Innovation and Threshold Effects

in Technology Responses to
Climate Change

Dennis Anderson and Sarah Winne

December 2003

Tyndall Centre for Climate Change Research Working Paper 43

 Innovation and
Threshold Effects
in Technology
Responses to Climate
Change
Professor Dennis Anderson and Sarah Winne

Imperial College London Centre for Energy Policy and Technology (ICEPT)
4th Floor RSM Building
Prince Consort Road
South Kensington
London, SW7 2BP

Email: [Link]@[Link]
[Link]@[Link]

Tyndall Centre Working Paper 43

December, 2003

1
Contents

Abstract 3
I. Introduction 3
II. Non-Linearities and Price Substitution Effects 4
III. Sensitivity to Initial Conditions, Parameters and Exogenous Variables 6
a) Additional Sources of Non-Linearities 7
b) Emissions, Accumulations and the Marginal Costs of Abatement 10
IV. A More General Model 12
a) Energy Demand Equation 13
b) Investment in New Technology 15
c) Cost of Old Technology Relative to New 15
d) Relative Prices 17
e) Market Shares 17
f) Price of Energy 18
g) Demand Met by Non-Carbon Technologies 18
V. Results 19
a) Model Runs 19
VI. Threshold Effects, External Benefits and Option Values of Innovation 31
a) Externalities 31
b) Threshold Effects in Marginal Benefit and Cost Curves 33
c) Uncertainties and the Economic Value of Exploring Options 34
VII. The Costs of Non-Carbon Technologies Relative to Fossil Fuels 36
VIII. Conclusions and Next Steps 40
Acknowledgements 42
References 42
Appendix 1: Parameter Values for the Base Case 44

2
Summary

The paper develops a way of characterising technological change in energy systems in a form
suitable for economic analysis. The focus is on technological responses to climate change. A
large array of non-carbon options is emerging, though their costs are generally high relative to
those of fossil fuels. However, costs are also declining relatively with innovation, investment and
learning-by-doing. The process of substitution is also argued to be highly non-linear, involving
threshold effects.
One consequence of this is that policies may be capable of engendering changes in the industry
out of all proportion to the scale of the policy itself; and it is suggested that the external benefits
of the policies, when they take root, may be far larger than marginal analysis often assumes.
Much of course depends on the scale of the policy impetus—and equally on the durability of the
policy, for policy initiatives followed by reversals or a lack of willingness on the part of
policymakers to persevere, is no better than no policy at all. A number of implications for policy
and economic analysis are drawn.
1. Introduction
When a new technology or practice is being substituted for another, it is possible for highly non-
linear or threshold effects to arise when relative prices become comparable. This is the case for
example when a new technology or practice with low polluting characteristics is being developed
to displace an existing technology with high levels of pollution while providing the same product
or service, such as kWh of useful energy or passenger miles of transport. The evidence suggests
that (a) the difference in costs between low polluting technologies and the polluting ones they
displace are often relatively small, and (b) the extent of substitution arising from correspondingly
small shifts in prices, brought about for example by environmental policies to favour the low
polluting options, may rise to 100%.1 In these situations substitution elasticities may reach extra-
ordinary levels, and threshold or switching effects arise. Examples are the emergence of
combined cycle gas-fired power stations in the 1990s, the substitution unleaded for leaded fuels
in vehicles, and the rapid uptake of electrostatic precipitators to reduce PM emissions from coal
fired power plant in the 1950s and 1960s; but there are many others.
The possibility of threshold effects is compounded if, in addition, costs change with discovery,
with the innovations associated with ‘learning-by-doing’, and with scale economies. Even
technologies with similar learning curves may exhibit appreciably different rates of cost reduction
if they are at different stages of development. For example, a technology with a 20% reduction in
costs with each doubling of cumulative investment would experience a three-to four-fold drop in
costs if its market share rose from, say, 0.1% to 5. In contrast, mature technologies having the
same learning curve co-efficient but occupying say 30% of the market would experience only a
one-third decline even if they rose to occupy the whole of the market. In other words, the effects
of learning are most pronounced in the early phases of a technology’s development and use.
This paper examines these effects in a dynamic model of technology substitution in which the
costs of both the new technologies and the technologies in place decline in a non-linear way with
cumulative investment, and in which the price substitution effects are also highly non-linear. The
model has chaotic features similar to that of an ‘S-curve’ model encountered in the physical
sciences (May, 1976), with the difference that innovation, demand growth and policy shifts

1
There are numerous examples of this happening with respect to local pollution from industrial activities,
electricity generation and transport, and water pollution from industrial and municipal effluents. Anderson,
D (2001). “Technical Progress and Pollution Abatement: an economic view of selected technologies and
practices.” Environment and Development Economics 6: 283-311

3
introduce further degrees of non-linearity, and sometimes threshold effects. It is applied to the
analysis of technology responses to climate change.
Much of the analysis of such models is concerned with their chaotic behaviour. This is, however,
not the main concern of this paper, since as will be seen chaotic behaviour in economic systems
can be mitigated by reducing time lags between cause and effect, for example by more prompt
feedback of market information. Its main concern is a second characteristic of non-linear systems,
which is their extreme sensitivity to small changes in initial conditions and the exogenous
variables, such as tax and innovation policies. It is shown that, when a technology takes root in a
system in which switching or threshold effects are possible, relatively small investments in its
development may lead to levels of use in the long term out of all proportion to the initial scale of
effort (and the initial scale of the policies); correspondingly, the external benefits of the initial
effort, which take the form of the benefits of the technology developments themselves, plus the
benefits of pollution abatement (where they relate to low-polluting technologies) may also be
exceptionally large in relation to the initial costs—appreciably larger than marginal analysis often
suggests. By the same token, small events, such as might arise from equivocations or reversals in
policy, may lead to failure, whatever the economic promise of a technology. The willingness of
innovators and policymakers alike to persist with an approach until the evidence is clear one way
or the other seems crucial for successful technologies to emerge.
The next two Chapters introduce the approach using simplifying assumptions. The assumptions
are relaxed in Chapters 4 and 5 when a more general model is developed and applied to policies
for addressing climate change; it is shown that threshold effects or ‘disruptive change’ in
technology responses to climate change are not beyond the bounds of possibility, as indeed the
scenarios of industry, the IPCC and many others have concluded. Chapter 6 turns to the problem
of estimating external costs and benefits in such non-linear systems. Chapter 7 summarises
evidence on innovation and the expected costs of low carbon technologies; it is placed late in the
text since the preceding chapters pose new questions for cost analysis based on learning curve
that frequently escape close scrutiny when modelling responses to climate change; in particular,
estimates of the levels to which costs might decline in the long-term are neglected by learning
curve analysis, which implicitly assumes no lower bound to costs; the analysis in this paper
combines evidence from learning curves with engineering evidence on this possible (if uncertain)
lower bound. Chapter 8 presents the conclusions.

2. Non-Linearities and Price-Substitution Effects

Consider the way one technology might be substituted for another as relative prices change.
Schematically, we might expect the effects to be as in Figure 2.1. Let Pt denote the ratio of the
price of the old (polluting) technology to that of the new (low polluting) technology, inclusive of
the effects of taxes and subsidies on prices. We might expect (a) low rates of substitution when Pt
is low, (b) high rates when Pt is close to unity, tapering off to (c) low rates again when Pt is high
and the market for the new technology is saturated. In the figure, market shares are denoted by St
(0 ≤ St ≤ 1). The market shares relate to the demands for new investment not to shares in total
capital stock. The term P(dS/dP) is shown indicate the effects of a per unit change in prices on
the share of the new technology in investment.2

2
The actual elasticity of substitution is given by:
η = ( p / r )(∂r / ∂P ) , where r = X n / X o , the ratio of investment in new to new technology. For the
logistic form given in equation (2) below η = aP . For the value of a = 15 in Figure 1 it rises from 7.5
when P = 0.5 to 22.5 when P = 1.5.

4
Figure 2.1: Effects of Relative Prices on Market Share (St )

1.20 4

3.5
1.00

3
(St left scale)
0.80
2.5

0.60 2

Pt. dS/dP (right scale) 1.5

0.40

0.20
0.5

0.00 0
0 0.5 1 1.5 2 2.5
Pt

The relationship between St and Pt postulated here is the familiar logistic or ‘S’ curve long used
for the analysis of the demand for new durable goods, except that the rate of change in market
share is expressed with respect to the change relative prices rather than the change in time, which
seems to us to have more economic appeal. In continuous form:

dS / dP = aS (1 − S ) (1)
The solution to which is:
S (t ) = e a ( P ( t ) −1) /(1 + e a ( P ( t ) −1) ) (2)

Both S and P change over time. Rearranging and using discrete time steps (1) takes the form:

St = St −1 + aSt −1 (1 − St −1 )( Pt − Pt −1 ) (3)

For certain values of the parameter a equation (3) is chaotic. A more familiar form of such
equations, which have received extensive treatment in the literature3, is X t = αX t −1 (1 − X t −1 ) .
For values of α ≥ 3.57 the long run solution starts to become chaotic; in fact, for α ≥ 3.8284
“there are cycles for every integer period, as well as an unaccountable number of asymptotically
aperiodic trajectories.”4 Figure 2.2 illustrates the effect for the following variant of this
equation, S t = S t −1 + αS t −1 (1 − S t −1 ) , with α = a(Pt - Pt-1). In this case an unstable and chaotic
region begins for values of α above ≈ 2.3.

3
See R. M. May (1976) “Simple mathematical models with very complicated dynamics.” Nature, 261:151-
159.
4
Ibid. p 154.

5
Figure 2.2: Chaotic Behaviour of Equation 2, and its Stabilisation Using Monthly Series

1.40

1.20
B ased on m onthly intervals

1.00
St

0.80

0.60

0.40
B ased on yearly intervals

0.20

0.00
0 2 4 6 8 10 12 14 16 18 20
T im e (years)

The unstable behaviour of the equation has an economic interpretation. The assumption
underlying (3) is that the growth of markets in the current period relates to estimates of market
shares in prices in the previous period (since changes in prices affect the value of α). If the
demand for a technology is going through a rapid growth phase, and if the time interval between t
– 1 and t is long (a year say) then boom and bust cycles may appear if investors fail to spot that
the market is approaching saturation and assume a continuation of ‘buoyant’ growth—a not
uncommon phenomenon. Such effects can be neutralised by the continual feedback of more up-
to-date information, as can be seen when we use the same equation with the same parameters and
initial conditions but using monthly intervals (the dotted curve in Figure 2). Looked at another
way, we can effectively reduce the value of what is a parameter in the simpler logistic model,
namely α = a(Pt - Pt-1), by reducing the time interval and the change in prices.5

3. Sensitivity to Initial Conditions, Parameters and Exogenous Variables

This paper is less concerned with the problem of chaotic behaviour, however, except to alert us to
the complexities it might introduce when relationships such as (3) are used larger-scale models,
than with another familiar feature of highly non-linear relationships such as (1) and (3). This is
the extra-ordinary sensitivity of the trajectories to the initial conditions, parameters and the
exogenous variables (such as policy variables). The effects may be compounded in such models if
relative prices Pt also drop steeply with the innovations induced by the rise in market share, e.g.
with further invention and learning-by-doing.

