Content uploaded by Rene A. Carmona
Author content
All content in this area was uploaded by Rene A. Carmona
Content may be subject to copyright.
Key words. Emissions markets, Cap-and-trade schemes, Equilibrium models, Environmental
Finance.
MARKET DESIGN FOR EMISSION TRADING SCHEMES
REN´
E CARMONA ∗, MAX FEHR †, JURI HINZ ‡,AND ARNAUD PORCHET §
Abstract. The main thrust of the paper is the design and the numerical analysis of new cap-
and-trade schemes for the control and the reduction of atmospheric pollution. The tools developed
are intended to help policy makers and regulators understand the pros and the cons of the emissions
markets. We propose a model for an economy where risk neutral firms produce goods to satisfy an
inelastic demand and are endowed with permits by the regulator in order to offset their pollution
at compliance time and avoid having to pay a penalty. Firms that can easily reduce emissions
do so, while those for which it is harder buy permits from those firms anticipating that they will
not need them, creating a financial market for pollution credits. Our model captures most of the
features of the European Union Emissions Trading Scheme. We show existence of an equilibrium and
uniqueness of emissions credit prices. We also characterize the equilibrium prices of goods and the
optimal production and trading strategies of the firms. We choose the electricity market in Texas
to illustrate numerically the qualitative properties observed during the implementation of the first
phase of the European Union cap-and-trade CO2emissions scheme, comparing the results of cap-
and-trade schemes to the Business As Usual benchmark. In particular, we confirm the presence of
windfall profits criticized by the opponents of these markets. We also demonstrate the shortcomings
of tax and subsidy alternatives. Finally we introduce a relative allocation scheme which despite of
its ease of implementation, leads to smaller windfall profits than the standard scheme.
1. Introduction. Emission trading schemes, also known as cap and trade sys-
tems, have been designed to reduce pollution by introducing appropriate market mech-
anisms. The two most prominent examples of existing cap and trade systems are the
EU-ETS (European Union Emission Trading Scheme) and the US Sulfur Dioxide
Trading System. In such systems, a central authority sets a limit (cap) on the total
amount of pollutant that can be emitted within a pre-determined period. To en-
sure that this target is complied with, a certain number of credits are allocated to
appropriate installations, and a penalty is applied as a charge per unit of pollutant
emitted outside the limits of a given period. Firms may reduce their own pollution
or purchase emission credits from a third party, in order to avoid accruing potential
penalties. The transfer of allowances by trading is considered to be the core principle
leading to the minimization of the costs caused by regulation: companies that can
easily reduce emissions will do so, while those for which it is harder buy credits.
In a cap-and-trade system, the initial allocation (i.e. the total number of al-
lowances issued by the regulator) should be chosen in order for the scheme to reach a
given emissions level. This total initial allocation is indeed the crucial parameter that
the regulator uses as a knob to control the emission level. But while the value of the
total initial allocation is driven by the emissions target, the specific distribution of
these allowances among the various producers and market participants can be chosen
∗Department of Operations Research and Financial Engineering, Princeton University, Prince-
ton, NJ 08544. Also with the Bendheim Center for Finance and the Applied and Computational
Mathematics Program. (rcarmona@princeton.edu).
†Institute for Operations Research, ETH Zurich, CH-8092 Zurich, Switzerland
(maxfehr@ifor.math.ethz.ch).
‡National University of Singapore, Department of Mathematics, 2 Science Drive, 117543 Singa-
pore (mathj@nus.edu.sg), Partially supported by WBS R-703-000-020-720 / C703000 of the Risk
Management Institute at the National University of Singapore
§Timbre J320, 15 Boulevard Gabriel Pri, 92245 MALAKOFF Cedex, FRANCE,
(arnaud.porchet@gmail.com).
1
2R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
in order to create incentives to design and build cleaner and more efficient production
units.
Naturally, emissions reduction increases the costs of goods whose production
causes those emissions. Part or all of these costs are passed on to the end con-
sumer and substantial windfall profits are likely to occur. Based on an empirical
analysis of power generation profitability in the context of EU-ETS, strong empir-
ical evidence of the existence of such profits is given in [14]. The authors of this
study come to the conclusion that power companies realize substantial profits since
allowances are received for free while they are always priced into electrical power at a
rate that depends upon the emission rate of the marginal production unit: producers
seem to take advantage of the trading scheme to make extra profit. This phenomenon
can even happen in a competitive setting. What follows is a simple illustration in a
deterministic framework.
Let us consider a set of firms that must satisfy a demand of D= 1 MWh of
electricity at each time t= 0,1,· · · , T −1, and let us assume that there are only
two possible technologies to produce electricity: gas technology which has unit cost
2 $ and emits 1 ton of CO2per MWh, and coal technology which has unit cost 1 $
and emits 2 tons of CO2per MWh. In this simple model, the total capacity of gas
is 1 MWh and coal’s capacity is also 1 MWh. We also suppose that producers face
a penalty π > 1 $ per ton of CO2not offset by credits, and that a total of T−1
credits are distributed to the firms, allowing them to offset altogether T−1 tons of
CO2. In this situation, we arrive at two conclusions. First, as demand needs to be
met, total emissions will be higher or equal than Ttons, even if all firms use the clean
technology (gas). Second, firms are always better off reducing emissions than paying
the penalty. As a consequence, the optimal generation strategy is to only use the
gas technology and emit Ttons of CO2. At least one firm has to pay the penalty,
and the price of emission credits is necessarily equal to πat each time. Indeed the
missing credit has a value πfor both the buyer and the seller. The price of electricity
is then 2 + πbecause a marginal decrease in demand will induce a marginal gain in
generation cost and a marginal decrease of the penalty paid. The total profit for the
producers is π(T−1), the penalty paid by the producers to the regulator is π, and the
total cost for the customers is (2 + π)T. Consider now the Business As Usual (BAU)
situation: the demand is met by using coal technology, the price of electricity is 1, the
total profit for producers is 0 and the total cost for the customers is T. In this simple
example the producers cost induced by the trading scheme is T+π: producers must
buy more expensive fuel, so a profit Tis made by the fuel supplier and they have to
pay the penalty π. The increase in fuel price, or switching cost, is a marginal cost
that must factor into the electricity price. The penalty is a fixed cost paid at the end,
but we see that in this trading scheme, this fixed cost is rolled over the entire period
and paid by the customers at each time, inducing a windfall profit for the producers.
