. We offer
our assessment of the original article and trace its legacy, in particular Nordhaus’s later series of
‘DICE’ models. From this work, many have drawn the conclusion that an efficient global emissions
abatement policy comprises modest and modestly increasing controls. We use DICE itself to provide
an initial illustration that, if the analysis is extended to take more strongly into account three
essential elements of the climate problem – the endogeneity of growth, the convexity of damage and
climate risk – optimal policy comprises strong controls.

1. To Slow or Not to Slow


‘To slow or not to slow’ by Bill Nordhaus (1991) is a landmark in economic research.
As the first analysis of the costs and benefits of policies to abate greenhouse gas
emissions,1 it opened the profession to a new field of application – climate change. Its
importance is partly illustrated by the number of times that it has been cited – on 1,150
occasions according to Google Scholar; 398 times according to the narrower, journals-
only measure in ISI Web of Knowledge.2
The context within which Nordhaus’s paper was written helps us understand its
contribution. While the basic science of the greenhouse effect was set out in the
nineteenth century by Fourier, Tyndall and Arrhenius, discussions surrounding the
possible role of humans in enhancing it – and therefore causing global warming and
climate change – began in earnest in the 1970s. For at least a decade, climate change
remained largely a scientific/environmentalist’s issue, debated in specialist confer-
ences and networks (Agrawala, 1998). Indeed, it is important to stress that the science
of climate change was running years ahead of the economics (something that
arguably remains the case today in understanding the impacts of climate change;
Stern, 2013).

* Corresponding author: Simon Dietz, Department of Economics, London School of Economics and
Political Science, Houghton Street, London WC2A 2AE, UK. Email: s.dietz@lse.ac.uk.
We thank Emanuele Campiglio, Antoine Dechezlepr^etre, Baran Doda, David Greenaway, Tom McDer-
mott, Elisabeth Moyer, Antony Millner, Bill Nordhaus and Bob Pindyck for helpful comments and
discussions, and the editor, Rachel Griffiths. We also acknowledge the financial support of the Grantham
Foundation for the Protection of the Environment and the Economic and Social Research Council. We alone
are responsible for the content.
This is an open access article under the terms of the Creative Commons Attribution License, which permits
use, distribution and reproduction in any medium, provided the original work is properly cited.
1
Shortly afterwards Bill Cline (1992) published what is generally considered to be the other foundational
analysis of climate mitigation benefits and costs.
2
Both accessed on 24 March 2014. However, these citation counts likely understate the paper’s legacy
considerably, since many will instead cite later work that is based on it (see Section 1).
[ 574 ]
[MARCH 2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 575
By the late 1980s, however, climate change was becoming both a policy issue and
increasingly political. In 1988, the Intergovernmental Panel on Climate Change
(IPCC) was established and in 1990 it published the first of its regular and
influential Assessment Reports to member governments. In 1989, the first meeting
of (22) Heads of State to discuss climate change was held in the Netherlands and
various other major international summits that year also put it on the agenda. Most
OECD countries already had their first climate-change targets by 1990 (Gupta,
2010), for instance the European Community, as it was then, had pledged to

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


stabilise its carbon dioxide emissions at 1990 levels by 2000. In 1992, virtually all
countries signed up to the United Nations Framework Convention on Climate
Change (UNFCCC) at a major summit on the environment and development in Rio
de Janeiro, with its objective to achieve ‘stabilization of greenhouse gas concentra-
tions in the atmosphere at a level that would prevent dangerous anthropogenic
interference with the climate system’ (Article 2).
Yet despite the obvious ecological risks of unmitigated climate change, the question
remained whether the benefits of avoiding these risks would outweigh the perhaps
substantial cost of cutting emissions.3 This is the central question that ‘To slow or not
to slow’ sought to tackle, by combining a simple model of social welfare and
production with an externality from greenhouse gas emissions, in general equilibrium.
This model took ‘existing models and simplified them into a few equations that are
easily understood and manipulated’ (p. 920), something that has become a hallmark
of Nordhaus’s work in the area. In summary, the main components of the model are:
(i) a single equation of motion for the global mean temperature, which rises in
response to the difference between the temperature that would be obtained in
long-run equilibrium, given the current atmospheric stock of greenhouse
gases, and the current temperature;
(ii) an equation of motion for the atmospheric stock of greenhouse gases, in
which some fraction of current emissions adds to the stock, at the same time as
some fraction of the current stock ‘decays’ by diffusing into the deep ocean;4
(iii) a social welfare function that is the discounted sum over time of utility per
capita;5
(iv) utility takes the form of the logarithm of consumption per capita of a single,
aggregate good;

3
There is a problem in using the language of benefit–cost analysis, if it is interpreted in its common and
narrow, marginal, fairly undynamic way and where risk is also treated narrowly. Climate-change policy raises
major questions of the strategic management of potentially immense risks and where different paths will have
different endogenous learning and discovery. This broader perspective is a major focus of this study and
should be central to economic research on the topic.
4
To get an idea of the simplicity of the modelling framework, especially the science module, note that a
fully fledged atmosphere–ocean general circulation model such as that of the UK Hadley Centre would
comprise hundreds of thousands of equations.
5
There is little plausibility in moral philosophy for a social welfare function that is the sum across
generations of the (discounted) utility per capita of each generation, irrespective of the number of people in a
generation, unless population is constant. Adding the (undiscounted) total utility of each generation is
essentially utilitarian. Pure-time discounting can be given a utilitarian interpretation if the discounting is
based on the probability of existence as a function of time, and that becomes an exponential function in
continuous time if the end of the world is the first event in a Poisson process.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
576 THE ECONOMIC JOURNAL [MARCH
(v) consumption per capita is given by (exogenous) output, less the total cost of
abating emissions, and the total cost of climate change;
(vi) a reduced-form abatement cost function, in which the total cost of abatement
depends on global aggregate emissions and emissions abatement; and
(vii) reduced-form damage, in which the total cost of climate change depends on
global mean temperature but where global mean temperature is an index of
a wider set of climatic changes including changes in precipitation and sea
level.

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


This modelling framework has had a lasting influence on the field and indeed
several elements of it still constitute the ‘industry standard’ today. The most notable
example of this is perhaps the idea of reduced-form damage.
According to the model, Nordhaus found that an optimal cut in the current flow of
global emissions of 11% relative to the base level should be made in a medium scenario
(given a rate of pure-time preference of 1% per annum and ‘medium’ damage). In a
‘high’ scenario, with no pure-time discounting and ‘high’ damage, a cut of global
emissions of around one third would be optimal. The concluding Section of this article
lays out these results, without commenting on the plausibility of the various scenarios.
Nonetheless, that the optimal emissions cuts were not more than one third implied that
only modest targets could be supported by economic analysis of this kind, in comparison
with some targets being discussed in the political arena. As the editor of the issue in
which the paper appeared wrote, it is ‘certainly a sobering antidote to some of the more
extravagant claims for the effects of global warming’ (Greenaway, 1991, p. 903).

2. The DICE Model Framework


While it was very much the purpose of ‘To slow or not to slow’ to cast climate-change
mitigation as a dynamic, investment problem, in which abatement costs could be paid up
front, so that climate change could be avoided several decades into the future, the model
dynamics were unsatisfactory – the economy was assumed to be in a so-called ‘resource
steady state’, in which all physical flows are constant. Therefore, we were asked to
consider the setting as being the middle of the twenty-first century, when such
conditions might plausibly hold (we can now see that this is highly unlikely). Optimal
emissions abatement was calculated by evaluating a marginal change to the steady-state
level (and thus the optimal cuts mentioned above were in the steady state). Time was still
relevant though, because, while the change in abatement costs was instantaneous, the
change in damage costs would be experienced only after a delay (Equations 7–9, p. 926).
Nordhaus himself was well aware of the shortcomings and indeed a preliminary
version of a more fully dynamic model had already been presented at a workshop by
the time ‘To slow or not to slow’ had been published. This new model was called DICE
(for a ‘dynamic integrated climate-economy’ model) (Nordhaus, 1992, 1993a,b, 1994).
Many elements of ‘To slow or not to slow’ could still be found in the original DICE
model, including the equation of motion of the atmospheric stock of CO2, log
utility and reduced-form abatement and damage costs. But at the core was a Ramsey–
Cass–Koopmans model of economic growth, allowing evaluation not only of the
optimal steady state but also of the optimal transition path. The social welfare function

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 577
was modified to include population, with the objective becoming the (pure-time)
discounted sum of total, instantaneous social utility, while a slightly more complex
model of temperature change was also added. Once again, the results of the analysis
with DICE pointed to modest emissions controls, modestly increasing over time – from
10% initially to 15% in the later twenty-first century.
Since these first studies with the DICE model, it has become the pre-eminent
integrated assessment model (IAM) in the economics of climate change. New versions
have been published periodically (Nordhaus and Boyer, 2000; Nordhaus, 2008), and a