5
Alternative forms of stabilisation are also possible, for example, short-run supply shortages may arise
when markets are growing rapidly and lead to short run increases in the prices of the new technology;
analysis of these is beyond the scope of this paper.

6
3.1 Additional Sources of Non-Linearities: Learning Curves and Scale Economies

It is necessary to look at the investment and cost equations underlying Pt and its initial values in
more detail. In studies of innovation, a relationship that has often been fitted to cost data is the
‘learning curve’.6 This has the form
Ct = C0 ( X t / X 0 ) − b (4)

where Ct are the unit costs at time t, C0 initial costs, Xt the cumulative investment (taken as an
indicator of experience) in the technology by time t from the time of its first introduction and b is
the ‘learning-curve parameter’. This relationship is highly non-linear, especially in the early
phases when Xt is small and experience accumulates rapidly. Figure 3.1 below shows the effect,
in which market share is taken as a proxy for Xt.7 Since prices are equal to costs plus or minus
any taxes or subsidy, it is evident that innovation introduces an extra order of non-linearity is
introduced in relationships like (3). Let us consider the problem this poses further.

Figure 3.1: Threshold Effects in Technology Development

(learning rates in percentages)

Initial Costs, when markets are small

5

4
Cost Ratio,
relative to
3
existing
10%
technologies
2

20%
1
30%

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Market Share (fraction of 1.0)

6
A. McDonald and L. Schrattenholzer (2001), “Learning Rates for Energy Technologies”, Energy Policy
29: 255-261, who give estimates for several other technologies and from other sources.
7
The term learning rate in Figure 3 refers to the per unit decline in costs with each doubling of cumulative
investment, which is equal to 1-2-b .

7
In finite difference form, the learning curve relationship assumes that the change in cost for each
per unit change in cumulative investment (∆X y / X t ) , is proportional to the difference between
the current costs and their lowest possible level, Cmin say, or:
∆Ct /( ∆X t / X t −1 ) = −b(Ct −1 − Cmin )
Rearranging: Ct = Ct −1 − b(Ct −1 − Cmin )( X t − X t −1 ) / X t −1 (5)

By using the form shown in (4), all studies so far have tacitly put Cmin to zero, and perhaps have
only escaped misfitting in the relationship because the values of Ct in the early phases of a
technology’s development, when statistical fitting is most often applied, tend to be high relative
to Cmin. Extrapolations based on (4) may be reasonably accurate over short periods, and in the
early phases of a technology’s development, but there is a danger of large inaccuracies if applied
over longer periods.
To take the analysis further, let us substitute (5) into (3) on two assumptions, both of which will
be relaxed later. First, that the existing technology with which the new one is competing is on the
flat part of its learning curve such that changes in its price can be ignored and Pt = (1+Tt)/Ct ,
where Tt is a pollution tax on the existing technology; for convenience, the cost of the existing
technology is put at unity and the costs of the new technology are expressed relative to the
existing technology. The second assumption is that the size of the market is constant such that Xt
is proportional to the cumulative value of the (annual) market share, or:

X t ∝ ∑0 Sv
t −1
and X t − X t −1 ∝ St −1

Substituting for Xt in (5) and repeating (3) for convenience, we now have two coupled non-linear
dynamic equations, plus a third relating prices to costs and a time dependent policy variable;
market shares are now determined by prices, and prices by market shares, relative costs and tax
policies:
St = St −1 + aSt −1 (1 − St −1 )( Pt − Pt −1 ) (3)

St −1
Ct = Ct −1 − b(Ct −1 − Cmin ) (6)
∑ S
v < t −1 v

Pt = (1 + Tt ) / Ct (7)

where the denominator in (6) represents the cumulative experience up to t = 0.

In these relationships there are two parameters in question (a and b), two initial conditions (C0
and ∑ S )8, a vector of exogenous policy variables (Tt , t = 0, 1, 2 ….), and an exogenously
v < t −1 v
estimated cost-expectation parameter (Cmin). None of these quantities are known accurately: the
range of a might plausibly be 5-15 (possibly wider, see below); estimated values for b for non-
carbon technologies ranges is 0.06 for nuclear power, 0.08-0.32 for wind energy, and 0.18-0.32
for solar photovoltaics; there are appreciable disagreements in the engineering literature as to
what the long term costs (Cmin) might decline to (see Chapter 7); even the initial conditions for
costs and taxes are difficult to estimate reliably since quoted costs vary greatly with bids while

8
There is also an initial condition for S0, which needs to be specified such that the difference equation
roughly satisfies the condition S t ≈ 0.5 when Pt = 1.0. In the following analysis, S0 is derived from
equation (2) using a value of P0 derived from C0 and T0 using (7).

8
governments have used a range of instruments—feed in laws, tradable permits, capital grants,
regulatory and regulatory targets—which make it difficult to ascertain with precision the imputed
value of T0. Yet as will be seen only small changes in estimates and assumptions may lead to
radically different results.
Figure 3.2 illustrates the point with two runs of the model. It considers cases where an incentives
policy is introduced sufficient to raise the ratio of the costs of fossil to non-fossil fuels by 50% (T
= 0.5), either by taxing the former, subsidising the latter of by a combination of the two. In one
case the policy is cut after 5 years, in the other after 20 years. In the former, the ‘coppicing’ of the
policy leads to the collapse of the non-carbon industry, from which it never recovers. In the
former, the industry is sufficiently well developed to survive the shock and to recover and grow.
It shows how important the willingness of policy makers to persevere with a policy will be; we
might term this the durability of a policy, which in this case is as important as the scale of the
policy itself.

Figure 3.2 Technology Choice and the Durability of Policies

1.0

0.8
St

0.6
Policy cut after 20 years

0.4

0.2
Policy cut after 5 years

0.0
1 11 21 31 41 51
Years
-0.2

The effects of alternative assumptions on such results will be explored once the fuller model has
been described below.9 Suffice it to note here that (i) higher values of the threshold effect
parameter a merely make the contrast between the two policies greater, in the sense that with
higher values of a switching would occur sooner and more rapidly in the case where the policies
take root. (ii) Lower values of the learning curve parameter, b, not surprisingly imply that
stronger policies would be needed for longer periods, as do lower values of a. (iii) A permanent
policy of taxing carbon would be unnecessary if the long-run costs of the non-fossil fuels (Cmin)
were to become significantly less than those of fossil fuels, and necessary if they were to be

9
The parameters and assumption for the runs shown in Figure 3.2 are as follows: a = 10.0, b = 0.3, C0 =
2.5, Cmin= 0.8, ∑ v<0
S v = 0.02 , about four times S0, which is calculated from the initial price ratio, Tt
= 0.5 for the first 5 years in one case and the first 20 in the other, after which it is zero.

9
higher. However, it is the initial policies themselves, and the willingness of policy makers and
industry to persevere with them, that are of over-riding importance.
Figure 3.3 shows the possible effects of perseverance (or the lack of it) in another way by plotting
the market share in year 50 (S50) against the year when the policy is cut. It illustrates how a
reversal of a policy might lead to its failure when it is on the brink of success.
Figure 3.3: Effects of perseverance on long term market shares

1.0
0.9
0.8
0.7
0.6
S50 0.5
0.4
0.3
0.2
0.1
0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29
Year when policy is cut

3.2 Emissions, Accumulations and the Marginal Costs of Abatement

Consider the case where a non-carbon energy source is displacing fossil fuels. The extra cost of
investments made in non-carbon sources in year t is (Ct − 1) St . If all investments were in fossil
fuels (St = 0) the costs would be unity (since the costs are being expressed relative to those of
fossil fuels). Hence the present worth of the total relative costs (PWTRC) over the period of
interest (t = 1…. Z say), as compared with the present worth of the costs of using fossil fuels if
investments there were no investments in non-carbon energy sources, is given by:

∑
Z
at (C t − 1) S t
≈ r ∑1 at (C t − 1) S t
Z
PWTRC = 1
(8)
∑
Z
1
at

where at is the discount factor and ∑a t t ≈ 1/r, where r is the discount rate.

Turning to emissions and accumulations, investments in fossil fuels will have emissions
proportional to their contribution to output, or to (1 – St) in the above notation. However, climate
change is a function of accumulations or concentrations in the atmosphere, so we need to add up
the emissions over its lifetime, from the year it is installed through to the end of the period of
interest (Z, say). For investments in fossil fuels introduced in t the cumulative emissions by Z will
thus be given by (1-St)(Z-t). Summing up over all vintages of investments, the contribution of

10
fossil fuels to carbon concentrations of investments made over the whole period is proportional
to:

∑ (1 − S )(Z − t ) = A
t =1
t z , say (10)

AZ is an index of the additions to carbon concentrations in the atmosphere over the period. The
index of additions in the ‘reference case’ when there is no investment in the non-carbon
alternative is given by putting St = 0, which gives a value of AZ = (1+Z)Z/2. Hence the per unit
abatement (PUA) of accumulations arising from the alternative technologies being introduced is
given by:

PUA = 1 − 2 AZ /(1 + Z ) Z (10)

To find an index of the marginal costs of abating (the cumulative volume of) emissions it is
necessary to estimate the change in PWTRC with respect to changes in AZ. This can be done by
varying the tax term T in (7). The results are shown in Figure 3.4, using the preceding equations
for three values of the cost parameter Cmin. The index of the marginal cost of abatement (MCA) =
100x∆(PWTRC)/∆ AZ.

Figure 3.4: An Index of the Marginal Costs of Abatement (MCA) versus the Cumulative Volume
of Emissions Abated on Different Assumptions for Cmin (hundred year period)

160.0

140.0

Cm in
120.0
= 1.5

100.0
MCA
(Inde x 80.0 Cm in
in %) = 1.0

60.0
Cm in
= 0.7
40.0

20.0

0.0
0 20 40 60 80 100
Pe rce nt a ba te m e nt

The marginal costs of abatement rise steeply as the level of abatement of cumulative emissions
reaches high levels, as is commonly discussed in economic texts. However, at low levels of

11
abatement, marginal costs are also high, since the technologies to reduce emissions are initially
far more expensive10-- as we are seeing with current efforts to reduce CO2 emissions in many
countries. It takes appreciable experience, and opportunities for innovation and scale economies
in the manufacture and installation of the required technologies, before marginal costs decline
significantly; expectations as to the costs and the performance of the technologies in the long-
term, and of course as to the incentives provided, are crucially important, as are alternative
assumptions about Cmin in the above figure show. Not the least of the dilemmas facing
policymakers and industry is that the values of Cmin are highly uncertain; in the case of climate
change, the range of possibilities is even greater than assumed above, both on account of
uncertainties as to the prices of fossil fuels (assumed constant so far) and the scope for
innovation. A further discussion of possible values is provided in Chapter 7.

It is not only the marginal cost curve which may have unfamiliar shapes in the presence of
threshold effects and non-linear cost functions; the marginal benefit curves may well have them
too. This topic is discussed in Chapter 6.