This windfall profit is exactly equal to the market value of the T−1 credits: these
credits are given for free by the regulator but their market value is actually funded
by the customer.
Another feature of emissions trading schemes is the risk of non compliance faced
by the producers and the regulator. The EU-ETS was introduced as a way of comply-
ing with the targets set by the Kyoto Protocol. Phase 1 of the Kyoto Protocol sets a
fixed cap for annual emissions of CO2by year 2012 to all industrialized countries that
ratified the protocol (Annex I countries). This reduction should guarantee on average
a level of emissions of 95 % of what it was in year 1990. All countries are free to
Market Designs for Emissions Trading Schemes 3
adopt the emission reduction policy of their choice, but in case of non-compliance in
2012, they face a penalty (payment of 1.3 emission allowances for each ton not offset
in Phase 1). The EU-ETS was designed to ensure compliance for the whole EU zone.
However, in an uncertain environment, there exists the possibility that the scheme
will fail its goal and that the producers will exceed the fixed cap set at the beginning
of the compliance period. In this case, it is the regulator’s responsibility to com-
ply with the target by buying allowances from other countries or generate additional
allowances by investing in clean projects under the Clean Development Mechanism
(CDM for short) or the Joint Implementation (JI for short) mechanism, or otherwise,
to pay the penalty. The design of emission trading schemes must also address this
question.
In the present work, we give a precise mathematical foundation to the analysis of
emission trading schemes and quantitatively investigate the impact of emission regu-
lation on consumers costs and company’s profits. Based on an equilibrium model for
perfect competition, we show that the action of an emission trading scheme combines
two contrasting aspects. On the one hand, the system reduces pollution at the low-
est cost for the society, as expected. On the other hand, it forces a notable transfer
of wealth from consumers to producers, which in general exceeds the social costs of
pollution reduction.
In a perfect economy where all customers are shareholders, windfall profits are
redistributed, at least partially, by dividends. However, this situation is not the
general case and the impact of regulation on prices should be addressed. There
are several other ways to return part of the windfall profits to the consumers. The
most prominent ones are taxation and charging for the initial allowance distribution.
Beyond the political risks associated with new taxes, we will show that one of the main
disadvantages of this first method is its poor control of emissions under stochastic
abatement costs. Concerning auctioning, it is important to notice that, in the first
phase of the EU-ETS, individual countries did not have to give away the totality of
their credit allowances for free. They could choose to auction up to 10% of their
total allowances. Strangely enough, except for Denmark, none of them exercised this
option. On the other hand using auctioning as a way to abolish windfall profits,
one looses one of the main features of cap-and-trade schemes, namely the mechanism
which allows to control the incentives to invest in and develop cleaner production
technologies. Indeed, a significant reduction of windfall profits through auctioning,
if at all possible, requires that a huge amount or even the total initial allocation
is auctioned. Further it involves a significant risk for companies since the capital
invested to procure allowances at the auction may be higher than the income later
recovered from allowances prices.
In this work, we argue that cap-and-trade schemes can work, even in the form
implemented in the first phase of EU-ETS, at least as long as allowance distribution is
properly calibrated. Moreover, we prove that it is possible to design modified emission
trading schemes that overcome these problems. We show how to establish trading
schemes that reduce windfall profits while exhibiting the same emission reduction
performance as the generic cap and trade system used in the first implementation
phase of the EU-ETS. These schemes also have the nice feature that a significant
amount of the allowances can be allocated as initial allocation to encourage cleaner
technologies.
Despite frequent articles in the popular press and numerous speculative debates
in specialized magazines and talk-shows, the scientific literature on cap-and-trade
4R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
systems is rather limited. We briefly mention a few related works chosen because of
their relevance to our agenda. The authors of [3] and [9] proposed a market model for
the public good environment introduced by tradable emission credits. Using a static
model for a perfect market with pollution certificates, [9] shows that there exists a
minimum cost equilibrium for companies facing a given environmental target. The
conceptual basis for dynamic permit trading is, among others, addressed in [2], [15],
[11], [7], [12] and [13]. Meanwhile, the recent work [13] suggests also a continuous-time
model for carbon price formation. Beyond these themes, there exists a vast literature
on several related topics, including equilibrium [1], empirical evidence from already
existing markets [6], [14], and uncertainty and risk [5], [8], [16]. The model we present
below follows the baseline suggested in [4].
We close this introduction with a quick summary of the contents of the paper.
Section 2 gives the details of the mathematical model used to capture the dynamic
features of a cap-and-trade system. We introduce the necessary notation to describe
the production of goods and the profit mechanisms in a competitive economy. Ex-
ogenous demand for goods is modeled by means of adapted stochastic processes. We
assume that demand is inelastic and has to be met exactly. This assumption could
be viewed as unusually restrictive, but we argue that it is quite realistic in the case
of electricity. We also introduce the emissions allowance allocations and the rules of
trading in these allowances.