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


regionally disaggregated model (RICE) was also developed (Nordhaus and Yang,
1996). However, to look only at Nordhaus’s own studies with DICE is to understate its
contribution hugely, because, by virtue of its simple and transparent unification of
growth theory with climate science (not to mention Nordhaus’s considerable efforts to
make the model code publicly available), it has come to be very widely used by others.
The uses to which it has been put are too numerous to cover in a comprehensive
manner. Some of the more significant examples include: the introduction of induced
innovation in the energy sector (Popp, 2004); explicit evaluation of optimal adaptation
policy (de Bruin et al., 2009); consideration of uncertainty and learning (Kolstad, 1996;
Keller et al., 2004); and treating consumption of material goods and environmental
quality separately, thus allowing evaluation of relative price changes (Sterner and
Persson, 2008).
Some of these extensions have challenged the broad conclusion that optimal
emissions control is modest. And indeed it is important to stress two things. First,
through his own updating of DICE, Nordhaus’s position, as formalised in the model
and its results, has shifted over the years towards stronger emissions reductions,
albeit incrementally. Second, one can readily see in Nordhaus’s writings an
awareness of the limitations of IAMs like DICE. Nonetheless, it is fair to say the
perception remains that an analysis of the costs and benefits of climate change in
an IAM does not support strong emissions cuts, under standard assumptions. For
instance, in the wake of the publication of the Stern Review on the Economics of Climate
Change (Stern, 2007) (which in fact used an IAM other than DICE), it has been
suggested that the difference in policy recommendations between the Review and
other studies lies very largely in the specification of a low pure-time discount rate
(Nordhaus, 2007), a rate that some have questioned.6 A central purpose of the rest
of this article is to explore whether a recommendation of modest emissions
reductions does indeed follow from using the DICE framework. We ask, can the
framework support strong controls on emissions, if restrictive assumptions about
growth, damage and climate risk are relaxed? These assumptions arguably lead to
gross underestimation of the benefits of emissions reductions in DICE and other
IAMs (Stern, 2013).
First, we incorporate endogenous drivers of growth and we allow climate change to
damage these drivers. This is in stark contrast to the current generation of IAMs, which
rests directly or indirectly on the Ramsey–Cass–Koopmans model, where the major

6
A careful exploration of the strong basis in moral philosophy for low pure-time discounting is provided
in Stern (2014a,b). In many IAM studies, high pure-time discounting is introduced without much discussion.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
578 THE ECONOMIC JOURNAL [MARCH
source of growth per capita in the long run is exogenous improvements in productivity,
but where climate change only impacts on current output.7 There are compelling
reasons for thinking that climate change could have long-lasting impacts on growth
(Stern, 2013) and there is now an emerging body of empirical evidence pointing in this
direction (Dell et al., 2012), even though climatic conditions in the recent past have
been relatively stable compared with what we now have to contemplate.
Second, we assume that the damage function linking the increase in global mean
temperature with the instantaneous reduction in output is highly convex at some

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


temperature. Consideration of some of the science, for example, on tipping points,
leads us in this direction (Weitzman, 2012). By contrast, most existing IAM studies
assume very modest curvature of the damage function. The DICE default is quadratic
and it is well known that with the standard values of the functions’ coefficients an
implausible 18°C or so of warming is required in order to reduce global output by 50%.8
Third, we allow for explicit and large climate risks. We do so by considering the
possibility of high values of the climate-sensitivity parameter; i.e. the increase in global
mean temperature, in equilibrium, accompanying a doubling in the atmospheric
concentration of carbon dioxide. We conduct sensitivity analysis on high values but
also specify a probability distribution reflecting the latest scientific knowledge on the
climate sensitivity as set out in the recent IPCC report (IPCC, 2013). Its key
characteristic is a fat tail of very high temperature outcomes that are assigned low
probabilities. By contrast, most IAM studies have ignored this key aspect of climate risk
by proceeding with a single, best guess value for the climate sensitivity, typically
corresponding to the mode of the IPCC distribution. We note, linking the second and
third points here, that the model temperature increase under business as usual a
century or so from now of 3.5 or 4°C (IPCC, 2013) could be extremely damaging – this
is not just a ‘tail’ issue.
Otherwise we remain faithful to the standard DICE framework, in order to make as
clear as possible the difference that these three extensions make. Most notably, we
retain its usual parameterisation of social values, where the rate of pure-time
preference is 1.5% and the elasticity of marginal social utility of consumption is 1.5,
so that with growth of consumption per capita of, say, 2%, the social discount rate would
be 4.5%. We have written elsewhere about why we think it is inappropriate to posit such
a high rate of pure-time preference (Stern, 2013, 2014 a,b) – and we return to explain
why in Section 5 – but for the purpose of clarity of comparison we set aside our
misgivings, concerning this and other features, in the modelling that comprises the
core of this study. More generally, there is a powerful case for arguing that this type of
model, with one good and exogenous population, has very serious defects in its ability
to capture key aspects of a problem for which destruction of the environment and
potential loss of life on a major scale are central.

7
While a reduction in current output may impact future growth via reduced savings – for a given savings
rate – we hypothesise that this effect is weak compared with direct reductions in the capital stock and
reductions in productivity. Fankhauser and Tol (2005) also find a weak impact of climate change on growth
via savings, using DICE. They did not, however, consider that climate damage could work on the capital stock
or on productivity.
8
Nordhaus sees the specification of the damage function for warming above 3°C as a ‘placeholder’ (Stern,
2013) but it is a placeholder that can have a powerful effect on the conclusions as we will see below.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 579
3. Extending DICE
3.1. Endogenous Growth
In standard DICE, the production function is:
Yt ¼ F ðKt ; Lt Þ ¼ ð1  D^t Þð1  Kt ÞAt Kta Lt1a ; (1)
where At is the exogenous element of total factor productivity (TFP) at time t, K is
capital, L is labour and a 2 (0, 1) is the capital exponent. D^ is the standard DICE

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


damage multiplier (see below for a definition) and the key point is that this is the only
pathway through which climate change affects growth – directly by multiplication with
gross output in each period. Λ represents emissions abatement costs. In all of our
analysis, we maintain standard assumptions about Λ and L, detailed in Appendix A
alongside many other aspects of the model.
In (1) the long-run growth rate of output, ignoring for one moment the role of
climate change, is given by the sum of the growth rates of At and Lt as in the standard
Solow (1956) model. Climate-change damage D^t (and abatement costs Λt) affect the
level of output in each period, which means that they can have two effects on the long-
run growth rate of output. First, period-to-period changes in D^t can effectively change
the long-run output growth rate. Second, depending on the rates of saving and capital
depreciation, D^t can impact the long-run growth rate by affecting capital investment
and in turn the stock of K in future periods.
Yet one of the central points of this study is that this is a very narrow story of how
climate change impacts on growth. We, therefore, consider two extensions to (1). Both
are endogenous growth models, incorporating knowledge spillovers from the
accumulation of capital by firms. And in both models, damage from a changing
climate not only fall on gross output at a particular point in time, they also
permanently reduce output possibilities at future points in time through their effect on
endogenous determinants of growth.

3.1.1. A model of capital damage, and knowledge proportional to the capital stock
Our first growth model incorporates knowledge spillovers via the capital stock in the
tradition of Arrow (1962), Romer (1986) and others. We combine this formulation
with a partitioning of the damage multiplier between output and capital. The
production function becomes
Yt ¼ ð1  DtY Þð1  Kt ÞAt Ktaþb Lt1a ; (1.K)
Y
where D now denotes the damage that directly reduce annual output. In this model,
we think of the economy as being composed of a number of firms, each making
investments. Growth is driven in part by learning-by-doing, which in turn depends on
each firm’s net investment, so that when the firm’s capital stock increases, so does
economy-wide productivity. We also make the standard assumption in this tradition
that knowledge is a pure public good. The elasticity of output with respect to
knowledge is b > 0, so that the knowledge process has a productivity factor Kb. These
assumptions have the effect of increasing the overall capital exponent to a + b. We
continue to assume an exogenous element of TFP A. This could be taken to represent
elements of productivity not captured in knowledge spillovers but we use it here
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
580 THE ECONOMIC JOURNAL [MARCH
principally for the narrower, instrumental purpose of calibrating (1.K) on (1) in the
absence of climate-change damage and emissions abatement costs, thus achieving a
controlled comparison of different production specifications.
We suppose there is further damage from climate change that reduces the
capital stock, which we label D K , so we obtain the following equation of motion of
capital:
Ktþ1 ¼ ð1  DtK Þð1  dK ÞKt þ It ; (2)

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


where d 2 [0, 1] is the depreciation rate on capital and It = sYt is investment, given
K

savings rate s (see Appendix A). In specifying D K we have in mind the representation of
two phenomena. First, D K includes permanent, direct climate damage to the capital
stock, for example, if climate change increases the likelihood of storms and those
storms damage infrastructure, or the abandonment of capital in coastal areas due to
sea-level rise. Second, D K could indirectly include broader impacts of climate change
on productivity via the endogenous growth mechanism (1.K). One effect it could pick
up is of a changing climate on the productivity of capital stocks, accumulated during a
different and more stable climatic regime. For example, water supply infrastructure
may become less productive, given a long-run change in precipitation. Another could
be that, if investment is increasingly diverted towards repair and replacement of capital
damaged by extreme weather, it may produce fewer knowledge spillovers. Appendix A
contains further details of how, for our simulation work, we partition damage D
between D Y and D K .
In sum, according to this model of growth and climate damage, some part of the
instantaneous impacts of climate change falls on capital rather than output, so that this
type of damage represents a permanent reduction in output possibilities in the future.
Moreover since the economy’s stock of knowledge is proportional to its stock of capital,
the negative effect on future output possibilities is magnified.