4. A More General Model

The following extensions are made to the above model:

(i) The size of the market is no longer constant, but increases with incomes, populations and
decreases in energy prices. The market in question is the world market, reflecting the global
nature of energy markets.
(ii) The world market is divided into two: the mature, low growth but still large scale markets
of the rich countries (OECD) and the high growth, populous markets of the developing
countries.
(iii) The retirement of old energy technologies as they come to the end of their economic
lifetimes provides further opportunities for new investment.
(iv) Reductions in the costs of the hydro-carbon technologies with which the non-carbon
technologies compete also decline with learning-by-doing.
(v) The costs per unit of capacity (e.g. per kW of electrical capacity) are similar across
countries on account of the influences of trade, investment and the international exchange
of knowledge and experience. The learning curve function for costs is therefore based on
world markets. However, the benefits of using energy technologies may differ between
countries on account of differing resource endowments; for instance the energy produced
by solar technologies in the tropics is three times that produced in the northern industrial
countries because the solar input is three times higher. To allow for such possibilities, a
yield parameter is introduced in the equations relating market shares to prices.
(vi) The learning curve equation is based on world markets, again reflecting the global nature of
the energy industry, and the rapid transmission of experience and technological information
across countries. Coupled with the differing cost parameters, this raises the possibility of
economically mutually beneficial externalities across countries arising from trade,
investment and international co-operation.
(vii) Relative prices are derived from costs and taxes as before.

10
In the above case, the initial costs of the non-carbon technologies are assumed to be 2.5 times those of
fossil fuels. Other than the different values of Cmin, shown in Figure 6, all other parameters and conditions
are the same as those assumed for Figures 4 and 5.

12
(viii) The prices of fossil fuels are assumed to be constant for the time being.

The equations remain deterministic, which is of course a limitation; the stochastic case, in which
the equations have error terms and the parameters are uncertain will be considered in a future
paper.

4.1 Energy Demand Equations

We have used a form commonly found in the economic literature:11

α β
Dt  Yt   Pet   Nt  −γt
=     e (4.1a)
D0  Y0   Pe0   N0 

αD β
DtD  Yt D   PetD   N tD  −γt
=   D   D e (4.1b)
D0D  Y0D   Pe
 0


N
 0



Yt: per capita income, in purchasing power parity prices

Nt: population
Pet: price of energy
α t : Per capita income elasticity
β : Energy price elasticity ( β =-0.5 for industrial countries, β =-0.3 for developing)
γ : Rate of technical progress in energy efficiency ( γ = 0.005)
Equation 4.1a represents energy demand in the industrial countries, 4.1b demand in the
developing countries (for which the superscript D is used).
For commercial energy income elasticities are zero below a certain threshold or ‘poverty line’,
rise to around 1.5 as countries become more industrialized and motorized and household demand
for energy using appliances increases, and then decline to one-fifth less of this at high per capita
incomes as markets mature.12 To reflect these changes we have used α t = 0.88e −0.00004Yt , based
on the work of Judson et al. The initial value of alpha for industrial countries is 0.30
corresponding to an initial income of $25,000, and for developing countries it is 0.78
corresponding to an initial income of $3,000.
There is a problem with the use of the so-called energy efficiency parameter γ : it has to our
knowledge not been estimated econometrically, since economists have widely ignored the energy
efficiency term in econometric work, which must necessarily lead to biases in the estimates of
price and per capita income elasticities; available estimates of γ come from engineering studies,
which as Grubb et al. (1992) noted range from under 0.005 to over 0.015. We plan to explore the
effects of such uncertainties on equation structure and parameters later.13
The per capita incomes for the industrial and developing countries (Yt and Yt D respectively) are
determined exogenously through a simple growth relationship:

11
E.g. Bates and Moore (1992) and Dargay and Gately (1995)
12
Bates and Moore (1992), Dargay and Gately (1995) and Judson, et al. (1999).
13
See Papathanasiou and Anderson (2001) , which explore the effects using Monte Carlo analysis, using an
earlier version of the model presented below.

13
 Yt = Yt −1 + g t Yt −1 (Industrial Countries) (4.2a)
Yt = Y
D D
t −1 +g Yt
D D
t −1 (Developing Countries) (4.2b)
Where:
g t = g t −1 (1 − v)
Y  (4.3a and 4.3b)
g tD = g t [1 + 0.32 t −D1 − 1]
 Yt −1 
and ν is a parameter representing the long-term rate of decline of economic growth in the
industrial countries. For countries such as the UK and the US, it has not declined at all in more
than a century, having averaged around 2% per year. The late starters in the industrial
revolution—such as France, Germany and, especially, Japan—experienced higher growth rates
over their long periods of ‘catch up’, but their long-term growth rates too converged to that of the
US and UK in recent decades, which has given a downward trend to the long-run weighted
average rate of per capita income growth in the industrial countries. Whether or not a 2% long-
term rate of growth can be sustained over the course of another century (it would imply a further
seven fold increase in per capita incomes and wealth) is not known. We have put the initial value
of ν = 0.01, implying that the long-run growth rate would decline to around 0.4 of its present level
over the century.
The equation for g tD allows for catch up of the developing countries, which is notable especially
in East Asia, and to a lesser extent in parts of South Asia and Latin America, though not alas in
most of Africa. The closer their per capita incomes approach those of the industrial countries, the
more their growth rates converge to those of the industrial countries.
Population levels are projected from the following, the assumptions for which can be estimated
from the projections of the World Bank and United Nations:
MN 0
Nt = (4.4a)
N 0 + ( M − N 0 )e − kt
D
D M D N0
Nt = D D D
(4.4b)
N 0 + ( M D − N 0 )e − k t

M is the stable population level and k a growth parameter. The populations are estimated
separately for the industrial and the developing countries. The initial level for the population of
industrial countries is 0.89 billion, whereas the initial level for developing countries is 5.1 billion.
The growth parameter k is given as 0.006 for industrial countries and 0.016 for developing
countries. The stable population level is 2.0 billion for industrial countries and 8.4 billion for
developing countries.14
The price variables for the demand equations will be presented later, once the equations for costs
have been derived.

14
World Bank. 2001. World Development Report 2000/2001: Attacking Poverty. New York: Oxford
University Press.

14
4.2 Investment in New Technology during Period t

Investment opportunities stem from demand growth and the retirement of old assets:
I t = ( Dt − Dt −1 ) + δDt −1 (4.5a)

I t = ( Dt − D D t −1 ) + δD D t −1
D D
(4.5b)

where δ = the retirement rate of assets that have come to the end of their economic lifetime. For
energy production investment typically has a 25-35 year lifetime ( δ ≈ 0.03). Investment is
calculated for each region using the regional demand equations (4.1).

4.3 Cost of Old Technology Relative to New: (a) using existing forms for learning curves

Consider first the case where the model is not separated into regions. From the equations
representing changes in demand, new investment and market shares (St, to be derived from the
equations on relative prices below), it is possible to estimate the cumulative investments in the
new and old technologies, denoted by U tN and U tO respectively:

U tN = U tN−1 + S t −1 I t −1 (4.6a)
U t
O
=U O
t −1 + (1 − S t −1 ) I t −1 (4.6b)

U 0N and U 0O are initial conditions, to be estimated separately from historical records.15 Costs can
then be derived from learning curve formula discussed in the preceding section. (Note that since
they depend in period t on cumulative experience up to the end of the preceding period, U tN and
U tO are defined in 4.6a and 4.6b to include investments only up to t – 1.) Expressing the costs of
old relative to new technologies:
N
C O C (U N / U N ) b
C t = tN = 0 Ot O0 b o (4.7)
Ct (U t / U 0 )
Similar learning rates in the runs reported in the next section are used for both old and new
technologies: bN = 0.3 (for new technology), bO = 0.3 (for old technology), and C0 is the initial
relative cost at time t = 0.
When the model is separated into the two world regions, the values for U are calculated for each
region (industrial and developing) and then added to calculate a relative cost function based on
world markets. The justification is that with trade, international investment and the sharing of
experience and knowledge across countries, the learning-by-doing effect is more closely related
to world than to regional market experience:

15
For the old technologies, which dominate initial supplies, a reasonable estimate of cumulative investment
 1 
can be derived from U 0 = D0   , where g* is the historical growth rate (we have taken a value
O

 g * +δ 
of g* = 0.08

15
 N
C tO C 0 (Wt N / W0N ) b
Ct = N = o
(4.8)
Ct (Wt O / W0O ) b
where Wt N = U tN (industrial )+ U tN (developing ) (4.9a)
and Wt = U (industrial ) + U (developing )
O
t
O
t
O
(4.9b)

4.3 Cost of Old Technology Relative to New: (b) Alternative equation including a minimum value
for the cost of new technology

A shortcoming of such cost equations based on learning curves (though the form in 4.8 is widely
used) is that the projected costs can decline to very low levels. In fact, they are asymptotically
zero, which is implausible. They may offer good approximations in the early phases of a
technology’s development, or even over particular periods if refitted to mature technologies, such
as has been done for gas turbines; but a relationship which holds well over say the first 5% of a
technology’s ultimate market cannot be assumed to hold over say the entire following 95%. It is
more appropriate have them converge toward some non-zero value, which for many technologies
can be estimated separately from engineering studies. To do this, first re-express the relationship
for C tN in a finite difference form to make the underlying dynamics more apparent. Above, the
cost of new technology was given by
−b N
W N 
C tN = C 0N  t N 

 W0 
Differentiating with respect to Wt N :

dC tN N N

N
= C 0N (W0N ) b (−b N )(Wt N ) −b −1
dWt
N N
⇒ ∆C tN = ∆Wt N C 0N (W0N ) b (−b N )(Wt N ) − b −1

N N
Or C tN = (Wt N − Wt −N1 )C 0N (W0N ) b (−b N )(Wt N ) −b −1
+ C tN−1

Rearranging:

CtN = (Wt N − Wt −N1 )(CtN )(−b N )(Wt N ) −1 + (CtN−1 )

To incorporate a minimum value into the equation, replace the term C tN with C tN−1 − C min .
( C tN is lagged one period to facilitate solving the equations of the model.) This gives:

 − bN  N
C tN =  N ( )
(Wt − Wt −N1 ) C tN−1 − C min + C tN−1 (4.10a)
 Wt 

An equation for C tO , the cost of old technology, can be derived in the same way:
 − bO  O
C tO =  O ( )
(Wt − Wt O−1 ) C tO−1 − C min + C tO−1 .
 (4.10b)
 Wt 

16
As before, b N is set at 0.3 and b O is set at 0.3. Estimates of Cmin can be derived from
engineering studies (see Chapter 7).