Section 3 defines the notion of competitive equilibrium for risk neutral firms in-
volved in our cap-and-trade scheme. Preliminary work shows that most of the theoret-
ical results of this paper still hold for risk averse firms if preferences are modeled with
exponential utility. However, in order to avoid muddying the water with unnecessary
technical issues which could distract the reader from the important issues of pollution
abatement, we restrict ourselves to the less technical case of risk neutral firms. For
the sake of completeness, we solve the equilibrium problem in the Business As Usual
(BAU from now on) case corresponding to the absence of market for emissions per-
mits. In this case, as expected, the prices of goods are given by the standard merit
order pricing typical of deregulated markets. The section closes with the proof of a
couple of enlightening necessary conditions for the existence of an equilibrium in our
model. These mathematical results show that at compliance time, the equilibrium
price of an emission certificate can only be equal to 0 or to the penalty level chosen
by the regulator. The second important necessary condition shows that in equilib-
rium, the prices of the goods are still given by a merit order pricing provided that
the production costs are adjusted for the cost of emissions. This result is important
as it shows exactly how the price of pollution gets incorporated in the prices of goods
in the presence of a cap-and-trade scheme. The following Section 4 is devoted to the
rigorous proof of the existence of an equilibrium. The proof uses classical functional
analysis results on optimization in infinite dimensional spaces. It follows the lines
of a standard argument based on the analysis of what an informed central planner
(representative agent) would do in order to minimize the social cost of meeting the
demand for goods.
Section 5 is devoted to the analysis of the standard cap-and-trade scheme featured
in the implementation of the first phase of the EU-ETS. By comparison with BAU
scenarios, we show that properly chosen levels of penalty and pollution certificate
allocations lead to desired emissions targets. However, our numerical experiments
on a case study of the electricity market in Texas show the existence of excessive
windfall profits. As explained earlier in our literature review, these profits have been
Market Designs for Emissions Trading Schemes 5
observed in the first phase of EU-ETS, giving credibility to the critics of cap-and-
trade systems. Section 6 can be viewed as the main thrust of the paper beyond the
theoretical results proven up to that point. We propose a general framework includ-
ing taxes and subsidies along the standard cap-and-trade schemes. We demonstrate
the shortcomings of the tax systems which suffer from poor control of the windfall
profits and unexpected expensive reduction policies when it comes to emissions re-
duction targets under stochastic abatement costs. We concentrate our analysis on
several new alternative cap-and-trade schemes and we show numerically that a rela-
tive allocation scheme can resolve most of the issues with the other schemes. Such a
relative allocation scheme is easy to describe and implement as pollution allowances
are distributed proportionally to production. Even though the number of permits is
random in a relative scheme, and hence cannot be known in advance, its statistical
distribution is well understood as it is merely a scaled version of the distribution of
the demands for goods. Consequently, setting up caps to meet pollution targets is not
much different from the standard cap-and-trade schemes. Moreover, the coefficient of
proportionality providing the number of permits is an extra parameter which should
make the calibration more efficient. Indeed, one shows that properly calibrated, the
relative schemes reach the same pollution targets as the standard schemes while at
the same time, they keep social costs and windfall profits in control.
Section 7 gathers more mathematical properties of the generalized cap-and-trade
schemes introduced in the previous section. Our results demonstrate the versatility
and the flexibility of such a generalized framework. It shows that regulators can con-
trol cap-and-trade schemes in order to reach pre-assigned pollution targets with zero
windfall profits and reasonably small social costs, or even to force equilibrium elec-
tricity prices to be equal to target prices. However, because of the level of complexity
of their implementations, it is unlikely that the schemes identified there will be used
by policy makers or regulators. The paper concludes with Section 8 which reviews
the main results of the paper recasting them in the perspective of the public policy
challenging issues uncovered by the results of the paper.
2. Standard Cap-and-Trade Scheme. In this section we present the elements
of our mathematical analysis. We consider an economy where a set of firms produce
and supply goods to end-consumers over a period [0, T ]. The production of these goods
is a source of pollutant emissions. In order to reduce this externality, a regulator
distributes emissions allowances to the firms at time 0, allows them to trade the
allowances on an organized market between times 0 and T, and at the end of this
compliance period, taxes the firms proportionally to their net cumulative emissions.
In what follows (Ω,F,{Ft, t ∈ {0,1, . . . , T }},P) is a filtered probability space.
We denote by E[.] the expectation operator under probability Pand by Et[.] the
expectation operator conditional to Ft. The σ-field Ftrepresents the information
available at time t. We will also make use of the notation Pt(.) := Et[1{.}] for the
conditional probability with respect to Ft.
2.1. Production of Goods. A finite set Iof firms produce and sell a set K
of different goods at times 0,1, . . . , T −1. Each firm i∈Ihas access to a set Ji,k
of different technologies to produce good k∈K, that are sources of emissions (e.g.
greenhouse gases ). Each technology j∈Ji,k is characterized by:
•a marginal cost e
Ci,j,k
tof producing one unit of good kat time t;
•an emission factor ei,j,k measuring the volume of pollutants emitted per unit
of good kproduced by firm iwith technology j;
•a production capacity κi,j,k.
6R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
For the sake of notation we introduce the index sets
Mi={(j, k) : k∈K, j ∈Ji,k}, i ∈I ,
M={(i, j, k) : i∈I , k ∈K, j ∈Ji,k}.
In this paper, our main example of produced good is electricity. We make the assump-
tion that the production costs are non-negative, adapted and integrable processes.