3.1.2. A model of endogenous TFP and damage to TFP


One constraining feature of production functions like (1.K) is that, since knowledge
is in one-for-one correspondence with the aggregate capital stock, it will depreciate
just as fast. If one considers a typical depreciation rate for economy-wide capital of
10% per year (indeed dK = 0.1 in DICE), the implication is a rapid diminution of
economy-wide knowledge over time. While the literature on measuring the returns
to R&D investment points to annual depreciation of around 15% of private, firm-
level R&D capital – see Hall et al. (2009) for a review, what we have here is a much
broader construct of knowledge concerned with overall skills and know-how.
Therefore, we offer an alternative formulation of endogenous growth – new as far
as we are aware – in which TFP is endogenous and depreciates more slowly than
capital.
We revert to the standard production function, modelling TFP through a separate
relation. The production function is hence:
Yt ¼ ð1  DtY Þð1  Kt ÞAt Kta Lt1a : (1.TFP)
Capital and TFP have different dynamics. The equation of motion of the capital
stock is simply given by

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 581

Ktþ1 ¼ ð1  dK ÞKt þ It : (2 0 )
Notice that in this specification, we do not allow climate damage to impact the
capital stock, although doing so would be straightforward by reverting to (2). The
equation of motion of TFP is given by
Atþ1 ¼ ð1  DtA Þð1  dAt ÞAt þ aðIt Þ; (3)
A
where dA is the net depreciation rate for productivity. We can think of d as

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


encapsulating both:
(i) depreciation of productivity through erosion or displacement of skills and
know-how; and
(ii) implicit, autonomous growth of TFP, which captures among other things
institutional innovations, beyond the scope of this model.
Given these two effects, dA could in principle be negative but here we assume it is
positive and less than dK.
D A is the part of damage that reduces productivity. It captures the productivity
effects of climate change mentioned above. Appendix A again explains how we
partition D between D Y and D A .
a(It) is a ‘spillovers’ function that converts the flow of capital investment in each
period into a flow of knowledge externalities across activities as a whole. This means
that the stock of TFP is augmented by knowledge spillovers, as well as changing over
time according to the balance of depreciation and autonomous growth due to other
factors, which is encoded in dA. In general, assume a0 ≥ 0. More specifically, in order to
again calibrate this model to standard DICE in the absence of climate damage and
abatement costs, it is necessary to assume further a0 > 0, a00 < 0, since in the standard
DICE model the growth rate of TFP falls rapidly in the initial periods.9 These
properties can be satisfied by
c
aðIt Þ ¼ c1 It 2 ;
where c1 > 0 and c2 2 ð0; 1Þ. Summing up, in this formulation some part of the
instantaneous impacts of climate change falls on TFP, permanently reducing future
output possibilities.

3.2. Convexity of Damage


The standard DICE damage function is of a convenient reduced form that has come to
be widely used in the field:
D^t ¼ 1  1=ð1 þ p1 Tt þ p2 Tt2 Þ; (4)
where T is the global mean atmospheric temperature relative to the period just
before the industrial revolution. The coefficients p1 and p2 are estimated by fitting
the function on data points, which comprise the sum of underlying sectoral studies
of climate damage at particular degrees of global warming (mostly T = 2.5  3°C),

9
Whether such concavity is theoretically or empirically plausible is not for this study.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
582 THE ECONOMIC JOURNAL [MARCH
for example, studies of crop losses and changing energy demand for space cooling
and heating.10 We should recognise, however, that these are ‘quasi’ data points,
since T = 3°C has not been seen on the planet for around 3 million years and
might lead to radical transformations in global climatic patterns. Making assump-
tions about the form of (2) is made still more difficult by the complete absence of
evidence on aggregate impacts for T ≥ 3°C. The quadratic form was originally
selected largely for convenience11 but it results in implausibly low damage at high
temperatures (Stern, 2008; Weitzman, 2012). This has prompted Weitzman (2012)

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


to suggest the following modification:

Dt ¼ 1  1=ð1 þ p1 Tt þ p2 Tt2 þ p3 Tt6:754 Þ; (4 0 )

where the coefficient p3 and its corresponding exponent are together used to satisfy
the assumption that, at T = 6, 50% of output is lost.12 This is the functional form
we use in this study13 but, in addition to Weitzman’s calibration of p3, we offer a
second, alternative calibration such that Dt = 0.5 when T = 4. Science and impact
studies tell us that, not only could we cross several key physical tipping points in the
climate system by the time the 4°C mark is reached (Lenton et al., 2008), the
impacts of such warming on the natural environment, economies and societies
could be severe, with reason to believe in the risk of vast movements of population
and associated conflict, unrest and loss of life (Stern, 2013). Global mean
temperatures regularly exceeding 4°C above pre-industrial have probably not been
seen for at least 10 million years (Zachos et al., 2008) and are within the range of
difference between today and the peak of the last Ice Age, when large ice sheets
covered northern Europe and North America (IPCC, 2013), radically influencing
where people could be. Given the potential magnitude of transformation illustrated
by this example, the assumption that Dt = 0.5 when T = 4 may be no less plausible,
to put it cautiously, than assuming, as (2) does with the standard parameterisation,
that Dt = 0.04 when T = 4, i.e. only 4% of output is lost as a result of temperatures
not seen for 10 million plus years.
In our first growth model, we partition damage as expressed in (40 ) between
damage affecting output D Y and those affecting capital D K , while in our second
model damage are partitioned between output and TFP as in (3). We do so in a
similar way to Moyer et al. (forthcoming) and the procedure is described in detail
in Appendix A.

10
Note that within this set of studies are some estimates of the money value of direct welfare losses due to
climate change, e.g. impacts on health and the amenity value of the environment.
11
Which is why Nordhaus himself describes such functions and the assumptions they embody about
damage at different temperatures as ‘placeholders subject to further research’ (Stern, 2013). However, we
will see data points of 4, 5 or 6°C, if we are negligent and unlucky, within decades. Hence, it makes sense to
try different formulations as representing different possibilities, including of the extremely damaging
circumstances the science suggests as possible.
12
A quadratic function could not be made to simultaneously fit the existing data, while satisfying this
additional assumption; it would give excessive damage for smaller temperature increases.
13
Elsewhere Dietz et al. (2007a,b,c); Stern (2007, 2008) we investigated models based on the PAGE IAM, in
which damage was a power function of temperature. We examined the sensitivity of damage to the exponent
of the power function up to a value of three.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 583
3.3. Climate Risk
Our last extension to the basic framework involves the climate sensitivity parameter.
We take two approaches here. First, we explore high values of this parameter in
sensitivity analysis. Second, we replace its sure value with a probability density function
(pdf). Climate sensitivity is a key factor in driving the change in temperature in DICE,
as it is in many other simple climate models. Thus, it is a natural example of large-scale
risk. Others would be relevant too, such as the scale of damage for a given temperature

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


increase, the scale of loss of life and so on.
The equation of motion of temperature is given by:
 