4.4 Relative Prices

Let Pt represent the price ratio of energy derived from old to new technologies in the industrial
countries and Pt D the corresponding ratio for the developing countries. These ratios will be
determined by the ratio of costs, the environmental taxes placed on the old (polluting)
technologies, and the subsidies for innovation provided for new technologies. In addition, it is
necessary to allow for differing natural resource endowments on the useful energy yield of the
technologies. As noted earlier, the capital costs per unit of capacity of the technologies used for
energy extraction and conversion tend to converge across countries with the influence of trade
and foreign investment—for example, the capital costs of oil and gas exploration, of electricity
generation from combined-cycle gas turbine plant, or the costs per kW of peak capacity of a solar
or wind plant. But the useful energy delivered by the plant may differ appreciably between
regions on account of differences in resource endowments, especially in the case of renewable
energy—wind, tidal streams, geothermal and solar energy are all cases in point. The following
expression makes an allowance for this through the factors F and FD by adjusting the effective
price paid per unit of energy delivered by an energy resource. In the model runs reported below,
F has been put to unity for the industrial countries, with FD for the developing countries thus
being expressed relative to that of the industrial countries.
C tO 1 + Tt 1
Pt = ⋅ ⋅ (4.13a)
C tN 1 − Gt F

C tO 1 + Tt D 1
and Pt D = ⋅ ⋅ (4.13b)
C tN 1 − GtD F D

Tt and Tt D = Tax rates (e.g. carbon tax) on old technology at time t, as a per unit of costs for
industrial and developing countries respectively
D
Gt and G t = Subsidies for innovation
F and Ft D = Resource endowment factors.

4.5 Market Shares

The market shares equation is of the same form as that presented in the previous chapter. There is
a separate equation for industrial and developing countries, linked to the price variables for each
region. Again letting S t and S tD respectively denote market shares of low carbon technology in
investment for the industrial and developing countries:
St = St −1 + aSt −1 (1 − St −1 )( Pt − Pt −1 ) (4.12a)

and S tD−1 = S tD−1 + aS tD−1 (1 − S tD−1 )( Pt D − Pt −D1 ) (4.12b)

As discussed earlier, these equations can introduce chaotic behaviour in the variable St, depending
on the product of the parameter a and the change in prices. A significant price shock, for example
can lead to values of St fluctuating chaotically above and below unity and above and below zero.
The equations are not structurally subtle enough to capture how markets might respond to the

17
overshooting and undershooting of investment, e.g. through more regular feedback of information
or by introducing additional feedbacks to neutralise the effects of high rates of change of prices.
But the problem can be circumvented by introducing the condition:
S 0 ≤ St ≤ 1.0 (4.13)

This is the approach used below.

4.6 Price of Energy (weighted average for energy prices)

The energy prices for each region ( Pet and PetD for the industrial and developing countries
respectively) are taken to be a weighted average of the new and the old technologies. For the
industrial countries:

Wt O − B O N Wt
N
N
C 0O ( O
) (1 + Tt )(1 − ( Rt −1 / Dt −1 )) + C 0 ( N
) − B (1 − Gt )( Rt −1 / Dt −1 )
Pet W0 W0
= (4.14a)
Pe0 C 0 (1 − G0 )( R0 / D0 ) + C 0 (1 + T0 )(1 − ( R0 / D0 ))
N O

Wt O − B O N Wt
N
−BN RtD−1
C 0O ( ) (1 + T D
t )(1 − ( Rt −1 / Dt −1 )) + C 0 (
D D
) (1 − G D
t )( )
Pe D t W0O W0N DtD−1
= (4.14b)
Pe D 0 C 0N (1 − G D 0 )( R0D / D0D ) + C 0O (1 + T D 0 )(1 − ( R0D / D0D ))

These equations enable us to close the loop and solve the energy demand equations 4.1 (for
industrial and developing countries) above.

4.7 Demand Met by Non-Carbon Technologies

Another useful term is Rt, the demand met by non-carbon technologies. This term is defined for
industrial and developing countries as:

Rt = Rt −1 (1 − δ ) + ( St I t −1 ) (4.15a)

RtD = RtD−1 (1 − δ ) + ( StD I tD−1 ) (4.15b)

Comparing Rt to Dt at any point in time allows us to determine how much of the demand for
energy is met by non-carbon technologies and how that demand is influenced by changing
policies and initial conditions.

18
5. Results
5.1 Model Runs
This chapter summarises the nine runs conducted. Table 1 shows the assumptions made for each
of the runs; the blank spaces for run 8 indicate that the level and duration of the carbon tax on
fossil fuels are being varied in the model. Unless specified, it is assumed that the other parameter
values are the same as in the base run, and that the parameter values are the same for both
developing and industrial countries. A full list of all the initial conditions and parameter values
for the base case is given in Appendix 1.

Table 1: Assumptions for the Nine Runs

Policy (incentive &

duration)
Parameter changes from
Run
T Yrs G Yrs base case
1. Base case (all regions) 0.3 100 0.3 20
2. No policy case (all regions) 0.0 0 0.0 0
3. Base case with policy reversal: a 0.3 10 0.3 10
4. No policy case + techn’gy surprise or a
0.0 0 0.0 0 Cmin-new = 0.75 Cmin-old
permanent fossil fuel price shock
5. Sensitivity to the minimum cost of new Cmin(new)=1.25
0.3 100 0.3 20
technology Cmin(old)
6. Sensitivity to initial conditions 0.3 100 0.3 20 C(0) = 0.3
7. Parameter space runs 0.3 100 0.3 20 0.1≤b≤0.4; 3≤a≤15
5≤Policy Duration≤50
8. Policy space runs 0.3 20
years; 0.1≤T≤0.6
9. No policies in developing countries
0.3 100 0.3 20 Industrial countries only

Run 1 (the base case)

The following figures show the model results when the initial cost ratio of fossil fuels relative to
the non-carbon alternative is 0.5. An innovation incentive of 30% reduces the prices of the non-
carbon alternatives for the first twenty years of the model run; a carbon tax increases the price of
fossil fuels by 30% for the entire 100 year period. Because the minimum cost of fossil fuels and
the minimum cost of non-carbon technology are both set at 0.8, the cost ratio decreases over the
100 year period to 1.0.
The removal of the innovation subsidy after 20 years is disruptive, but not fatal, to growth, since
the non-carbon technology has taken root by then, capable of surviving the shock.
The growth of demand for non-carbon technologies corresponds roughly with the mid range of
the IPCC scenarios.

19
Figure 5.1: Market Shares of the New (Non-Carbon) Technology

1
0.9
0.8
0.7
0.6
0.5
0.4
Market Shares of New Technology
0.3
0.2
0.1
0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.2: Relative Price of Fossil Fuels to Non Carbon Energy Sources

1.6

1.4

1.2

1
Relative Price
P(t)

0.8
0.6

0.4

0.2

0
1 9 17 25 33 41 49 57 65 73 81 89 97
Ye ar

20
Figure 5.3: Energy Demand Index (Industrial Countries)

1.4

1.2

1
Energy Demand (index)
0.8

0.6

0.4
Demand met by non-
0.2 carbon technologies

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.4: Energy Demand Index (Developing Countries)

2.5

2
Energy Demand (index)

1.5

1
Demand met by non-
carbon technologies
0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

21
Run 2 (the no-policy run)
There are no taxes or incentives implemented in this run. That is, T and G are set equal to zero for
the entire 100 year period. The non-carbon alternatives emerge endogenously, albeit very slowly,
driven by the steady effects of learning by doing and building on the initially small markets for
non-carbon technologies that exist today. As can be inferred from later runs (e.g. run 9) it is the
higher growth markets of developing countries that provide the main spur for continued
innovation when supportive policies are absent.

Figure 5.5: Energy Demand Index (Industrial Countries)

1.2

0.8 Energy Demand (index)

0.6

0.4 Demand met by non-

carbon technologies
0.2

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.6: Energy Demand Index (Developing Countries)

2.5

1.5 Energy Demand (index)

1
Demand met by non-
carbon technologies
0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

22
Run 3 (policy cut after 10 years)
This run shows the results if there is a 30% policy in place for 10 years only, the alternatives do
not emerge at all. Once the policy is taken away after 10 years, the drop in the relative price is
too sharp for the non-carbon market to recover. Comparing these results to Run 1 shows that
‘coppicing’ a policy too soon can be fatal, whatever the promise of the new technology (see also
the results reported in Figure 3.3 with the simplified model).

Figure 5.7: Energy Demand Index (Industrial Countries)

1.4

1.2

1
Energy Demand (index)
0.8

0.6

0.4 Demand met by non-

carbon technologies
0.2

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.8: Energy Demand Index (Developing Countries)

2.5

2
Energy Demand (index)
1.5

1 Demand met by non-

carbon technologies
0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

23
Run 4 (no policy + technology surprise, or alternatively a permanent shock to fossil fuel prices)
In this case, the possibilities of lower long-run costs for the non-carbon technology, relative to
those are explored. Alternatively, since Cmin is the costs of the non-carbon relative to those of
fossil fuel technologies, it can be considered as the case of a permanent rise in the latter. This run
shows the sensitivity of the model to the relative value of long run costs – even though there is no
policy, the non-carbon alternatives eventually emerge.

Figure 5.9: Energy Demand Index (Industrial Countries)

1.4

1.2

0.8 Energy Demand (index)

0.6

0.4
Demand met by non-
0.2 carbon technologies

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.10: Energy Demand Index (Developing Countries)

2.5

2
Energy Demand (index)
1.5

0.5 Demand met by non-

carbon technologies

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

24
Run 5 (sensitivity to minimum cost ratio)
To further examine the sensitivity of the model to the minimum cost ratio, this run specifies
Cmin(new) = 1.25Cmin(old). Even with a 30% carbon tax and a 30% innovation incentive for new
technology, the alternatives never emerge.

Figure 5.11: Energy Demand Index (Industrial Countries)

1.2

0.8 Energy Demand (index)

0.6

0.4
Demand met by non-
carbon technologies
0.2

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.12: Energy Demand Index (Developing Countries)

2.5

1.5 Energy Demand (index)

1
Demand met by non-
carbon technologies
0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

25
Run 6 (sensitivity to initial cost ratio)
In this run, the initial cost of new technology is three times that of the initial cost of fossil fuels,
as compared with an initial ratio of two in the base run. As a result, the new technology does not
emerge as quickly because it is more difficult to overcome this greater initial difference in cost.

Figure 5.13: Energy Demand Index (Industrial Countries)

1.2

Energy Demand (index)

0.8

0.6
Demand met by non-
carbon technologies
0.4

0.2

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.14: Energy Demand Index (Developing Countries)

2.5

Energy Demand (index)

1.5

1 Demand met by non-

carbon technologies

0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

26
Run 7 (Parameter Space Runs)
As discussed earlier, the model results are very sensitive to certain parameter values, most
notably the substitution parameter, a, and the learning rate for new technology, b. This run
focuses on the effect of varying both the learning rate and the substitution parameter on
R(50)/D(50), which is the percentage of the demand for energy that is met by non-carbon
technologies at year 2050. Low rates of learning in combination with low values of substitution
parameter a, yield a lower percentage of energy demand being met by non-carbon technologies.
The parameters themselves have threshold values for any particular set of policy assumptions.
However, as can be inferred from the next run on the policy space, the ‘cliff’ shown in Figure
5.15 can be shifted toward the origin by stronger and more durable policies; the general directions
of a policy are thus unaffected by the admittedly large uncertainties as to the parameters a and b.
A weaker substitution effect (lower a) or a lower learning rate (lower b) simply argue for a more
determined effort. Policies themselves need to include for a margin for error.