At each time 0 ≤t≤T−1, firm i∈Idecides to produce throughout the period
[t, t + 1) the amount ξi,j,k
tof good k∈K, using the technology j∈Ji,k . Since the
choice of the production level ξi,j,k
tis based only on present and past observations, the
processes ξi,j,k are supposed adapted and, since production cannot exceed capacity,
we require that the inequalities
0≤ξi,j,k
t≤κi,j,k, i ∈I, k ∈K, j ∈Ji,k, t = 0,1,· · · , T −1,(2.1)
hold almost surely. Our market is driven by an exogenous and inelastic demand for
goods. Since electricity production is a significant proportion of the emissions covered
by the existing schemes, this inelasticity assumption is reasonable. We denote by Dk
t
the demand at time tfor good k∈K. This demand process is supposed to be adapted
to the filtration {Ft}t. For each good k∈K, we assume that the demand is always
smaller than the total production capacity for this good, namely that:
0≤Dk
t≤X
i∈IX
j∈Ji,k
κi,j,k almost surely, k∈K. (2.2)
This assumption is a natural extension of the assumption of inelasticity of the de-
mand as it will conveniently discard issues such as blackouts which would only be a
distraction given the purposes of the paper.
2.2. Emission Trading. We denote by π∈[0,∞) the penalty per unit of pol-
lutant. For example, in the original design of the European Union Emissions Trading
Scheme (EU-ETS) πwas set to 40€per metric ton of Carbon Dioxyde equivalent
(tCO2e). For each firm, the net cumulative emission is the amount of emissions which
have not been offset by allowances at the end of the compliance period. It is com-
puted at time Tas the difference between the total amount of pollutants emitted over
the entire period [0, T ] minus the number of allowances held by the firm at time T
and redeemed for the purpose of emissions abatement. The net cumulative emission
is this difference whenever positive, and 0 otherwise.
For the sake of simplicity we assume that the entire period [0, T ] corresponds to
one simple compliance period. In particular, at maturity T, all the firms have to
cover their emissions by allowances or pay a penalty. Moreover, certificates become
worthless if not used as we do not allow banking from one phase to the next. So in this
economy, operators of installations that emit pollutants will have two fundamental
choices in order to avoid unwanted penalties: reduce emissions by producing with
cleaner technologies or buy allowances.
At time 0, each firm i∈Iis given an initial endowment of Λi
0allowances. So if
it were to hold on to this initial allowance endowment until the end, it would be able
to offset up to Λi
0units of emissions, and start paying only if its actual cumulative
emissions exceed that cap level. This is the cap part of a cap-and-trade scheme.
Depending upon their views on the demands for the various products and their risk
appetites, firms may choose production schedules leading to cumulative emissions in
Market Designs for Emissions Trading Schemes 7
excess of their caps. In order to offset expected penalties, they may engage in buying
allowances from firms which expect to meet demand with less emissions than their
own cap. This is the trade part of a cap-and-trade schemes.
Remark 1. A first generalization of the above allowance distribution scheme
is to reward the firms with allocations Λi
tat each time t= 0,1,· · · , T −1. Even
though modeling EU-ETS would only require one initial (deterministic) allocation Λi
0
for each firm, we shall assume that the distribution of pollution permits is given by
adapted stochastic processes {Λi
t}t=0,1,··· ,T −1. Indeed, all the theoretical results proven
in the paper hold for these more general permit allocation processes since existence,
uniqueness and characterization of the equilibrium price processes depend only upon
the total number of emission permits issued during the compliance period, not on the
way the permits are distributed over time and among the various economic agents.
However as we will demonstrate, the statistical properties of social costs and wind-
fall profits depend strongly on the way permits are allocated. The challenge faced by
policy makers is to optimally design these allocation schemes to minimize social costs
while satisfying emissions reduction targets, controlling producers windfall profits and
setting incentives for the development of cleaner production technologies. We shall
concentrate on these issues in Sections 6 and 7.
Allowances are physical in nature, since they are certificates which can be re-
deemed at time Tto offset measured emissions. But, because of trading, these
certificates change hands at each time t= 0,1,· · · , T , and they become financial
instruments. However in general the allocation of allowances does not take place at a
single timepoint 0. For example, in EU ETS, allowances are allocated in March each
year, while the 5 year compliance period starts in January. Therefore a significant
amount of allowances are traded via forward contracts. Because compliance takes
place at time T, and only at that time, we will restrict ourselves to the situation
where trading of emission allowances is done via forward contracts settled at time T.
Remark 2. Because compliance takes place at time T, a simple no-arbitrage ar-
gument implies that the forward and spot allowance prices differ only by a discounting
factor, such that trading allowances or forwards gives the same expected discounted
payoff at time T. Therefore under the equilibrium definition that will be introduced
in Section 3, considering only forward trading yields no loss of generality. Moreover
allowing trading in forward contracts in our model provides a more flexible setting: it
is more general than considering only spot trading, since it allows for trading pollution
permits even before these allowances are issued and allocated. This turns out to be an
important feature when dealing with general allocation schemes.
We denote by Atthe price at time tof a forward contract guaranteeing delivery of
one allowance certificate at maturity T. The terminology price at time tis misleading
as there is no exchange of funds at time t.Atis better seen as a strike than a price
in the sense that it is the price (in time Tcurrency) at which the buyer at time tof
the forward contract agrees to purchase the allowance certificate at time T.