F2CO 2  
Tt ¼ Tt1 þ j1 Ft  ðTt1 Þ  j2 Tt1  Tt1 ;
LO
(5)
S
where Ft is radiative forcing, F2CO2 is the radiative forcing resulting from a doubling
in the atmospheric stock of carbon dioxide, S is the climate sensitivity, T LO is the
temperature of the lower oceans, j1 is a parameter determining speed of adjustment
and j2 is the coefficient of heat loss from the atmosphere to the oceans. Calel
et al. (2014) contains a detailed explanation of the physics behind this equation.
In standard DICE S = 3°C. However, it has long been known that there is substantial
uncertainty about S (Charney, 1979). Moreover investigations in recent years (as
collected by Meinshausen et al., 2009) have tended to yield estimates of the pdf of S
that have a large positive skew and in most cases the right-hand tail can indeed be
defined as ‘fat’.14 In the latest IPCC report (IPCC, 2013), a subjective pdf is offered
that is the consensus of the panel’s many experts. According to this distribution, S is
‘likely’ between 1.5 and 4.5°C, where likely corresponds to a subjective probability of
anywhere between 0.66 and 1. It is ‘extremely unlikely’ to be less than 1C, where
extremely unlikely indicates a probability of ≤ 0.05, while it is ‘very unlikely’ to exceed
6°C, where this denotes a probability of ≤ 0.1. We thus choose values of S 2 {1.5, 3, 6}
for sensitivity analysis.
For our stochastic modelling we fit a continuous pdf to these data, using the mid-
points of the IPCC probability ranges. In doing so, we face a choice over the type of
function to fit. We performed a test of the fit of various functional forms, in terms of
root-mean-square error, to the IPCC probability statements and found that the log-
logistic function demonstrated the best fit among those we examined. The log-logistic
function also has the advantage of having a tail of intermediate ‘fatness’ relative to
other forms, thus, in this sense, it constitutes a middle-of-the-road assumption:
 a1
S
a
b
f ðSÞ ¼   a  ; (6)
S
b 1þ
b
where a  4.2 and b  2.6 are the shape and scale parameters respectively giving mean
S of 2.9, a standard deviation of 1.4 and the 95th percentile at 5.3.

14
Where the density in the upper tail approaches zero more slowly than the exponential distribution.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
584 THE ECONOMIC JOURNAL [MARCH
It is worth emphasising, before moving on to the results, that there are other
potentially significant sources of risk attending to the impacts of greenhouse gas
emissions. Some of these are in the climate system – for instance the effective heat
capacity of the oceans (Calel et al., 2014) – yet a focus on S captures the essence of
physical climate risk in a clear and simple way. Other sources of risk relate to damage
for any given temperature and could also be modelled with probabilities, were the
evidence to justify doing so. However, as we have argued, the damage functional form
and parameterisation are currently very poorly constrained by evidence, and therefore

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


it seems appropriate to instead pursue this, potentially very important source of risk,
via a more simple sensitivity analysis on different functions as proposed in subsection
3.2.

4. Results
4.1. Baseline
At the heart of this exercise is an investigation into the prospects for growth and damage
in a changing climate. Figure 1 plots baseline consumption per capita – i.e. in the
absence of controls on carbon dioxide emissions imposed by a social planner – under
various scenarios over the next two centuries. The upper panel plots the forecasts of the
model with production (1.K) and damage from climate change on the capital stock,
while the lower panel plots the forecasts of the model with production (1.TFP), where
TFP growth is endogenous and where climate change reduces TFP.
The ‘standard’ trajectory represents the forecast of the standard DICE model
without the various extensions we are considering in this article. The starting year is
2005. It is of course the same in both panels and notice immediately by how much
consumption per capita increases in it, powered largely by exogenous productivity
growth15 – in 2205 it is more than 15 times the 2005 level. This is despite a large
increase in the atmospheric stock of carbon dioxide and in the global mean
temperature (discussed below). Without large assumed improvements in the exoge-
nous element of TFP, the increase in per capita consumption would be much smaller.
Changing the model of growth begins to yield more pessimistic forecasts, although it
does not by itself qualitatively alter the tendency for the future to be much better off
than the present. Under the model with capital damage, consumption/head in 2205 is
13.3 times higher than in 2005, while under the model of productivity damage it is 11.4
times higher. Since the total damage multiplier Dt in (4) is the same in the two models,
simply being partitioned differently between damage on output, capital and TFP (see
Appendix A), the larger effect in the model of productivity damage partly reflects the
longer lasting impact of climate change in this model, where depreciation of
productivity is slow compared with capital.
The divergence in forecasts is much more marked, however, when we layer on
greater convexity of damage as in (40 ). With Weitzman’s (2012) calibration,
consumption per capita grows much more slowly after 2150 in the model of capital

15
With no growth in labour, the long-run output growth rate implied by (1) is simply that of exogenous
TFP.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 585
100 K Model
Standard
90 S = 3, Quardratic Damage
S = 3, Weitzman Damage
Consumption Per Capita (US $2005)

80 S = 3, High Damage
S = 6, Weitzman Damage
70

60

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


50

40

30

20

10

0
2000 2050 2100 2150 2200

100 TFP Model

90
Consumption Per Capita (US $2005)

80

70

60

50

40

30

20

10

0
2000 2050 2100 2150 2200

Fig. 1. Baseline Consumption Per Capita, 2005–2205


Notes. The upper panel corresponds to the model with capital damage and with knowledge
proportional to the aggregate capital stock, while the lower panel corresponds to the model of
endogenous TFP Growth and TFP damage.

damage, while in the model of TFP damage it peaks around 2150 before actually falling
thereafter. By 2205, it is only 8.3 and 5.8 times higher respectively than today. If the
damage function is set such that damage equivalent to 50% of global output are
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
586 THE ECONOMIC JOURNAL [MARCH
assumed to occur upon 4°C warming, the collapse in living standards is much stronger,
with consumption/head peaking before the end of this century and ending up in both
models around or below the present level in real terms. A similar forecast is generated
by Weitzman damage, when we instead increase the climate sensitivity parameter S to
6°C, which has a probability, as described above, of up to 0.1 according to IPCC. The
two growth models yield similar forecasts in these cases, demonstrating the diminished
importance of growth assumptions when instantaneous damage are severe and or
warming is very rapid.

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


Changes in the atmospheric stock of carbon dioxide and global mean temper-
ature, which drive these growth prospects, are shown in Figures B1 and B2
respectively in Appendix B. Baseline emissions will take the atmospheric stock of
carbon dioxide to nearly 800 ppm by the end of this century in all the scenarios
considered. The stock continues to increase after 2100 but there is some feedback
of climate damage on emissions, which works through the depressive effect of
climate damage on growth and of growth on emissions. The principal determinant
of global mean temperature is the value of the climate sensitivity parameter. With
the typical central estimate of 3°C, the global mean temperature is forecast to be in
the region of 3.5°C above the pre-industrial level by 2100, while if S = 6 it could be
more than 5°C above pre-industrial.

4.2. Optimal Controls


We now move to examining the optimal controls on emissions, set by a social planner.
As Appendix A explains, the social planner’s objective is to maximise the sum over time
of discounted total utility by choosing a set of emissions control quantities and prices
from 2015 until 2245, with a given abatement cost function (see Appendix A). We
present results covering the rest of this century. Table 1 lists the optimal emissions
control rate (the percentage or fractional reduction in emissions from the baseline)
under various scenarios, while Table 2 does the same for the optimal carbon price.16 It
is clear from the Tables that modifying the growth model and the associated pathways
through which climate change can affect the economy, as well as increasing the
convexity of the damage function, and increasing the climate sensitivity, can
significantly increase the optimal emissions control rate and the associated carbon
price, both initially and throughout.
Let us focus on initial control quantities and prices – these give us something
with which to compare current global policy efforts and debates. In standard DICE
the emissions control rate, that is the percentage reduction in industrial carbon
dioxide emissions, is 0.158 in 2015, with an associated carbon price of $44/tC in
2005 prices (divide by roughly 3.7 to obtain estimates/tCO2, and multiply by c. 1.16
to bring up to 2012 prices17). If we switch from this standard model of exogenous

16
Where the optimal carbon price is defined as the marginal cost of abatement at the optimal emissions
level calculated. Whether it is reasonable to interpret this as a price depends on the convexity of the
abatement cost curve, i.e. it depends on there being rising marginal costs. It has been contended that
marginal costs do not rise but these are issues for another paper.
17
World Bank data on GDP deflator, from http://data.worldbank.org/indicator/ NY.GDP.DEFL.KD.ZG,
retreived on 22 November 2013.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 587
Table 1
Optimal Emissions Control Rate, 2015–2105. S = 3 Unless Otherwise Indicated

2015 2025 2035 2045 2055 2065 2075 2085 2095 2105
Standard 0.158 0.184 0.211 0.240 0.270 0.302 0.335 0.370 0.407 0.446
Capital models
Quadratic damage 0.213 0.289 0.356 0.424 0.495 0.565 0.636 0.706 0.777 0.848
Weitzman damage 0.235 0.322 0.401 0.484 0.568 0.650 0.730 0.805 0.875 0.944
Weitzman damage, S = 6 0.360 0.494 0.619 0.751 0.883 1 1 1 1 1