Figure 5.15: Effects of Parameters a and b on Long-Run Energy Options: Industrial Countries

0.7
0.6
0.5
0.4

R(50)/D(50) 0.3

0.2
0.1 20
0.0 15
5

0.2
0 2
10 a
0.3 5
b 0.4
0.5

27
Figure 5.16: Effects of Parameters a and b on Long-Run Energy Options
Developing Countries

0.8
0.7
0.6
0.5
0 5
0.4
R(50)/D(50) 0.3
0.2
0.1 20
0.0 15
5

0.2
0 2
10 a
0.3 5
b 0.4
0.5

Run 8 (Policy Space Run)

The preceding runs have already shown that if a policy is too weak, or is not implemented for a
sufficiently long time period, then, barring an economic surprise or a permanent shock to fossil
fuel prices, new technologies are unlikely take root. In this run, both the strength of the tax and
the duration of the policy are varied in order to show the extent to which policy determines
whether new technology catches on in the market. The duration (that is, the number of years both
the incentive and the tax are in place before being removed) varies from between 5 and 50 years.
The strength of the tax varies from between 10% and 60%. Figure 5.17 and 5.18 show estimates
of R(50)/D(50) over this policy space. The threshold effects of policy are again apparent.
Whatever the long term promise of the alternatives might be, they do not emerge when the
policies are too weak and/or if they are reversed too soon. The general message is that weakness,
equivocation and dithering on the policy front is no better—and is perhaps worse—than no policy
at all.

28
Figure 5.17: Effect of Policy Strength and Duration on Long-run Energy Options
Industrial Countries

0.9
0.8
0.7
0.6
0.5

R(50)/D(50) 0.4
0.3
0.2
0.1 50
40
0.0 30
0.1 10
20 Duration
0.2
0 2 0.3
0 3 0.4 0
trength
Str 0.5 0.6

Figure 5.18: Effect of Policy Strength and Duration on Long-run Energy Options
Developing Countries

0.9
0.8
0.7
0.6
0.5

R(50)/D(50) 0.4
0.3
0.2
0.1 50
40
0.0 30
0.1 10
20 Duration
0.2
0 2 0.3
0 3 0.4 0
trength
Str 0.5 0.6

29
Run 9 (Separate Policies for Industrial and Developing Countries)
This run simulates the effect of implementing a 30% carbon tax and a 30% incentive for new
technology in just industrial countries, while developing countries are not subject to any policy.
In this situation, non-carbon technologies are not able to catch on even in industrial countries
because of the absence of policies in developing countries, which undercut investment
opportunities and the rate of innovation. There is some growth for the non-carbon technologies in
the developing countries, but it is generally low (the result is similar to that in the second run).
The results serve to show how important the participation of developing countries in climate
change policies will be—and especially the importance of market growth in these regions as a
stimulus for innovation.

Figure 5.19: Energy Demand Index (Industrial Countries)

1.2

0.8

0.6 Energy Demand (index)

0.4
Demand met by non-
0.2 carbon technologies

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

Figure 5.20: Energy Demand Index (Developing Countries)

2.5

1.5
Energy Demand (index)

1
Demand met by non-
carbon technologies
0.5

0
1 8 15 22 29 36 43 50 57 64 71 78 85 92 99
Ye ar

30
6. Threshold Effects and the External Benefits and Option Value of Innovation.
This section discusses some implications arising from the preceding analysis for the cost-benefit
analysis of pollution abatement policies. The three main questions it addresses are:
• How do non-linearities and threshold effects in technology development and substitution
affect estimates of the external benefits of innovation?
• What are the implications for the cost-benefit analysis of pollution abatement policies? (The
discussion also takes account of threshold effects in environmental damage functions.)
• How might estimates of option values of alternative policies be affected?
The intention is to lay the analytical basis for further computations using the preceding model,
which will be reported in a future paper.

6.1 Externalities
Let pt denote the price of the energy produced, kt the marginal private costs of production, and et
the present value of the marginal environmental benefits arising from an investment, and Bt total
social benefits. It is total investment. If an investment has no effects on the costs of others, then
the marginal social benefits of the investment are simply the marginal private benefits plus the
marginal environmental benefits:
dBt / dI t = ( pt − k t ) + et
However, this is to assume that investments today have no influence on the costs and benefits of
future investments. When future costs change with investment actual social marginal benefits of
an investment I0 in period 0 consists of five terms, not two:
(i) The marginal private benefits as above ( pt − kt )

(ii) Plus the marginal environmental benefit arising directly from the investment (e0), also as
above.

(iii) Plus the present value of the marginal reductions in future costs of investments, weighted
by the prospective volume of use.

(iv) Plus the present value of the increase in marginal private benefit of future increases in
investments induced by the reductions in future costs.

(v) Plus the present value of the marginal environmental benefits of future investments also
induced by the reductions in future costs.

The third term is sometimes known as the ‘learning externalities’ term; it is the cost savings
arising from the reductions in future costs brought about by the current investment.

More formally, let B0 represent the net present value of a time stream of investments with
environmental benefits, such that:
B0 = a 0 ( p 0 − k 0 + e0 ) I 0 + a1 ( p1 − k1 + e1 ) I 1 + ....... (6.1)

31
Consider the case where a change ∆I0 leads to changes in both the unit costs of future investments
and, on account of this, to changes ∆I1, ∆I2, .... Then:
dkt dI
∆kt = ⋅ ∆I 0 , which is negative, and ∆I t = t ⋅ ∆k t
dI 0 dk t
Then, with at being the discount factor:
∆B0
∆I 0
{
≈ a 0 ( p 0 − k 0 + e0 ) + ∑t ≥1 at ∆k t I t + ( pt − {k t − ∆k t } + et )∆I t } (6.2)

The term in the summation sign represents the external benefits of innovation, and corresponds to
(iii), (iv) and (v) above.16 With threshold effects this component of external benefits induced by
investments ∆I0 may be very large—far larger than the term e0∆I0, which dominates (almost to a
fault) traditional economic thinking on environmental policy.

Figure 6.1 Graphical illustration of the terms in equation 6.2

kt
MB,
MC
∆kt

pt + et

pt

0 ∆It Qt Investment in energy

It supply in year t

16
That the term kt -∆ kt is an average value over the interval It to It + ∆It is explained below in relation to
Figure 6.1.

32
 Figure 6.1 illustrates the result graphically. The downward sloping line represents the demand
curve for new investment, which is taken to be the marginal benefit curve. The energy price
before environmental taxes is pt (the pre-tax price of fossil fuels say) and rises to pt + et if
environmental taxes are set equal to the marginal costs of environmental damage. Investment in
energy supply in year t is Qt of which the portion supplied by the low carbon technology would
be It (if costs remained at kt) the balance Qt - It being provided by fossil fuels. The marginal cost
of supply from the non-carbon alternative is kt. If this falls by ∆kt on account of innovation and
the learning effects of earlier investments, the cost savings to investment in period t, all else
constant, is approximately equal to I t ∆kt , the first term in the summation sign in equation 6.2, or
item (iii) in the paragraph which preceded it. If however the elasticity of substitution between the
alternative technologies is large, as is implied in the figure, small shifts in relative costs will lead
to large substitution effects, leading to large environmental benefits, approximately equal to
et ⋅ ∆I t , and to an overall economic benefit in year t of ( pt − {k t − ∆k t } + et )∆I t , where
{k t − ∆k t } denotes the average value of this quantity over the total supply; this corresponds to
the remaining terms within the summation sign of 6.2 or items (iv) and (v).
The quantitative importance of the externalities term within the summation sign on the right hand
side of 6.2 is commonly neglected in economic studies. Yet there is evidence from engineering
studies of pollution abatement technologies to suggest that for a number of environmental
problems in the past the elasticities of substitution are indeed large once the technologies have
been developed, such that it is the term just noted that has been the dominant effect in practice. 17

6.2 Threshold Effects in Marginal Benefit and Cost Curves

Threshold effects are commonly associated in the scientific literature with environmental change,
in which large and irreversible damages are thought to occur with low levels of abatement, and
are only avoided by high levels of abatement. Pollution standards have rarely called for, say 20-
30 percent reductions, or even 50-70%, but for reductions by orders of magnitude, and in a large
number of historical cases, e.g. in water and atmospheric pollution, this is what has been achieved
in practice.18 In its recommendations on how the UK might show leadership by responding to the
problem of climate change, the Royal Commission on Environmental Pollution (2000) argued for
60% reductions in CO2 emissions by 2050; but it is widely conceded that there is a non-trivial
possibility of threshold effects occurring, e.g. in the thermohaline circulation and glacial climates,
in which case very high levels of abatement might be needed.19
When environmental threshold effects arise, the total damage function may take the form of a
sharp S-curve of the form discussed earlier (see Figure 2.1), and the total benefits of abating a
reversed S-curve. The marginal benefit curve of pollution abatement is then not the steadily
declining function with diminishing returns, so often depicted in economic texts and papers, but a
peaky function of the form shown in Figure 6.2. Included is a schematic representation of the
marginal costs of abatement, discussed in Chapter 3. The total costs of abatement are the area
under the marginal cost curve, while the total costs of environmental damage equal the area under
the marginal benefit curve plus the irreversible fixed costs of environmental damage at low levels
of abatement.

17
A review and analysis is provided in D. Anderson (2001) “Technical Progress and Pollution Abatement:
an economic view of selected technologies and practices.” Journal of Environment and Development
Economics. 6: 283-311
18
Anderson (2001).
19
Mastrandrea and Schneider (2001), Ganopolski and Rahmstorf (1999, 2001).

33
Figure 6.2: Marginal Benefits and Costs of Abatement with Threshold Effects

MC
& MB
MB

MC

Abatement A

It can be seen why the debate on climate change (and on a good many other environmental
policies) is often so polarised. If the view is that irreversible damages are unlikely or are small,
then the optimum policy is for no abatement, beyond that which can be achieved by the so-called
‘no regrets’ or ‘win-win’ options such as are often thought possible through energy efficiency.20
If on the other hand the view is that the irreversible damages would be large in the absence of
abatement, then the optimum level of abatement would be at the other extreme, point A in the
above figure. In this case is it desirable to pass through a (perhaps long) phase of policy in which
the marginal costs exceed the marginal benefits of abatement, a possibility that is often ignored.

6.3 Uncertainties and the Economic Value of Exploring Options

The complexities deepen when uncertainties are taken into account. There are two major
uncertainties, concerning the costs of environmental damage, and the costs of avoiding it. It is
well known that:21
(a) There is a value, termed the option value, to delaying a decision on an irreversible investment
until the uncertainties are reduced. This may (depending on probabilities and the estimated cost
and benefit streams) argue for delay;
(b) When, on the other hand, a delay may lead to irreversible damages, the case for avoiding
delay and bringing forward a policy is increased.

20
A discussion of this topic requires separate analysis, and is beyond the scope of this paper.
21
AK Dixit and RS Pindyck (1994). Investment Under Uncertainty. Princeton NJ: Princeton University
Press.