Each firm can take positions on the forward market, and we denote by θi
tthe
number of forward contracts held by firm iat the beginning of the time interval
[t, t +1). As usual, θi
t>0 when the firm is long and θi
t<0 when it is short. We define
a trading strategy of firm ias an adapted process {θi
t}t=0,··· ,T . If we denote by fi
t
the quantity of forward contracts bought or sold at time tand throughout the period
[t, t + 1), fibeing an adapted process, the position at time tverifies: θi
t+1 =θi
t+fi
t.
The net cash position resulting from this trading strategy, leading to a net position
8R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
of θi
Tcontracts at time T, is:
RA
T(θ) := −
T
X
t=0
fi
tAt=
T−1
X
t=0
θi
t(At+1 −At)−θi
TAT.(2.3)
We here make the assumption that allowances can be traded until time T, whereas
production of goods is decided at time tfor the whole period [t, t + 1), so that the
last production decision occurs at time T−1. This assumption is reasonable since
production of good is a less flexible process than trading.
2.3. Profits. As we argued earlier, it is natural to work with T-forward al-
lowance contracts because compliance takes place at time T. By consistency, it is
convenient to express all cash flows, position values, firm wealth, and good values in
time T-currency. As a side fringe benefit, this will avoid discounting in the computa-
tions to come. So we use for num´eraire the price Bt(T) at time tof a Treasury (i.e.
non defaultable) zero coupon bond maturing at T. We denote by {˜
Sk
t}t=0,1,··· ,T the
adapted spot price process of good k∈K, and according to the convention stated
above, we shall find it convenient to work at each time twith the T-forward price
Sk
t=˜
Sk
t/Bt(T)
and we skip the dependence in Tfrom the notation of the T-forward price as Tis the
only maturity we are considering.
Hence, a cash flow Xtat time tis equivalently valued as a cash flow Xt/Bt(T) at
maturity T. So if firm ifollows the production policy ξi={(ξi,j,k
t)k∈K j∈Ji,k }T−1
t=0 its
instantaneous revenues at time tfrom goods production is given by
X
(j,k)∈Mi
(˜
Sk
t−e
Ci,j,k
t)ξi,j,k
t
and its time T-forward value is given by:
X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t
provided we set Ci,j,k
t=e
Ci,j,k
t/Bt(T). The total net gains from producing and selling
goods are thus:
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t.(2.4)
In order to hedge their production decisions, firms trade on the emissions market
by adjusting their forward positions in allowances. In addition, at maturity T, each
firm iredeems allowances to cover its emissions and/or pay a penalty. Let
Πi(ξi) :=
T−1
X
t=0 X
(j,k)∈Mi
ei,j,kξi,j,k
t(2.5)
be the actual cumulative emissions of firm iwhen it uses production strategy ξi.
We also suppose that there exists another source of emissions on which firm ihas
Market Designs for Emissions Trading Schemes 9
no control, denoted ∆i, and supposed to be an FT-measurable random variable. If
we think of electricity as one of the produced goods for example, the presence of this
uncontrolled source of emissions can easily be explained. Usually electricity producers
are required to hold a reserve margin in order to respond to short time demand changes
and to protect against sudden outages or unexpectedly rapid ramps in demand. When
scheduling their plants it is not yet known how much of this reserve margin will
be used. Therefore in most markets there is an uncertainty on the exact emission
level when a production decision is made. Alternatively, we can see ∆ias a sink
of emissions, accounting for example for the credits gained from Clean Development
Mechanisms or Joint Implementation mechanisms. In this case it can take negative
values. In a first reading ∆ican be thought of as being 0 for the sake of simplicity. We
shall see later in the paper that its presence helps characterizing the equilibrium of
the economy and that it is a useful tool for modeling several variations of the model.
Introducing the net amount Γiof allowances that producer i∈Ican use to offset the
scheduled emissions by
Γi= ∆i−
T−1
X
t=0
Λi
t(2.6)
the total penalty paid by firm iat time Tis:
π(Γi+ Πi(ξi)−θi
T)+.(2.7)
Combining (2.4) and (2.7) together with (2.3), we obtain the expression for the
terminal wealth (profits and losses at time T) of firm i:
LA,S,i(θi, ξ i) :=
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t
+
T−1
X
t=0
θi
t(At+1 −At)−θi
TAT
−π(Γi+ Πi(ξi)−θi
T)+.(2.8)
To emphasize the mathematical technicalities of the model, we underline the fact
that demands and production costs change with time in a stochastic manner. The
statistical properties of these processes are given exogenously, and are known at time
0 by all firms. Moreover, we always assume that these processes satisfy the constraints
(2.1) and (2.2) almost surely. Agents adjust their production and trading strategies
in a non-anticipative manner to their observations of the fluctuations in demand and
production costs. In turn, the production and trading strategies ξiand θibecome
respectively adapted stochastic processes on the stochastic base of the demand and
production costs.
3. Market Equilibrium. In this section, we follow the common apprehension
that a realistic market state is described by prices which correspond to a so-called
market equilibrium, a situation, where the demand for each product is covered, all
financial positions are in the zero net supply, and each firm is satisfied by its own
strategy. We define such an equilibrium and provide necessary conditions for its
existence.
10 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
3.1. Definition of Equilibrium. For any 1 ≤p≤ ∞ and for any normed
vector space F, we introduce the following space of adapted processes:
Lp
t(F) := (Xs)t
s=0;F-valued, kXsk ∈ Lp(Fs), s = 0, . . . , t.(3.1)
We also introduce the spaces of admissible production strategies:
Ui:= n(ξi
t)T−1
t=0 ∈ L∞
T−1(RMi); 0 ≤ξi,j,k
t≤κi,j,k, t = 0, . . . , T −1o,
U:=
ξ∈Y
i∈I
Ui;X
i∈IX
j∈Ji,k
ξi,j,k
t≥Dk
t, k ∈K, t = 0,· · · , T −1
and the spaces of admissible trading strategies:
Vi(A) := (θi
t)T+1
t=1 ,adapted,kRA
T(θi)k ∈ L1(FT)
V(A) := Y
i∈I
Vi(A).