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


High damage 0.342 0.471 0.591 0.709 0.814 0.901 0.970 1 1 1
Productivity models
Quadratic damage 0.272 0.365 0.444 0.528 0.614 0.702 0.792 0.882 0.974 1
Weitzman damage 0.290 0.392 0480 0.573 0.667 0.761 0.852 0.942 1 1
Weitzman damage, S = 6 0.432 0.584 0.722 0.868 1 1 1 1 1 1
High damage 0.396 0.538 0.663 0.783 0.891 0.984 1 1 1 1

Table 2
Optimal Carbon Prices (2005 US$/tC), 2015–2105. S = 3 Unless Otherwise Indicated

2015 2025 2035 2045 2055 2065 2075 2085 2095 2105
Standard 44.4 57.0 71.2 87.8 106.2 127.1 150.0 175.8 204.6 236.6
Capital models
Quadratic damage 76 129 182 245 316 393 476 563 656 752
Weitzman damage 91 156 226 310 405 506 609 711 812 912
Weitzman damage, S = 6 196 337 495 684 895 1097 1074 1052 1032 1012
High damage 178 309 455 617 774 909 1017 1052 1032 1012
Productivity models
Quadratic damage 118 196 272 363 466 580 705 840 984 1012
Weitzman damage 133 222 313 420 541 670 806 945 1032 1012
Weitzman damage, S = 6 271 456 653 888 1121 1097 1074 1052 1032 1012
High damage 233 393 559 738 911 1066 1074 1052 1032 1012

growth to (1.K) with capital damage, the optimal emissions control rate rises to
0.213 (optimal carbon price = $76/tC). Further extending this model to
incorporate highly convex damage with Weitzman’s (2012) parameterisation, it
rises to 0.235 (optimal price = $91/tC), while with our high damage function
scenario it is 0.342 (optimal price = $178/tC). When Weitzman damage are
combined with a high climate sensitivity, the optimal control rate is 0.36, brought
about by an optimal price levied at $196/tC. Some caution should be exercised,
however, in interpreting the relevance of these strong initial control rates and
prices, because DICE, as a model of medium and long-run dynamics, lacks
adjustment costs, which could render such a rapid decarbonisation infeasible.
In the endogenous growth model (1.TFP) where instantaneous climate damage
work on TFP as well as output, the increase in the controls is even stronger. With
quadratic damage, the optimal control rate on emissions is 0.272 with an associated
carbon price of $118/tC. Moving to Weitzman damage increases this to 0.29
(optimal carbon price = $133/tC), while with our high damage function scenario
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
588 THE ECONOMIC JOURNAL [MARCH
the controls are respectively 0.396 and $233/tC. When Weitzman damage is
combined with a high climate sensitivity they are respectively 0.432 and $271/tC.
Notice for both growth models the marked rise in the carbon price when we
move from Weitzman to high damage or from S = 3 to S = 6, which reflects
convexity in the marginal abatement cost function. Nonetheless, the same remarks
regarding adjustment costs and their potential effect on the optimal controls apply
here.
Figures B3 and B4 in Appendix B show the consequences of the optimal controls

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


for global mean temperature and the atmospheric stock of CO2. Compared with the
baseline, it can be seen that maximisation of social welfare implies significant
reductions in both climate variables. With Weitzman damage, the build-up of
atmospheric CO2 is limited to 524 ppm in the model of capital damage and 489
ppm in the model of TFP damage, while with the more pessimistic parameterisation
of the damage function the corresponding maximum concentrations are 459 and
444 ppm. These numbers are broadly in line with the types of stabilisation
concentrations recommended by many scientists. In sharp contrast, with standard
DICE the optimal emissions controls allow atmospheric CO2 to rise throughout this
century and peak at around 735 ppm in the middle of the next century. The
resulting warming depends on the climate sensitivity.

4.3. Optimal Control Under Stochastic Warming


Thus, far we have computed the optimal controls, contingent on a set of point values of
the climate sensitivity parameter, S 2 {1.5, 3, 6}. A fuller specification of climate risk
involves characterising a probability distribution over different values of S, as we
described in Section 3, and solving for the optimal path of emissions controls. The
planner’s problem is specified as maximising expected social welfare, where expec-
tations are formed before the first period commences and are not revised (see
Appendix A for further details of the optimal control problem).18 Expected values are
formed in a Monte Carlo simulation, sampling (via the Latin Hypercube method) 500
times from f (S) in (6).
Tables 3 and 4 report the optimal control quantities and prices respectively for the
two growth models, each run with the various different damage functions. Since this
exercise constitutes a fuller specification of climate risk, these might be considered our
headline results. Notice that, comparing them with Tables 1 and 2, the effect of
randomising S depends on the damage function – the optimal controls are higher
under random S, given Weitzman or high damage, but lower given quadratic damage.
Remember that f (S) in (3) is not a mean-preserving spread around S = 3. Rather,
mean S is 2.9 and, as a distribution with a large positive skew, significantly more than
half of the probability mass lies below the mean. When one bears in mind that what

18
In line with much of the literature, we simplify the problem by omitting the possibility of learning about
the climate sensitivity from observations obtained after the first period has commenced. So the planner must
stick to optimal controls computed at the outset, a so-called open-loop control. Were it possible to learn
about climate sensitivity from observations and to change policy settings in response – a closed-loop policy –
the planner could of course achieve at least as high a level of social welfare, most probably much higher.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 589
Table 3
Optimal Emissions Control Rate Under Random S, 2015–2105

2015 2025 2035 2045 2055 2065 2075 2085 2095 2105
Standard 0.158 0.184 0.211 0.240 0.270 0.302 0.335 0.370 0.407 0.446
Capital models
Quadratic damage 0.204 0.277 0.341 0.406 0.474 0.543 0.611 0.681 0.750 0.819
Weitzman damage 0.250 0.344 0.434 0.532 0.631 0.732 0.831 0.925 0.993 0.993
High damage 0.393 0.542 0.688 0.841 0.986 0.998 0.998 0.998 0.999 0.999

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


Productivity models
Quadratic damage 0.261 0.350 0.427 0.508 0.591 0.677 0.765 0.854 0.944 1
Weitzman damage 0.307 0.417 0.518 0.627 0.743 0.862 0.982 0.999 1 1
High damage 0.481 0.656 0.825 1 1 1 1 1 1 1

Table 4
Optimal Carbon Prices (2005 US$/tC) Under Random S, 2015–2105

2015 2025 2035 2045 2055 2065 2075 2085 2095 2105
Capital models
Quadratic damage 70 119 169 226 293 365 442 528 614 707
Weitzman damage 101 176 261 368 490 625 769 914 1019 999
High damage 229 399 598 838 1093 1092 1071 1050 1030 1010
Productivity models
Quadratic damage 110 181 253 338 435 543 664 792 930 1012
Weitzman damage 147 248 359 494 657 839 1040 1050 1032 1012
High damage 329 563 830 1146 1121 1097 1074 1053 1032 1012

ultimately matters is the pdf of consumption per capita that results from f (S), it should
start to become clear that, when the damage function has modest curvature, the effect
of randomising S on the optimal controls can be to lower them, but when the damage
function has strong curvature the opposite is true, because the tail of high
temperatures exerts an ever larger relative effect on consumption per capita, utility
and social welfare.
Figures B5 and B6 in Appendix B show the consequences of the optimal controls
for the atmospheric stock of CO2 and global mean temperature respectively.19
Figure B5 shows that the optimal mean stock of atmospheric CO2 peaks in our
endogenous growth models at no more than about 500 ppm, and as little as 420
ppm, depending on the growth model and damage function. These stock levels are
well below those in the standard DICE model. Those combinations of growth model

19
Since the climate sensitivity is uncertain, so, obviously, is the change in the global mean temperature,
and since this goes on to affect emissions via damage, there is also some uncertainty in the longer run about
the atmospheric stock of CO2. Therefore, both figures report mean values from the Monte Carlo simulation.
In the case of the atmospheric stock of CO2, the uncertainty is very small (no more than 1 ppm), but in the
case of global mean temperature it is considerably larger. Therefore, in the latter case we also show the 90%
confidence interval, in 2205, from the Monte Carlo simulation to the right of the main chart.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
590 THE ECONOMIC JOURNAL [MARCH
and damage function yielding higher climate impacts support a lower optimal stock.
Compared with Figure B3 we can see that the optimal stock is lower under random
S than when S = 3. Figure B6 shows that mean temperature is kept to a maximum
of around 2°C except in two cases. First, in model (1.K) with capital damage, when
the damage function is quadratic, mean warming peaks at around 2.5°C early next
century. Second, in standard DICE mean warming peaks at c. 3.5°C. Notice the
spreads around mean warming and in particular the very large 90% confidence
interval around warming in standard DICE, where the 90th percentile reaches as

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


much as 5.6°C. Optimal emissions controls in our extended models of DICE cut
this tail of high temperatures significantly, due to their damaging consequences in
the short and long run.