34
(c) If the costs of abating damage are such that they cannot be ascertained reliably without
investment, then this too will argue for avoiding delay and for investments aimed at exploring
options.
These results are summarised for a simplified case in Table 6.1 in order to arrive at an elementary
point for policy. Let q denote the probability of one estimate of the annual damages arising from
climate change and (1-q) the probability of another estimate. The net annual benefits of avoiding
these damages are denoted by B1 and B2 respectively, and to simplify analysis they are assumed
constant over time. The costs of the investments are annualised and deducted from the net benefit
streams. B1 is taken to be > 0 and B2, which if we had more information we would avoid
incurring altogether, is <0 in the first four cases.
In case 1, there is no delay in policy; in case 2, the decision is delayed until better information is
available such that the possible economic losses (B2) are avoided. A comparison of the benefits
between the two cases suggests that it would always pay to delay in this situation. The situation is
different when irreversible losses arise, as can be seen from Case 3; investing now (Case 1) might
be the better option, depending on the balance of probabilities and net benefits.
Table 6.1.
Net benefits per period:
Probability 0 1 2 3,……. Expected net benefitsa/
Case 1—Invest now, with downside benefits B2 < 0:
q B1 B1 B1 B1,…. qB1 / r + (1 − q ) B2 / r
1-q B2 B2 B2 B2,….
Case 2—Delay one period (with no irreversible Losses B2 are avoided:
damages):
q 0 B1 B1 B1,…. qB1 / r (1 + r ) ≈ qB1 / r
1-q 0 0 0 0,….
Case 3—Delay one period (with irreversible damages): Losses B2 also avoided:
q - B1 B1 B1 B1,….
1-q 0 0 0 0,….
qB1 / r (1 + r ) − qB1 ≈ qB1 / r − qB1
Case 4—Invest I to explore options (irreversible
damages): qB1* / r (1 + r ) − qB1* − I
q -I - B1* B1* B1*,….
B1 *
1-q -I 0 0 0,….
Case 5—As for 4, but with net benefits B2* > 0 As in Case 4 plus:
q -I - B1* B1* B1*,…. (1 − q ) B2* / r (1 + r ) − (1 − q ) B2*
B1 *
1-q -I - B2* B2* B2*,….
B2 *
a/ The formulae assume long time horizons such that the present value of the benefit streams is
approximately 1/r times the annual benefit stream.
However, there is the further possibility of investing in the first period to explore options, the
point of which is to find novel and lower cost ways of addressing the problem; the effect is to
raise the net benefit stream from B1 to a higher level B1* and, by the same token, the lower
estimate of net benefits from B2 to a higher level B2*. This is summarised in case 4, in which
irreversible damage arises while options are being explored. Again, depending on probabilities
and the changes in the net benefit stream, this might be preferable to all other possibilities—and it

35
certainly would if, in addition, as illustrated in Case 5, it turned even the previously negative net
benefit stream B2 into a positive one.
It is apparent that virtually any policy position might be defended depending on assumptions as to
both the net benefit streams and the probabilities to be attached to them. Estimates of the net
benefits vary enormously in the case of climate change, from under $5 per ton of carbon emitted
to several hundred dollars per ton, and even these ignore the possibility of extreme threshold
events.22
However, it does seem that the policy of exploring options is likely remain robust under
reasonable assumptions, and may well be justified by reference to the benefits of innovation
alone. Let ∆C denote the present value of the annual cost savings benefits arising from the
exploration of options to address climate change. The net cost savings benefits of exploring
options would then lie somewhere in between q∆C − I and q∆C + (1 − q)∆C − I = ∆C − I .

7. The Costs of Non-Carbon Technologies Relative to Fossil Fuels

The cost equations require estimates of initial costs, the sizes of the market or market shares with
which these estimates are associated, learning rates, and an estimate of long-term costs (Cmin)
from engineering studies. For non-carbon alternatives to fossil fuels, all these quantities vary
appreciably between technologies.

Table 7.1 provides estimates of near and long term costs for a range of renewable energy
technologies, nuclear power, of hydrogen derived from these sources, and the costs of electricity
using fuel cells supplied by hydrogen.23 The table compares these costs with those of natural gas,
transport fuels and the generation of electricity from fossil fuels. The main sources of these
estimates are the reviews provided in the World Energy Assessment of the UNDP and the World
Energy Council (2000), the background studies and reviews commissioned by the PIU (now the
Prime Minister’s Strategy Unit) and the DTI for the UK Energy White Paper, and the review by
Anderson and Leach (forthcoming) on the use of hydrogen to solve the intermittency problem
posed by renewable energy. All the estimates shown are based on engineering economic analysis,
as opposed to the statistical analysis of costs more familiar to economists.

The last two columns of the table show the ratios of the costs of the non-carbon fuel relative to
the appropriate fossil fuel, which is taken to be petrol and diesel for transport fuels, natural gas
for electricity generation and the gas markets.24 Given the uncertainties in cost estimates—even
estimates of today’s costs vary greatly between studies—the ratios are indicative only. For the
purposes of the present study, five points stand out:
1. Current costs. These are generally higher, sometimes appreciably higher, than those of the
fossil fuel alternative. They vary from being about 50% higher in the case of wind, 70% higher
for nuclear power, to 200% higher for biofuels and offshore wind and marine energy, to 200-
700% higher for Photovoltaics (the lower figure being for high insolation areas, the lower one for
low insolation areas), to being an order of magnitude higher for hydrogen derived from renewable

22
A review is provided by Tol R.S. J. (1999), ‘The Marginal Costs of Greenhouse Gas Emissions’, The Energy
Journal, 20, 1, p. 61-81
23
The costs of hydrogen derived from fossil fuels with the carbon being sequestered are to be added in the
next draft.
24
These ratios are of course sensitive to assumptions about the costs of fossil fuels. It is planned to explore
these assumptions more fully in the next draft of the paper.

36
energy of nuclear power. It is clear that significant innovation, ‘learning’ and scale economies
will be needed if the costs of moving to a low carbon economy are not to be inordinately high.
2. Long-term costs. The engineering estimates shown in the table suggest that costs should
decline significantly over time. This assessment is supported independently by the statistical
analysis of learning-curves by McDonald and Schrattenholzer (2001), cited earlier, whose
estimates are summarized in Table 7.2.25 Note that the primary energy technology with the
highest initial cost (photovoltaics) is the one projected to have the lowest cost in the long-term.
3. The costs of using the renewable energy alternative vary greatly between countries. This is
most obviously the case for solar energy, but it is also true (though the estimates are not provided
in Table 7.1) for energy from wind and biomass and, by implication, hydrogen derived from these
resources. Indeed the costs of coal and gas vary appreciably between countries.
4. Uncertainties in the long-term costs of fossil fuels. The sensitivity of the cost ratios to
assumptions about the costs of fossil fuels is once again emphasized. Many studies assume that
the costs of fossil fuels will rise with growing demand and scarcity. The above estimates and the
model used in the present paper assume they will decline with resource discovery and technical
progress. The sensitivity of the results to these assumptions still needs to be explored.

25
Similar estimates of learning curve co-efficients are also to be found in the review by the IEA (2001).

37
 Table 7.1: The average costs of renewable energy compared with fossil fuels and nuclear
power: today and in prospect.
Projected Indicative ratio,
Technology costs as relative to relevant fossil
technology fuel comparator:a/
Current cost matures
(US c/kWh) Today Long-term
(US c/kWh)
Biomass Energy:
• Electricity 5-15 4-10
• Heat 1-5 1-5
• Ethanol for vehicle fuels 3-9 2-4 3.0 1.5
• (c.f. petrol and diesel) (1.5-2.2) (1.5-2.2)
Hydrogen fuelled vehicle To be added To be added
Wind Electricity
• onshore 3-5 2-3 1.5 1.0
• offshore 6 - 10 2-5 3.0 1.5
Solar Thermal Electricity (insolation of 2500kWh/m2/yr) 12-18 4-10 5.0 2.0
Geothermal Energy:
• Electricity 2-10 1-8
• Heat 0.5-5.0 0.5-5.0
Marine Energy:
• Tidal Barrage (e.g. the proposed Severn Barrage) 12 12
• Tidal Stream 8-15 4 3.0 1.3
• Wave 8-20 5-7 1.5 2.0
Grid connected photovoltaics, according to incident solar
energy (‘insolation’):
• 1000 kWh/m2 per year (e.g. UK) 50-80 ~8 8.0 1.0
• 1500kWh/m2 per year (e.g. southern Europe) 30-50 ~5 4.5 0.5
• 2500 kWh/m2 per year (tropics) 20-40 ~4 3.0 0.4

Nuclear Power 4-6 3-5 1.7 1.3

Electricity Grid supplies from fossil fuels (incl. transmission Declines in
and distribution): capital costs
• Off-peak 2-3 may be offset
• Peak 15-25 by rising fuel
• Average 8-10 costs
Costs of Central Grid Supplies, excl. transmission and
distribution:
• Generation from natural gas 2-4 Ditto
• Generation from coal 3-5
Hydrogen from Renewable Energy:
• Industrial consumers 9 6 18.0 12.0
• Residential consumers 12 8 4.0 3.0
Hydrogen from fossil fuels with carbon sequestration To be added To be added
• From coal
• From natural gas
Natural gas supplies:
• Industrial consumers 0.5 0.5
• Domestic consumers 3.0 3.0
Average costs of electricity with hydrogen as fuel.
a. Used for combustion in central generation:
• HV supplies 17 12 6.0 4.0
• LV supplies 25 14 3.0 1.5
b. With fuel cell for decentralized combnd ht & power:
• In homes 28 12 3.0 1.3
• In industry 23 9 8.0 3.0

38
Sources: World Energy Assessment: Energy and the Challenge of Sustainability. UNDP and World Energy
Council (2000) updated and extended based on data gathered for the UK government, PIU (2002), and for
the recent simulation studies undertaken for the UK, DTI (2003). The costs of hydrogen production and its
use for central and decentralized generation are based on the analysis of Anderson and Leach (forthcoming
in Energy Policy). The above estimates are based on a discount rate of 10%.
a/ The comparators are as follows: For ethanol: petrol and diesel fuels. For wind and marine energy:
combined cycle gas-fired generation. For grid connected PVs: the domestic price of electricity. For nuclear
power: combined cycle gas-fired generation. For hydrogen derived from renewable energy as a fuel: natural
gas. For the use of hydrogen (derived from renewable energy) as a combustion fuel for central generation,
and for use in fuel cells for decentralized generation: electricity from the grid. The estimates are rounded,
are based on the mid-points of ranges when ranges are available, and are intended to be indicative only.
---------------------------------
Table 7.2 Learning Rates for Selected Energy Technologies
Learning Rate,
Technology (and source of estimate) Period %
Wind:
• OECD 1981-95 17
• US 1985-94 32
• California 1980-94 18
• Denmark 1990-94 8
Solar PV:
• EU 1985-95 32
• World 1976-92 18
Ethanol (Brazil) 1979-95 20
Electrolytic Hydrogen from renewables (engineering studies) -- 18
Compact Florescent Lamps (US) 1992-98 16
Gas Turbine Combined Cycle Power Plants:
• OECD 1984-94 34
• EU n.g. 4
Gas Pipelines:
• Onshore 1984-97 4
• Offshore 1984-97 24
Oil Extraction from the North Sea n.g. 25
Coal for Electric Utilities 1948-69 25
Nuclear Power (OECD) 1975-93 6
Electric Power Production 1926-70 35
Source: Except for electrolytic hydrogen, which is based on Ogden’s review in the 1999 Annual Review of
Energy and the Environment, the estimates are quoted from A. McDonald and L. Schrattenholzer (2001),
“Learning Rates for Energy Technologies”, Energy Policy 29: 255-261, who give estimates for several
other technologies and from other sources.
n.g = not given.