In order to avoid problems with existence of expected values in (2.8), we suppose that
allowance demand and production costs are integrable:
assumption 1.
Γi∈L1,{Ci
t= (Ci,j,k
t)(j,k)∈Mi}T−1
t=0 ∈ L1
T−1(RMi)i∈I, (3.2)
In what follows, we also use a technical assumption on the nature of the uncontrolled
emissions. Even though this assumption is not needed for most of the equilibrium
existence results, it will help us characterize the prices in equilibrium by ruling out
pathological situations. This technical assumption states that up until the end of the
compliance period, there is always uncertainty about the expected pollution level due
to unpredictable events as described in Section 2.3 in the sense that conditionally
on the information available at time T−1, the sum of all the Γi’s has a continuous
distribution. More precisely, we shall assume that
assumption 2. the FT−1-conditional distribution of Pi∈I∆ipossesses almost
surely no point mass, or equivalently, for all FT−1-measurable random variables Z
P(X
i∈I
∆i=Z)= 0 (3.3)
As we already pointed out, this technical assumption will help us refine the statements
of some of the results leading to the equilibriums.
Following the intuition that given price processes A={At}T
t=0 and S={(Sk
t)k∈K}T−1
t=0
each firm aims at increasing its own wealth by maximizing
(θi, ξi)7→ E[LA,S,i (θi, ξi)],(3.4)
over its admissible investment and production strategies, we are led to define equilib-
rium in the following way:
Definition 1. The pair of price processes (A∗, S∗)∈ L1
T(R)× L1
T−1(R|K|)are
an equilibrium of the market if for each i∈Ithere exists (θ∗i, ξ∗i)∈ Vi(A∗)× Uisuch
Market Designs for Emissions Trading Schemes 11
that:
(i) All financial positions are in zero net supply, i.e.
X
i∈I
θ∗i
t= 0, t = 0, . . . , T (3.5)
(ii) Supply meets demand for each good
X
i∈IX
j∈Ji,k
ξi,j,k
t=Dk
t, k ∈K, t = 0, . . . , T −1 (3.6)
(iii) Each firm i∈Iis satisfied by its own strategy in the sense that
E[LA∗,S∗,i (θ∗i, ξ∗i)] ≥E[LA∗,S ∗,i(θi, ξ i)] for all (θi, ξi)∈ V i(A∗)× U i.(3.7)
3.2. Equilibrium in the Business As Usual Scenario . When the penalty
πis equal to zero, an equilibrium should correspond to the Business As Usual sce-
nario. As we explain below, it is characterized by the classical merit order production
strategy. At time tand for each good k, all the production means of the economy are
ranked by increasing production costs Ci,j,k
t. Demand is met by producing from the
cheapest production means and good k’s equilibrium spot price is the marginal cost
of production of the most expensive production means used to meet demand Dk
t.
More precisely, if (A∗, S∗) is an equilibrium, the optimization problem of firm iis
sup
(θi,ξi)∈Vi(A∗)×Ui
E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t+
T−1
X
t=0
θi
t(At+1 −At)−θi
TAT
.
Trading and production strategies are thus decoupled from each other and we are left
with a classical competitive equilibrium problem where each firm maximizes
sup
ξi∈Ui
E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t
,(3.8)
and the equilibrium prices S∗are set so that supply meets demand. The solution
of this equilibrium problem is given by the following linear program for each good
k∈K:
((ξ∗i,j,k
t)j∈Ji,k )i∈I= argmax((ξi,j,k
t)j∈Ji,k )i∈IX
i∈IX
j∈Ji,k
−Ci,j,k
tξi,j,k
t(3.9)
s.t. X
i∈IX
j∈Ji,k
ξi,j,k
t=Dk
t
ξi,j,k
t≤κi,j,k for i∈I, j ∈Ji,k
ξi,j,k
t≥0 for i∈I, j ∈Ji,k .
for all times t, and the associated equilibrium prices are
S∗k
t= max
i∈I, j ∈Ji,k(Ci,j,k
t)1{ξ∗i,j,k
t>0},(3.10)
12 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
This is exactly the merit order pricing mechanism of electricity that can be observed
in most deregulated electricity markets without emission trading scheme. Conversely,
it is easily seen that the above prices together with the above strategies define an
equilibrium. In Section 4 we will see that even under an emission trading scheme the
dispatching of production among producers is still a merit order-like dispatching with
costs adjusted to take into account the mark-to-market value of emissions.
3.3. Necessary Conditions for the Existence of an Equilibrium. Before
turning to the full characterization of the equilibriums, we present some necessary
conditions that will provide interesting insight.
Proposition 3.1 (Necessary Conditions). Let (A∗, S∗)be an equilibrium and
(θ∗, ξ∗)an associated optimal strategies, then following conditions hold:
(i) Then the allowance price A∗is a bounded martingale with values in [0, π]such that
{A∗
T= 0} ⊇ {Γ + Π(ξ∗)<0},{A∗
T=π} ⊇ {Γ + Π(ξ∗)>0}(3.11)
up to sets of probability zero.
(ii) If moreover Assumption 2 holds, then it follows that A∗is almost surely given by
A∗
t=πE[1{Γ+Π(ξ∗))≥0}|Ft] (3.12)
for all t= 0, . . . , T .