5. Conclusions
‘To slow or not to slow’ (Nordhaus, 1991) and its subsequent development into the
dynamic DICE model have given us what seems to be a coherent and powerful
framework for assessing the costs and benefits of climate-change mitigation. But it
has in-built assumptions on growth, damage and risk, which together result in gross
underassessment of the overall scale of the risks from unmanaged climate change
(Stern, 2013). This criticism applies with just as much force to most of the other
IAMs that DICE has inspired. The purpose of this article has been to show how
these unrealistic assumptions might be relaxed and what would be the conse-
quences of doing so, in terms of optimal emissions reductions and carbon prices,
atmospheric concentrations of carbon dioxide and global mean temperature.
The first assumption we have relaxed is that the underlying drivers of economic
growth are exogenous and unaffected by climate change. Instead we look at two
models of endogenous growth, in which the damage from climate change affect the
drivers of long-run growth, not just current output. The second assumption we have
relaxed is that the damage function relating instantaneous climate damage to the
increase in global mean temperature is only weakly convex. Instead, we allow for
the possibility that instantaneous damage increase rapidly, particularly once the
global mean temperature reaches 4–6°C above the pre-industrial level. We
suggest this representation is more plausible, given the scale of change that such
warming could bring; at the very least, simulations based on weak convexity should
not dominate our attention as they have come to do. The third assumption we have
relaxed is that the climatic response to greenhouse gas emissions is moderate and
moreover is precisely understood. Very few, if any, commentators would
explicitly claim that climate sensitivity is precisely understood, of course.
Nonetheless, most economic modelling is undertaken using only a single, central
estimate of the climate sensitivity parameter, fixed in the centre of the distribu-
tion of available estimates from the science. We explore risk in this crucial
parameter.
Overall, the scale of the risks from unmanaged climate change in this modelling
framework is the convolution of these three extensions. We show that, with the
models extended in this way, business-as-usual trajectories of greenhouse gas
emissions give rise to potentially large impacts on growth and prosperity in the
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 591
future, especially after 2100. Indeed, these impacts are large enough to feed back
into future emissions via reduced activity but the feedback is too small and too late
for the system to self-regulate. Thus, optimal emissions control is strong and
strongly increasing. As a guide, we find that these models suggest the carbon price
in a setting of globally coordinated policy, such as a cap-and-trade regime or a
system of harmonised domestic carbon taxes, should be in the range $32–103/tCO2
(2012 prices) in 2015. It must be remembered that the DICE model lacks
adjustment costs, so the high end of the range should be interpreted cautiously. On

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


the other hand, and potentially of great importance, we have, notwithstanding our
extensions, omitted important risks in relation to the distribution of damage, which
could give higher carbon prices. Within two decades, the carbon price should rise
in real terms to $82–260/tCO2. Doing so would, according to the model, keep the
expected atmospheric stock of carbon dioxide to a maximum of c. 425–500 ppm
and the expected increase in global mean temperature to c. 1.5–2°C above pre-
industrial.
The study is only a preliminary investigation, whose purpose was to illustrate or
sketch the consequences of relaxing assumptions that have limited plausibility and
possible large effects on policy conclusions. We have, for instance, restricted our
attention to knowledge, accumulated through learning-by-doing, as the driver of
long-run growth, though other sources of growth are important and other models
might be deployed. Our exploration of the implications of risk has, for the sake of
clarity, been limited to the climate sensitivity, though other sources of great risk exist
in the physical climate system, not to mention in the economy. The models that we
do use require the choice of parameter values, about some of which there is
currently very little relevant empirical evidence. Given slow rates of learning about
some IAM parameters, this should be regarded as an endemic problem, however
(Pindyck, 2013). It is not the case that the standard model parameters are well
constrained, whereas the new parameters we introduce are not. Future work building
on our framework should also pay attention to the costs of rapid adjustment to a low-
carbon economy and possible limits to the speed of decarbonisation. This work will
need to go well beyond the choice of parameter values to consider new model
structures.
This is not an article about the sensitivity of results to pure-time discounting, or
other parameters and structures relevant to discounting. As we found in the
technical Appendix to Stern (2007) and in Dietz et al. (2007a,c), lower pure-time
discounting does indeed favour stronger and earlier action to curb emissions. Those
results were from the ‘PAGE’ IAM (Hope, 2006) but we know from other work that
this is also true of DICE (Nordhaus, 2007). We have argued elsewhere that careful
scrutiny of the ethical issues around pure-time discounting points to lower values
than are commonly assumed (usually with little serious discussion). Pure-time
discounting is essentially discrimination by date of birth in the sense that a life,
which is identical in all respects (including time patterns of consumption) but
happens to start later, has a lower value. If, for example, the pure-time discount
rate were 2%, a life starting 35 years later, but otherwise the same, would have half
the value of a life starting now. The time horizon essential to a discussion of
climate change makes careful examination of these ethical issues unavoidable.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
592 THE ECONOMIC JOURNAL [MARCH
Preliminary calculations indicate that low pure-time discounting will significantly
increase the optimal controls in this article as well.
One cannot and should not expect a single model to capture all relevant issues and
neither should we be able to resolve all difficulties within a single framework.20 It is
enough for a model to help raise and understand key aspects of a problem. This means
that we should be grateful to Bill Nordhaus for providing one helpful vehicle. As it is
expanded and different perspectives are brought in, including the possibility of major
loss of life from climate change, then we would suggest the arguments for strong action

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


will look still stronger.

Appendix A. Extended Model Description


Here, we offer an extended description of the DICE model, focusing on the major model
equations and in particular on our modifications. Even more detail can be found on Nordhaus’s
model website at http://www.econ.yale.edu/~ nordhaus/homepage/index.html. Our analysis is
based on the 2010 version of the model.
The model represents a social planner maximising a classical utilitarian objective functional by
choosing the rate of control of industrial carbon dioxide emissions:

X
Tmax
max W ¼ uðct ÞLt ð1 þ qÞt ;
flt gTt¼1
max
t¼0

where l 2 [0, 1] is the emissions control rate, u(ct) is the instantaneous social utility of
consumption per capita at time t and q = 0.015 is the utility discount rate. Note that c is not only
time-dependent as the above equation implies, it is also state-dependent when we undertake
stochastic modelling. We suppress notation of state-dependence for simplicity; bear in mind that,
when running the model with a random parameter (the climate sensitivity S in (3)), we take the
expectation of social welfare. Tmax is the terminal period, which is 2595. The model proceeds in
time steps of ten years from 2005, so appropriate interpretations must be made in considering
the various equations of motion. Notice that, since 2005 is in the past, our first control period is
t = 1, i.e. 2015.21
The utility function is iso-elastic,
ct1g
uðct Þ ¼ ;
1g

where g is the elasticity of marginal social utility of consumption and is set to 1.5 to allow
comparison with standard DICE.
As set out in the main body of the article, we explore two alternative production functions:

Yt ¼ ð1  DtY Þð1  Kt ÞAt Ktaþb Lt1a ; (1.K)

Yt ¼ ð1  DtY Þð1  Kt ÞAt Kta Lt1a : (1.TFP)

20
For example, the paradoxes of social choice theory can be better understood by broadening the
philosophical perspective and are not easily resolved within the standard framework (Sen, 2009).
21
In fact, we need only solve lt from 2015 to 2245 inclusive, since DICE assumes that from 2255 onwards
lt = 1, because a zero-emissions backstop energy technology becomes competitive.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 593
The capital elasticity a = 0.3, while the elasticity of output with respect to knowledge, b = 0.3
(Mankiw et al., 1995).
The equation of motion of capital in model (1.K) is

Ktþ1 ¼ ð1  DtK Þð1  dK ÞKt þ It ;

where dK = 0.1, while in model (1.TFP) we simply drop (1D K ).