5. The diversity of non-carbon options. This is perhaps the most important and robust conclusion
to emerge from a plethora of studies by industry, government, the IPCC and the academic
community over the past decade. It is that there is an abundance of options for achieving a low
carbon economy should the need arise. The list in table 7.2 is already long, but it is far from
exhaustive, and does not reveal an often large number of variants within each technological
grouping—in photo-conversion for example, in solar thermal technologies, and in devices for

39
harnessing the offshore resource. The sheer range of options that is being explored gives further
ground for the believing that innovation will reduce costs further, as indicated in the table. The
range and diversity of possibilities is such that it is almost impossible to “pick winners” from
among them.

This last point has an implication for how the process of technology substitution is characterized
in large scale economic modeling efforts. It is possible to draw up a model which represents all
conceivable alternatives and to use an optimizing algorithm for choosing between them; this is
broadly the approach of the IEA’s Markal model, which was used extensively in background
studies for the UK Energy White Paper.26 Given the range of costs and possibilities, it is
necessary when using such models to embark on a large number of sensitivity exercises.
However, it is impracticable to incorporate such models in the large scale macro-economic
modeling exercises, which serve the different purposes of estimating the economic impact of
substitution on economic growth and other economic quantities of interest. Clearly some
decentralization of responsibilities is necessary. The present authors suggest characterizing the
alternatives to fossil fuels under a single, non-carbon option for each economic sector—transport,
electricity generation, and the gas markets—without going into detail as to precisely which of the
non-carbon options is likely to emerge. Analysis of the latter can properly be delegated to a
separate study using such models as Markal.

8. Conclusions and Next Steps

The preceding analysis has sought to explore how threshold effects in substitution, combined with
learning and scale economies in technology development, may influence the way energy systems
might respond to the problems posed by climate change. It was shown that even seemingly small
shifts in the level and duration of policy assumptions may lead from a situation where energy
systems would continue to depend on fossil fuels for so long as fossil fuels remain abundant and
the least cost resource, to one in which, after allowing for the inevitably long lags associated with
the building up of a new capital stock, fossil fuels become almost wholly displaced by the non-
carbon alternative. It is not surprising therefore that the more than 150 scenarios of energy futures
in the peer-reviewed studies reviewed by the IPCC differ by orders of magnitude—from the
world’s economies continue to remain nearly 100 percent dependent on fossil fuels over very
long time periods, to becoming nearly wholly independent of them.
The difference, which in and of itself is hardly surprising, turns on policies, and of course the rate
of innovation. But the influence of policy may be far larger than marginal analysis might lead us
to believe. An initial impetus provided by policy is, under two conditions, capable of initiating a
dynamic process of discovery, investment, a rise in market shares, further investment, innovations
and reductions in costs through learning-by-doing and scale economies, further investment… and
so on. The two conditions are that:
• The policy signal is sufficiently strong, given the cost parameter.
• The policy is sufficiently durable.
Otherwise, the dynamic process may spiral inwards, no matter how promising the technological
alternatives might be, and lead to nothing. It was shown that a policy reversal in the early phases,
after, say a 5 year experiment, or a policy signal that is too weak or tentative, is no better than no
policy at all: a degree of perseverance is required for any innovation-oriented policy to succeed,
as it is for innovation itself. In every sense of the term, the future is what we make it. The only

26
Department of Trade and Industry (2003).

40
exception is when a technological surprise is in the offing; this cannot be ruled out, of course, but
to gamble on it would not seem to be a wise policy in the light of what we know of costs and
technology developments so far (see Chapter 7).
The implications of such effects are that both the positive externalities of innovation and the
environmental benefits of a policy are also far larger than marginal analysis might lead us to
believe. Instead of being marginal, if the effects of a policy take root, the overall social
benefits—the sum of the positive externalities of innovation, and the negative externalities of
pollution that are avoided—may be out of all proportion to the strength and duration of the initial
impetus.
In the presence of threshold and learning effects, the marginal cost curves for pollution
abatement take on a form quite different to the rising convex form so often depicted in economic
texts; similarly, given the possibility of threshold effects in the climate change damage function,
the curve depicting the marginal benefits of abatement takes on quite a different form. (See
chapter 6.) The upshot is that polar opposite conclusions may sometimes be drawn on the role of
policy and its effects, depending on one’s assessment of the scale of the threshold effects and
innovation: on the one hand the optimal level of abatement may be near-zero, on the other nearly
100%. Once again, however, the difference between these extremes may be a relatively small
step in policy, above all a willingness to explore options thoroughly.
The markets for energy technologies are global markets, and learning experiences are rapidly
transferred internationally both through investment and through the transmission of knowledge.
It follows that international co-operation through trade and investment, facilitated by
international organisations, conveys appreciable positive externalities across countries. It was
shown that the rate of technology development and uptake is likely to be far slower—and in
some cases may grind to a halt altogether—if policies fail to recognise the opportunities of
international co-operation in the area of innovation and technology development. So far, the
international policy dialogue on climate change has focussed on the Kyoto Accords, the Global
Environment Facility (the financing arm of the UN Framework Convention on Climate Change)
and the Clean Development Mechanism. A new and promising dimension would be to put more
onus on international co-operation in the area of innovation, as the UK Prime Minister himself
has recognised.27
The preceding analysis is also intended to provide a framework for thinking about technological
change in large-scale economic models—in particular the Cambridge model—for the analysis of
the economic implications of responding to climate change. In turn, this will feed into the
research on Integrated Assessment by the Tyndall Centre for Climate Change. A number of
analytic findings were reported on in the text, e.g. on how to avoid the chaotic features
associated with non-linear models (Chapters 2, 3 and 4) and on how to characterise non-carbon
options without getting into impossible tangles when handling what is, in fact, a very large
number and diverse array of non-carbon options (Chapter 7). There are already a number of ways
of studying the latter in depth, e.g. by using the IEA’s Markal model; but this entails sacrifices
when it comes to analysing the macro-economic impact. There is a natural division of labour
between the two approaches, and for the macro-economic analysis it should be sufficient to
concentrate on the substitution and costs of non-carbon options in general for each of the key
sectors—e.g. for transport, and the gas and electricity markets. Given the huge array non-carbon
options, and the variations of costs between countries, the actual task of analysing which ones
are likely to emerge, and what kinds of policy support are merited, is best left to country-level
analysis using such tools as Markal.

27
See ICEPT (2002): “Assessment of Technological Options to Address Climate Change: A Report for the
Prime Minister’s Strategy Unit.” The report is available on the websites of the Strategy Unit and ICEPT.

41
There is still much analytical work to be done to complete the analysis. (1) Equations to
represent carbon emissions and accumulations need to be introduced (a straightforward step). (2)
There is further work to be done on costs, as indicated in Chapter 7. (3) The arguments on
positive externalities in Chapter 6 need to be quantified. (4) How to handle uncertainties and risk
needs to be explored (a big step). Related to this, (5) we plan to look at the option value of
policies in light of uncertainties and the analysis in Chapter 6.
Acknowledgements
This research for this paper was made possible by a grant from the Tyndall Centre for Climate
Change. We offer our thanks to the Centre for the opportunities it has provided, and to all the
following colleagues named below.
The project is part of a larger involving the Cambridge Department of Applied Economics
(Jonathan Köhler, Terry Barker and Haoran Pan), the Policy Studies Institute (Paul Ekins and
Paulo Agnolucci), and the University of Manchester Business School (Paul Dewick, Marcela
Miozzo, and Ken Green). Rachel Warren of the Tyndall Centre has also participated actively in
all our meetings and seminars, since she is responsible for the development of Tyndall’s
Integrated Assessment Model for the analysis of climate change, and is not only tracking our
thinking closely, but has engineered a follow-up grant to enable us to compare our approach with
those of other international scholars. Our colleague Tim Foxon has both participated in all the
discussions of the development of our approach, and has contributed a lot to our thinking through
his work on innovation, which is currently being sponsored by the ESRC’s Sustainable
Technologies Programme.
References
Anderson, D. (2001). “Technical Progress and Pollution Abatement: an economic view of
selected technologies and practices.” Environment and Development Economics 6: 283-
311.
Bates, R.W. & Moore, E.A. (1992). “Commercial Energy and the Environment.” Policy
Research Working Paper. Washington, DC: The World Bank, Environmental Policy and
Research Division.
Dargay, J. & Gately, D. (1995). “The Response of World Energy and Oil Demand to Income
Growth and Changes in Oil Prices.” Annual Review of Energy and the Environment 20:
145-78.
DTI (2003). DTI Economics Paper Number 4: Options for a Low-Carbon Future.
London: Department of Trade and Industry.
Dixit, A.K. & Pindyck, R.S. (1994). Investment Under Uncertainty. Princeton NJ:
Princeton University Press.
Ganopolski, A. & Rahmstorf, S. (1999). “Long-term Global Warming Scenarios Computed with
an Efficient Coupled Climate Model.” Climatic Change 43: 353-367.
Ganopolski, A. & Rahmstorf, S. (2001). “Rapid Changes of Glacial Climate Simulated in
a Coupled Climate Model.” Nature 409: 153-158.
Grubb, M. et al. (1992). “Induced Technical Change in Energy and Environmental
Modelling: Analytic Approaches and Policy Implications.” Annual Review of Energy and
the Environment 27: 271-308.
Grubb, M. et al. (1993). “The Costs of Limiting Fossil-Fuel CO2 Emissions: A survey and
analysis.” Annual Review of Energy and the Environment 18: 397-478.