(iii) The spot prices S∗kand the optimal production strategy ξ∗icorrespond to a merit
order-type equilibrium with adjusted costs Ci,j,k
t+ei,j,kA∗
t.
Proof. First let us show that A∗has to be a martingale. This is seen as follows:
if not, there exists a time tand a set A∈Ftof non-zero probability such that
Et[A∗
t+11A]>1AA∗
t(resp. <). Then for each agent i∈Ithe trading strategy given
by ¯
θi
s=θ∗i
sfor all s6=tand ¯
θi
t=θ∗i
t+ 1A(resp ¯
θi
t=θ∗i
t−1A) outperforms the
strategy θ∗i, contradicting the third property of an equilibrium.
To prove (3.11) notice that according to the definition of the equilibrium, θ∗i
T(ω)
coincides for almost all ω∈Ω with the maximizer of
z→ −A∗
T(ω)z−π(Γi(ω)+Πi(ξ∗i)(ω)−z)+.(3.13)
As a consequence, we obtain A∗
T∈[0, π] almost surely, since if A∗
T(ω)6∈ [0, π] then
there exists no maximizer to (3.13). Further, observe that if A∗
T(ω)∈(0, π] then the
maximizer is less than or equal to Γi(ω) + Πi(ξ∗i)(ω) and if A∗
T(ω)∈[0, π) then the
maximizer is greater than or equal to Γi(ω)+Πi(ξ∗i)(ω). This holds for each ihence
following inclusions are satisfied almost surely
{A∗
T∈(0, π]} ⊆ ∩i∈I{θ∗i
T≤Γi+ Πi(ξ∗)} ⊆ {X
i∈I
θ∗i
T≤Γ + Π(ξ∗)}(3.14)
{A∗
T∈[0, π)} ⊆ ∩i∈I{θ∗i
T≥Γi+ Πi(ξ∗)} ⊆ {X
i∈I
θ∗i
T≥Γ + Π(ξ∗)}.(3.15)
That is
{A∗
T∈(0, π]}∩{Γ + Π(ξ∗)<0} ⊆ {X
i∈I
θ∗i
T<0},(3.16)
{A∗
T∈[0, π)}∩{Γ + Π(ξ∗)>0} ⊆ {X
i∈I
θ∗i
T>0}.(3.17)
Market Designs for Emissions Trading Schemes 13
Observe that due to the first equilibrium condition Pi∈Iθ∗i
T= 0, the events on the
right hand sides of (3.16) and (3.17) are sets of probability zero which shows that the
inclusions (3.11) hold almost surely. Condition (ii) is a direct consequence of (i) and
Assumption 2.
Finally, the optimization problem of agent ican be written as:
sup
θi∈Ui
E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−ei,j,kA∗
T)ξi,j,k
t
= sup
θi∈Ui
E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−ei,j,kA∗
t)ξi,j,k
t
(3.18)
thanks to the martingale property of A∗. Comparing the above optimization problem
with (3.8), we observe that the equilibrium can be seen as a competitive production
equilibrium with adjusted costs Ci,j,k
t+ei,j,kA∗
t.
This concludes the proof.
The above results provide a better understanding of what a potential equilibrium
should be. The allowance price must always be in [0, π], which is very intuitive since
buying an extra allowance at time twill result in a gain of at most πat time T.
As highlighted in the previous section, the equilibrium in the BAU scenario can be
related to a global cost minimization problem. We shall see in the next section that
the equilibrium in the presence of a trading scheme enjoys the property of social
optimality in the sense that any equilibrium corresponds to the solution of a certain
global optimization problem, where the total pollution is reduced at minimal overall
costs. We call this optimization problem the representative firm problem. Beyond the
economic interpretation of social-optimality, the importance of the global optimization
problem is that its solution helps calculate the allowance prices in equilibrium. We
now explore this connection in detail.
4. Equilibrium and Global Optimality. In this section, we show rigorously
the existence of an equilibrium as defined in Definition 1. We do so by re-framing
the problem as an equivalent global optimization problem involving a hypothetical
informed central planner (which we call a representative agent). We prove the equiv-
alence of the two approaches, and as a by-product of the necessary condition proven
in the previous section, we derive the uniqueness of the allowance price process.
4.1. The Representative Agent Problem . For each admissible production
strategy ξ={ξi}i∈I∈ U, the overall production costs are defined as
C(ξ) :=
T−1
X
t=0 X
(i,j,k)∈M
ξi,j,k
tCi,j,k
t.
and the overall cumulated emissions as
Π(ξ) :=
T−1
X
t=0 X
(i,j,k)∈M
ei,j,kξi,j,k
t.(4.1)
14 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Using the notation
Γ := X
i∈I
Γi
for the aggregate uncontrolled emissions and allowance endowments, the total costs
from production and penalty payments can be defined as
G(ξ) := C(ξ) + π(Γ + Π(ξ))+, ξ ∈ U .(4.2)
We introduce the global optimization problem
inf
ξ∈U E[G(ξ)] (4.3)
which corresponds to the objective of an informed central planner trying to minimize
overall expected costs. Recall that ξis admissible if ξ∈ U , i.e. if the demand is met
and the production constraints are satisfied. The reason for the introduction of this
global optimization problem is contained in the second necessary condition for the
existence of equilibrium.
Proposition 4.1. If (A∗, S∗)is an equilibrium with associated strategies (θ∗, ξ∗),
then ξ∗is a solution of the global optimization problem (4.3) .