Output is either consumed or invested,

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


Y t ¼ Ct þ I t

where Ct = ctLt is aggregate consumption and It = sYt, where s = 0.23 is the savings rate
(calibrated to long-run average optimal savings in standard DICE, without climate damage and
emissions abatement costs and, in principle, some private inter-temporal objectives). We specify
exogenous, constant savings in order to capture in a simple way the second-best context implied
by fitting our models of endogenous growth to current macroeconomic data. In growth models
with knowledge spillovers, the savings rate chosen by a planner will be greater than the savings
rate emerging from a decentralised equilibrium of firms and households, because the marginal
private return to investment does not include the spillovers.
A more elaborate analysis would permit households to choose their optimal savings rate in
equilibrium with firms’ private marginal product of capital (in response to the planner’s
emissions controls), but it is worth noting that, in standard DICE, endogenising the savings rate
has been shown to make little difference to the optimal policy (Fankhauser and Tol, 2005; and
Nordhaus’s laboratory notes on DICE), so our simplification is unlikely to matter.22 In any case,
whether households are currently taking into account the effects of climate policy on future
consumption prospects when choosing how much to save is unclear.
In model (1.TFP), productivity is endogenous and its equation of motion is

Atþ1 ¼ ð1  DtA Þð1  dA ÞAt þ c1 It 2 ;


c
(3)

where dA = 0.01 is the rate of depreciation of the stock of TFP, while c1  0.0003 and
c2  0:373 are parameters of the spillovers function. c1 and c2 are calibrated so that output
in (1.TFP), in the absence of climate damage and emissions abatement costs, is the
same as in standard DICE. In model (1.K), TFP is an exogenous time series, so (3) does not
apply.
The climate damage function is

Dt ¼ 1  1=ð1 þ p1 Tt þ p2 Tt2 þ p3 Tt6:754 Þ; (4’)

where p1 = 0 and p2  0.00284 throughout. p3 = 0 when we compute results for the standard
setting (i.e. Dt ¼ D^t ),  5.07 9 106 when we use Weitzman’s parameterisation, or
 8.199105 according to our high damage specification, where Dt is assumed to be equal to
0.5 when the atmospheric temperature is 4°C above the pre-industrial level.
Damage is then partitioned between output and capital, or output and TFP, depending on the
growth model:
Dti ¼ f i  Dt ;
ð1  Dt Þ
DtY ¼ 1 
ð1  Dti Þ

22
See also Mirrlees and Stern (1972), who first illustrated this feature in simple optimal growth models.
© 2015 The Authors.
The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
594 THE ECONOMIC JOURNAL [MARCH
where f is the share of damage to i = A or i = K. IAMs do not in general explicitly address the
allocation of damage between capital and output, and vary widely in what they implicitly assume
about it. Nordhaus and Boyer (2000) analysis might be read to suggest that f K is in the region of
1/3, so 0.3 is the value we choose. The calibration problem is even more acute in the case of
allocating damage between output and TFP – there are, as Moyer et al. (forthcoming) also point
out, currently severe modelling, data and estimation problems in carrying out such an allocation.
Moyer et al. (forthcoming) consequently explore a range of values of f A between 1% and 100%.
We make the relatively conservative assumption that f A = 0.05.
The total abatement cost function is

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


Kt ¼ h1;t lht 2 ;

where h1,t is a time-varying coefficient and h2 ¼ 2:8, hence marginal abatement costs are
increasing in emissions control.
Cumulative industrial carbon dioxide emissions are constrained by remaining fossil fuel
reserves,

X
Tmax
EIND
t  C Cum;
t¼0

where C Cum = 6000 gigatonnes of carbon is the constraint, and total emissions of carbon are
the sum of industrial emissions of carbon dioxide and exogenous emissions of carbon dioxide
from land use:

Et ¼ EIND
t þ ELAND
t :

Industrial carbon dioxide emissions at time t are proportional to gross output in the same
period, hence there is a different function depending on the growth model:

EIND
t ¼ rt ð1  lt ÞAt Ktaþb Lt1a ; or
¼ rt ð1  lt ÞAt K a L 1a ;
t t

where rt is the ratio of uncontrolled emissions to output and is an exogenous, time-varying


coefficient. It is assumed that ∂r/∂t < 0, representing autonomous improvements in carbon
productivity that arise from technical progress and structural change, and that ∂2r/∂t2 > 0.
The atmospheric stock of carbon is driven by total emissions, in a system of three equations
representing the cycling of carbon between three reservoirs, the atmosphere MAT, a quickly
mixing reservoir comprising the upper ocean and parts of the biosphere MUP, and the lower
ocean MLO:

MtAT ¼ Et þ /11 Mt1


AT
þ /21 Mt1
UP

MtUP ¼ /12 Mt1


AT
þ /22 Mt1
UP
þ /32 Mt1
LO

MtLO ¼ /23 Mt1


UP
þ /33 Mt1
LO
:

Cycling is determined by a set of coefficients /jk that govern the rate of transport from
reservoir j to k per unit of time.
The change in the atmospheric stock of carbon from the pre-industrial level determines
radiative forcing,
 
M AT
Ft ¼ F2CO2  log2 t þ FtEX ;
Md AT

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 595
where MdAT is the stock of carbon in the atmosphere before the industrial revolution (i.e. in

1750) and FtEX is exogenous radiative forcing (capturing among other things the forcing due to
greenhouse gases other than carbon dioxide) and is time-dependent. The equation of motion of
temperature is given by:
 
F2CO2  
Tt ¼ Tt1 þ j1 Ft  ðTt1 Þ  j2 Tt1  Tt1
LO
; (5)
S

where T LO is the temperature of the lower oceans and evolves according to:

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


 
TtLO ¼ Tt1
LO
þ j3 Tt1  Tt1
LO
:

Appendix B. Further Results

1200
Atmospheric Concentration of Carbon Dioxide (ppm)

1000

800

Standard
600 TFP Model; Weitzmen Damage
TFP Model; High Damage
Capital Model; Weitzman Damage
400 Capital Model; High Damage

200

0
2000 2050 2100 2150 2200

Fig. B1. Baseline Atmospheric Stock of Carbon Dioxide, 2005–2205. S = 3

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
596 THE ECONOMIC JOURNAL [MARCH
8.0 K Model
s=6
Atmospheric Temperature (Degrees Celsius 7.0 Standard
s=3
s = 3, High Damage
6.0 s = 1.5
Above Pre-industrial)

5.0

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


4.0

3.0

2.0

1.0

0.0
2000 2050 2100 2150 2200

7.0 TFP Model


Atmospheric Temperature (Degrees Celsius

6.0

5.0
Above Pre-industrial)

4.0

3.0

2.0

1.0

0.0
2000 2050 2100 2150 2200

Fig. B2. Baseline Global Mean Temperature (Degrees Centigrade Above Pre-industrial), 2005–2205
Notes. The upper panel corresponds with the model of capital damage, while the lower panel
corresponds with the model of TFP damage. The damage function calibration is ‘Weitzman’
unless otherwise indicated.

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 597
800
Atmospheric Concentration of Carbon Dioxide (ppm)
700

600

500

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


400

300
TFP Model; Weitzman Damage
200 TFP Model; High Damage
Capital Model; Weitzman Damage
Capital Model; High Damage
100 Standard

0
2000 2050 2100 2150 2200

Fig. B3. Optimal Atmospheric Stock of Carbon Dioxide, 2005–2205. S = 3

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
598 THE ECONOMIC JOURNAL [MARCH
4.0 K Model
S=6
Atmospheric Temperature (Degrees Celsius 3.5 S=3
S = 3, High Damage
Standard
3.0
Above Pre-industrial)

2.5

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


2.0

1.5

1.0

0.5

0.0
2000 2050 2100 2150 2200

4.0 TFP Model


Atmospheric Temperature (Degrees Celsius

3.5

3.0
Above Pre-industrial)

2.5

2.0

1.5

1.0

0.5

0.0
2000 2050 2100 2150 2200

Fig. B4. Optimal Global Mean Temperature (Degrees Centigrade Above Pre-industrial), 2005–2205
Notes. The upper panel corresponds with the model of capital damage, while the lower panel
corresponds with the model of TFP damage. The damage function calibration is ‘Weitzman’
unless otherwise indicated.

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 599
800
Atmospheric Concentration of Carbon Dioxide (ppm)
700

600

500

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


400

300 TFP Model; Weitzman Damage


TFP Model; High Damage
Capital Model; Weitzman Damage
200 Capital Model; High Damage
Standard
100

0
2000 2050 2100 2150 2200

Fig. B5. Optimal Atmospheric Stock of Carbon Dioxide, 2005–2205, Mean Over Random S

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
600 THE ECONOMIC JOURNAL [MARCH

Atmospheric Temperature (Degrees Above Pre-industrial)


6
Quadratic K Model
Weitzman
5 High
Standard DICE

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


3

0
2000 2050 2100 2150 2200
Atmospheric Temperature (Degrees Above Pre-industrial)

6
TFP Model

0
2000 2050 2100 2150 2200
Fig. B6. Optimal Global Mean Temperature (Degrees Centigrade Above Pre-industrial), 2005–2205, Mean
Over Random S
Notes. The upper panel corresponds with the model of capital damage, while the lower panel
corresponds with the model of TFP damage. The bars on the right-hand side give the 90%
confidence interval in 2205.