42
ICEPT (2002). “Assessment of Technological Options to Address Climate Change: A Report for
the Prime Minister’s Strategy Unit.” (Available on
[Link]
Judson, R.A., Schmalensee, R., & Stoker, T.M. (1999). “Economic Development and the
Structure of the Demand for Commercial Energy.” The Energy Journal 20(2): 29-
57.
Mastrandrea, M., & Schneider, S. (2001). “Integrated Assessment of Abrupt Climatic
Changes.” Climate Policy 1: 433-449.
May, R. (1976). “Simple mathematical models with very complicated dynamics.” Nature
261: 459-467.
McDonald, A. & Schrattenholzer, L. (2001). “Learning Rates for Energy Technologies.” Energy
Policy 29: 255-261.
Papathanasiou, D. & Anderson, D. (2001). “Uncertainties in Responding to Climate
Change: On the economic value of technology policies for reducing costs and creating
options.” The Energy Journal 22: 79-114.
PIU (February 2002). Energy Review: A Performance and Innovation Unit Report.
London: Performance and Innovation Unit, the Cabinet Office.
Royal Commission on Environmental Pollution (2000) Energy – The Changing Climate.
London: Royal Commission on Environmental Pollution.
Tol, R.S.J. (1999). “The Marginal Costs of Greenhouse Gas Emissions.” The Energy Journal
20(1): 61-81.
UNDP and World Energy Council (2000). World Energy Assessment: Energy and the Challenge
of Sustainability. New York, NY: United Nations Development Programme.
World Bank (2001). World Development Report 2000/2001: Attacking Poverty. Washington,
DC: World Bank.

43
Appendix 1: Parameter Values for the Base Case Run
Parameter Base Case Value Value for Developing Value for Industrial
Countries Countries*
a: substitution parameter 10
b old: learning rate for old
technology 0.3
b new: learning rate for new
technology 0.3
g*: historical growth rate 0.08
Gamma: energy efficiency
parameter 0.005
M: stable population value 8.4 billion 2.0 billion
beta: energy price elasticity
-0.3 -0.5
alpha(0): initial per capita
income elasticity 0.78 0.3
k: population growth rate 0.016 0.006
delta: retirement rate 0.03
v: rate of decrease in income
growth 0.01
T: Carbon Tax 30% for 100 years 30% for 100 years
G: Innovation Subsidy 30% for 20 years 30% for 20 years
Cmin (old): minimum cost
of old technology 0.8
Cmin (new): minimum cost
of new technology 0.8
Cnew(0): initial cost of new
technology 2.0
Cold(0): initial cost of old
technology 1.0
D(0): initial demand for
energy 160 exajoules 240 exajoules
Pop(0): initial population 4.95 billion 1.04 billion
Y(0): initial income $3410 ppp $24430 ppp
S(0): initial market shares in
new technology 0.025
g(0): income growth rate 0.05 0.02
R(0): demand met by non-
carbon technologies 0.0097
I(0): initial investment 0.06
* The former USSR is included with the industrial countries.

44
The trans-disciplinary Tyndall Centre for Climate Change Research undertakes integrated research
into the long-term consequences of climate change for society and into the development of
sustainable responses that governments, business-leaders and decision-makers can evaluate and
implement. Achieving these objectives brings together UK climate scientists, social scientists,
engineers and economists in a unique collaborative research effort.
Research at the Tyndall Centre is organised into four research themes that collectively contribute
to all aspects of the climate change issue: Integrating Frameworks; Decarbonising Modern
Societies; Adapting to Climate Change; and Sustaining the Coastal Zone. All thematic fields
address a clear problem posed to society by climate change, and will generate results to guide the
strategic development of climate change mitigation and adaptation policies at local, national and
global scales.
The Tyndall Centre is named after the 19th century UK scientist John Tyndall, who was the first to
prove the Earth’s natural greenhouse effect and suggested that slight changes in atmospheric
composition could bring about climate variations. In addition, he was committed to improving the
quality of science education and knowledge.
The Tyndall Centre is a partnership of the following institutions:
University of East Anglia
UMIST
Southampton Oceanography Centre
University of Southampton
University of Cambridge
Centre for Ecology and Hydrology
SPRU – Science and Technology Policy Research (University of Sussex)
Institute for Transport Studies (University of Leeds)
Complex Systems Management Centre (Cranfield University)
Energy Research Unit (CLRC Rutherford Appleton Laboratory)
The Centre is core funded by the following organisations:
Natural Environmental Research Council (NERC)
Economic and Social Research Council (ESRC)
Engineering and Physical Sciences Research Council (EPSRC)
UK Government Department of Trade and Industry (DTI)

For more information, visit the Tyndall Centre Web site ([Link]) or contact:
External Communications Manager
Tyndall Centre for Climate Change Research
University of East Anglia, Norwich NR4 7TJ, UK
Phone: +44 (0) 1603 59 3906; Fax: +44 (0) 1603 59 3901
Email: tyndall@[Link]
Recent Working Papers

Tyndall Working Papers are available online at

[Link]

Mitchell, T. and Hulme, M. (2000). A Dessai, S., (2001). The climate regime
Country-by-Country Analysis of Past from The Hague to Marrakech: Saving
and Future Warming Rates, Tyndall or sinking the Kyoto Protocol?, Tyndall
Centre Working Paper 1. Centre Working Paper 12.
Hulme, M. (2001). Integrated Dewick, P., Green K., Miozzo, M., (2002).
Assessment Models, Tyndall Centre Technological Change, Industry
Working Paper 2. Structure and the Environment, Tyndall
Centre Working Paper 13.
Berkhout, F, Hertin, J. and Jordan, A. J.
(2001). Socio-economic futures in Shackley, S. and Gough, C., (2002). The
climate change impact assessment: Use of Integrated Assessment: An
using scenarios as 'learning machines', Institutional Analysis Perspective,
Tyndall Centre Working Paper 3. Tyndall Centre Working Paper 14.
Barker, T. and Ekins, P. (2001). How High Köhler, J.H., (2002). Long run technical
are the Costs of Kyoto for the US change in an energy-environment-
Economy?, Tyndall Centre Working Paper economy (E3) model for an IA system:
4. A model of Kondratiev waves, Tyndall
Centre Working Paper 15.
Barnett, J. (2001). The issue of 'Adverse
Effects and the Impacts of Response Adger, W.N., Huq, S., Brown, K., Conway,
Measures' in the UNFCCC, Tyndall Centre D. and Hulme, M. (2002). Adaptation to
Working Paper 5. climate change: Setting the Agenda for
Development Policy and Research,
Goodess, C.M., Hulme, M. and Osborn, T.
Tyndall Centre Working Paper 16.
(2001). The identification and evaluation
of suitable scenario development Dutton, G., (2002). Hydrogen Energy
methods for the estimation of future Technology, Tyndall Centre Working Paper
probabilities of extreme weather 17.
events, Tyndall Centre Working Paper 6.
Watson, J. (2002). The development of
Barnett, J. (2001). Security and Climate large technical systems: implications
Change, Tyndall Centre Working Paper 7. for hydrogen, Tyndall Centre Working
Paper 18.
Adger, W. N. (2001). Social Capital and
Climate Change, Tyndall Centre Working Pridmore, A. and Bristow, A., (2002). The
Paper 8. role of hydrogen in powering road
transport, Tyndall Centre Working Paper
Barnett, J. and Adger, W. N. (2001).
19.
Climate Dangers and Atoll Countries,
Tyndall Centre Working Paper 9. Turnpenny, J. (2002). Reviewing
organisational use of scenarios: Case
Gough, C., Taylor, I. and Shackley, S.
study - evaluating UK energy policy
(2001). Burying Carbon under the Sea:
options, Tyndall Centre Working Paper 20.
An Initial Exploration of Public
Opinions, Tyndall Centre Working Paper 10. Watson, W. J. (2002). Renewables and
CHP Deployment in the UK to 2020,
Barker, T. (2001). Representing the
Tyndall Centre Working Paper 21.
Integrated Assessment of Climate
Change, Adaptation and Mitigation,
Tyndall Centre Working Paper 11.
Watson, W.J., Hertin, J., Randall, T., Gough, Xueguang Wu, Jenkins, N. and Strbac, G.
C. (2002). Renewable Energy and (2003). Integrating Renewables and CHP
Combined Heat and Power Resources in into the UK Electricity System:
the UK, Tyndall Centre Working Paper 22. Investigation of the impact of network
faults on the stability of large offshore
Paavola, J. and Adger, W.N. (2002). Justice
wind farms, Tyndall Centre Working Paper
and adaptation to climate change,
32
Tyndall Centre Working Paper 23.
Xueguang Wu, Jenkins, N. and Strbac, G. Pridmore, A., Bristow, A.L., May, A. D. and
(2002). Impact of Integrating Tight, M.R. (2003). Climate Change,
Renewables and CHP into the UK Impacts, Future Scenarios and the Role
Transmission Network, Tyndall Centre of Transport, Tyndall Centre Working Paper
Working Paper 24 33
Xueguang Wu, Mutale, J., Jenkins, N. and
Dessai, S., Hulme, M (2003). Does climate
Strbac, G. (2003). An investigation of
policy need probabilities?, Tyndall Centre
Network Splitting for Fault Level
Working Paper 34
Reduction, Tyndall Centre Working Paper
25
Tompkins, E. L. and Hurlston, L. (2003).
Brooks, N. and Adger W.N. (2003). Country Report to the Cayman Islands’
level risk measures of climate-related Government. Adaptation lessons learned
natural disasters and implications for from responding to tropical cyclones by
adaptation to climate change, Tyndall the Cayman Islands’ Government, 1988
Centre Working Paper 26 – 2002, Tyndall Centre Working Paper 35
Tompkins, E.L. and Adger, W.N. (2003).
Kröger, K. Fergusson, M. and Skinner, I.
Building resilience to climate change
(2003). Critical Issues in Decarbonising
through adaptive management of
Transport: The Role of Technologies,
natural resources, Tyndall Centre Working
Tyndall Centre Working Paper 36
Paper 27
Dessai, S., Adger, W.N., Hulme, M., Köhler, Ingham, A. and Ulph, A. (2003)
J.H., Turnpenny, J. and Warren, R. (2003). Uncertainty, Irreversibility, Precaution
Defining and experiencing dangerous and the Social Cost of Carbon, Tyndall
climate change, Tyndall Centre Working Centre Working Paper 37
Paper 28
Brooks, N. (2003). Vulnerability, risk and
Brown, K. and Corbera, E. (2003). A Multi-
adaptation: a conceptual framework,
Criteria Assessment Framework for
Tyndall Centre Working Paper 38
Carbon-Mitigation Projects: Putting
“development” in the centre of
Tompkins, E. L. and Adger, W.N. (2003).
decision-making, Tyndall Centre Working
Defining response capacity to enhance
Paper 29
climate change policy, Tyndall Centre
Hulme, M. (2003). Abrupt climate Working Paper 39
change: can society cope?, Tyndall
Centre Working Paper 30 Klein, R.J.T., Lisa Schipper, E. and Dessai, S.
(2003), Integrating mitigation and
Turnpenny, J., Haxeltine A. and O’Riordan,
adaptation into climate and
T. (2003). A scoping study of UK user
development policy: three research
needs for managing climate futures.
questions, Tyndall Centre Working Paper 40
Part 1 of the pilot-phase interactive
integrated assessment process (Aurion
Project), Tyndall Centre Working Paper 31
Watson, J. (2003), UK Electricity
Scenarios for 2050, Tyndall Centre
Working Paper 41

Kim, J. A. (2003), Sustainable

Development and the CDM: A South
African Case Study, Tyndall Centre
Working Paper 42

Anderson, D. and Winne, S. (2003),

Innovation and Threshold Effects in
Technology Responses to Climate
Change, Tyndall Centre Working Paper 43