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 601

London School of Economics

References
Agrawala, S. (1998). ‘Context and early origins of the intergovernmental panel on climate change’, Climatic
Change, vol. 39(4), pp. 605–20.
Arrow, K.J. (1962). ‘The economic implications of learning by doing’, Review of Economic Studies, vol. 29(3),
pp. 155–73.
Calel, R., Stainforth, D.A. and Dietz, S. (2014). ‘Tall tales and fat tails: the science and economics of extreme

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


warming’, Climatic Change. DOI: 10.1007/s10584-013-0911-4.
Charney, J.G. (1979). Carbon Dioxide and Climate: A Scientific Assessment, Washington, DC: National Academy of
Science.
Cline, W.R. (1992). The Economics of Global Warming, Washington, DC: Peterson Institute.
de Bruin, K.C., Dellink, R.B. and Tol, R.S. (2009). ‘Ad-dice: an implementation of adaptation in the dice
model’, Climatic Change, vol. 95(1–2), pp. 63–81.
Dell, M., Jones, B.F. and Olken, B.A. (2012). ‘Temperature shocks and economic growth: evidence from the
last half century’, American Economic Journal: Macroeconomics, vol. 4(3), pp. 66–95.
Dietz, S., Anderson, D., Stern, N., Taylor, C. and Zenghelis, D. (2007a). ‘Right for the right reasons: a final
rejoinder on the Stern review’, World Economics, vol. 8(2), pp. 229–58.
Dietz, S., Hope, C. and Patmore, N. (2007b). ‘Some economics of dangerous climate change: reflections on
the Stern review’, Global Environmental Change, vol. 17(3), pp. 311–25.
Dietz, S., Hope, C., Stern, N. and Zenghelis, D. (2007c). ‘Reflections on the Stern review (1): a robust case for
strong action to reduce the risks of climate change’, World Economics, vol. 8(1), pp. 121–68.
Fankhauser, S. and Tol, R. S. (2005). ‘On climate change and economic growth’, Resource and Energy
Economics, vol. 27(1), pp. 1–17.
Greenaway, D. (1991). ‘Economic aspects of global warming: editorial note’, ECONOMIC JOURNAL, vol. 101
(407), pp. 902–3.
Gupta, J. (2010). ‘A history of international climate change policy’, Wiley Interdisciplinary Reviews: Climate
Change, vol. 1(5), pp. 636–53.
Hall, B.H., Mairesse, J. and Mohnen, P. (2009). ‘Measuring the returns to R&D’, Technical Report, National
Bureau of Economic Research.
Hope, C. (2006). ‘The marginal impact of CO2 from page 2002: an integrated assessment model
incorporating the ipcc’s five reasons for concern’, Integrated Assessment, vol. 6(1), pp. 19–56.
IPCC (2013). ‘Summary for policymakers’, in ( T.F. Stocker, D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen,
J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley, eds.), Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate
Change, Cambridge and New York: Cambridge University Press.
Keller, K., Bolker, B.M. and Bradford, D.F. (2004). ‘Uncertain climate thresholds and optimal economic
growth’, Journal of Environmental Economics and Management, vol. 48(1), pp. 723–41.
Kolstad, C.D. (1996). ‘Learning and stock effects in environmental regulation: the case of greenhouse gas
emissions’, Journal of Environmental Economics and Management, vol. 31(1), pp. 1–18.
Lenton, T.M., Held, H., Kriegler, E., Hall, J.W., Lucht, W., Rahmstorf, S. and Schellnhuber, H.J. (2008).
‘Tipping elements in the earth’s climate system’, Proceedings of the National Academy of Sciences, vol. 105(6),
pp. 1786–93.
Mankiw, N.G., Phelps, E.S. and Romer, P.M. (1995). ‘The growth of nations’, Brookings Papers on Economic
Activity, vol. 1995(1), pp. 275–326.
Meinshausen, M., Meinshausen, N., Hare, W., Raper, S.C., Frieler, K., Knutti, R., Frame, D.J. and Allen, M.R.
(2009). ‘Greenhouse-gas emission targets for limiting global warming to 2c’, Nature, vol. 458(7242), pp.
1158–62.
Mirrlees, J.A. and Stern, N.H. (1972). ‘Fairly good plans’, Journal of Economic Theory, vol. 4(2), pp. 268–88.
Moyer, E., Woolley, M., Glotter, M. and Weisbach, D.A. (forthcoming). ‘Climate impacts on economic growth
as drivers of uncertainty in the social cost of carbon’, Climatic Change.
Nordhaus, W.D. (1991). ‘To slow or not to slow: the economics of the greenhouse effect’, ECONOMIC JOURNAL,
vol. 101(407), pp. 920–37.
Nordhaus, W.D. (1992). ‘An optimal transition path for controlling greenhouse gases’, Science, vol. 258
(5086), pp. 1315–9.
Nordhaus, W.D. (1993a). ‘Optimal greenhouse-gas reductions and tax policy in the “dice” model’, American
Economic Review, vol. 83(2), pp. 313–7.
Nordhaus, W.D. (1993b). ‘Rolling the “dice”: an optimal transition path for controlling greenhouse gases’,
Resource and Energy Economics, vol. 15(1), pp. 27–50.
Nordhaus, W.D. (1994). Managing the Global Commons: The Economics of Climate Change, Cambridge, MA: MIT
Press.

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
602 THE ECONOMIC JOURNAL [MARCH
Nordhaus, W.D. (2007). ‘A review of the “Stern review on the economics of climate change”’, Journal of
Economic Literature, vol. 45(3), pp. 686–702.
Nordhaus, W.D. (2008). A Question of Balance: Weighing the Options on Global Warming Policies, New Haven, CT:
Yale University Press.
Nordhaus, W.D. and Boyer, J. (2000). Warming the World: Economic Models of Global Warming, MA: MIT Press.
Nordhaus, W.D. and Yang, Z. (1996). ‘A regional dynamic general-equilibrium model of alternative climate-
change strategies’. American Economic Review, vol. 86(4), pp. 741–65.
Pindyck, R.S. (2013). ‘Climate change policy: what do the models tell us?’, Journal of Economic Literature, vol. 51
(3), pp. 860–72.
Popp, D. (2004). ‘Entice: endogenous technological change in the dice model of global warming’, Journal of

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019


Environmental Economics and Management, vol. 48(1), pp. 742–68.
Romer, P.M. (1986). ‘Increasing returns and long-run growth’, Journal of Political Economy, vol. 94(5), pp.
1002–37.
Sen, A.K. (2009). The Idea of Justice, Cambridge, MA: Harvard University Press.
Solow, R.M. (1956). ‘A contribution to the theory of economic growth’, Quarterly Journal of Economics, vol. 70
(1), pp. 65–94.
Stern, N. (2007). The Economics of Climate Change: The Stern Review, Cambridge: Cambridge University Press.
Stern, N. (2008). ‘The economics of climate change’, American Economic Review: Papers and Proceedings, vol. 98
(2), pp. 1–37.
Stern, N. (2013). ‘The structure of economic modeling of the potential impacts of climate change: grafting
gross underestimation of risk onto already narrow science models’, Journal of Economic Literature, vol. 51
(3), pp. 838–59.
Stern, N. (2014a). ‘Ethics, equity and the economics of climate change. Paper 1: science and philosophy’,
Economics and Philosophy, vol. 30(3), pp. 397–444.
Stern, N. (2014b). ‘Ethics, equity and the economics of climate change. Paper 2: economics and politics’,
Economics and Philosophy, vol. 30(3), pp. 445–501.
Sterner, T. and Persson, U.M. (2008). ‘An even sterner review: introducing relative prices into the
discounting debate’, Review of Environmental Economics and Policy, vol. 2(1), pp. 61–76.
Weitzman, M. (2012). ‘GHG targets as insurance against catastrophic climate damages’, Journal of Public
Economic Theory, vol. 14(2), pp. 221–44.
Zachos, J.C., Dickens, G.R. and Zeebe, R.E. (2008). ‘An early cenozoic perspective on greenhouse warming
and carbon-cycle dynamics’, Nature, vol. 451(7176), pp. 279–83.

Appendix C. Nordhaus, W.D. (1991). ‘To slow or not to slow: the economics of the
greenhouse effect’, ECONOMIC JOURNAL, vol. 101(407), pp. 920–37.

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 603

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
604 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 605

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
606 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 607

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
608 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 609

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
610 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 611

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
612 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 613

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
614 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 615

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
616 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 617

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
618 THE ECONOMIC JOURNAL [MARCH

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
2015] GROWTH, CONVEX DAMAGES AND CLIMATE RISK 619

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.
620 THE ECONOMIC JOURNAL [ M A R C H 2015]

Downloaded from https://academic.oup.com/ej/article-abstract/125/583/574/5076996 by guest on 02 October 2019

© 2015 The Authors.


The Economic Journal published by John Wiley & Sons Ltd on behalf of Royal Economic Society.

You might also like