Proof. Obviously, it suffices to show that
E(G(ξ∗)) ≤E(G(ξ)) for all ξ∈ U.(4.4)
In order to do so we notice that:
X
i∈I
E[LA∗,S∗,i (θ∗i, ξ∗i)] = ET−1
X
t=0 X
i,j,k∈M
(S∗k
t−Ci,j,k
t)ξ∗i,j,k
t
+
T−1
X
t=0 X
i∈I
θ∗i
t(A∗
t+1 −A∗
t)−X
i∈I
θ∗i
TA∗
T
−πX
i∈I
(Γi+ Πi(ξ∗i)−θ∗i
T)+
=ET−1
X
t=0 X
k∈K
S∗k
tX
i∈IX
j∈Ji,k
ξ∗i,j,k
t−C(ξ∗)
−πX
i∈I
(Γi+ Πi(ξ∗i)−θ∗i
T)+
where we used the fact that in equilibrium, Pi∈Iθ∗i
t= 0 holds for all t= 0, . . . , T
due to condition (i) of Definition 1. Next we use the convexity inequality
X
i∈I
x+
i≥X
i∈I
xi+
Market Designs for Emissions Trading Schemes 15
and once more the fact that the financial positions are in zero net supply to conclude
that
X
i∈I
E[LA∗,S∗,i (θ∗i, ξ∗i)] ≤
T−1
X
t=0 X
k∈K
E[S∗k
tDk
t]−E[C(ξ∗)]
−πEX
i∈I
Γi+X
i∈I
Πi(ξ∗i)+
=
T−1
X
t=0 X
k∈K
E[S∗k
tDk
t]−E[C(ξ∗)] −πE[(Γ + Π(ξ∗))+]
=
T−1
X
t=0 X
k∈K
E[S∗k
tDk
t]−E[G(ξ∗)].
Now, for each ξ∈ U we define θ(ξ) as
θi
t(ξ) = 0 for all i= 1, . . . , N ,t= 0, . . . , T −1,
θi
T(ξ) = Γi+ Πi(ξi)−Γ + Π(ξ)
|I|.
Repeating the above argument for (θ(ξ), ξ) yields
X
i∈I
E[LA∗,S∗,i (θi(ξ), ξi)] = X
t,k
E[S∗k
tDk
t]−E[G(ξ)].(4.5)
Applying the third property (each agent is satisfied with its own strategy) of the
(A∗, S∗) equilibrium to the optimal investment and production strategies (θ∗i, ξ∗i)
and (θi(ξ), ξi) yields
E[G(ξ∗)] ≤X
t,k
E[S∗k
tDk
t]−X
i∈I
E[LA∗,S∗,i (θ∗i, ξ∗i)]
≤X
t,k
E[S∗k
tDk
t]−X
i∈I
E[LA∗,S∗,i (θi(ξ), ξ)] = E[G(ξ)].
This holds for all ξ∈ U completing the proof.
The existence of an optimal ξfor the global optimization problem (4.3) follows
from standard functional analytic arguments.
Proposition 4.2. Under Assumption 1, there exists a solution ξ∈ U of the
global optimal control problem (4.3).
Our proof relies on two simple properties which we state and prove as lemmas for the
sake of clarity. First, we note that L1:= Qi∈IL1
T−1(RM), equipped with the norm
kXk=
T−1
X
t=0 X
(i,j,k)∈M
E[|Xi,j,k
t|]
is a Banach space with dual L∞:= Qi∈IL∞
T−1(RM), the duality form being given by
hX, ξi:=
T−1
X
t=0 X
(i,j,k)∈M
E[Xi,j,k
tξi,j,k
t], X ∈ L1, ξ ∈ L∞.
16 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Next, we consider the weak∗topology σ(L∞,L1) on L∞(see [10]), namely the weakest
topology for which all the linear forms
L∞3ξ7−→ hX, ξi ∈ R,(4.6)
for X∈ L1are continuous.
Lemma 4.3. The real valued function
L∞3ξ7−→ E[G(ξ)] = E[C(ξ)] + πE[(Γ + Π(ξ))+] (4.7)
is lower semi-continuous for the weak∗topology.
Proof. Obviously, the real valued function
L∞3ξ7−→ E[C(ξ)]
is continuous for the weak∗topology since it is of the form ξ7−→ hX, ξifor some
X∈ L1since X=C={Ci,j,k
t}is a fixed element in L1by assumption. So we only
need to prove that the real valued function
L∞3ξ7−→ E[(Γ + Π(ξ))+] (4.8)
is lower semi-continuous. Using the fact that for any integrable random variable X
one has
E[X+] = sup
0≤Y≤1
E[XY ]
one sees that
E[(Γ + Π(ξ))+] = sup
0≤Y≤1
(E[ΓY] + E[YΠ(ξ)])
and hence that the function (4.8) is the supremum of a family of continuous function
since for fixed Y,E[ΓY] is a constant and ξ7−→ E[YΠ(ξ)] is continuous for the weak∗
topology by the very definition of this topology. Since the supremum of any family
of continuous functions is lower semi-continuous, this concludes the proof that (4.7)
is lower semi-continuous.
Lemma 4.4. For the convex subset Uof L∞it holds that:
(i) Uis norm-closed in L1
(ii) Uis weakly∗closed in L∞.
Proof. (i)If (ξn)n∈Nis a sequence in Uconverging in L1to some random vari-
able ξ, then (ξi,j,k
t,n )n∈Nconverges in mean for each (i, j, k)∈Mand t= 0, . . . , T −
1, and extracting a subsequence if necessary, one concludes that (ξi,j,k
t,n )n∈Nand
(Pi∈IPj∈Ji,k ξi,j,k
t,n )n∈Nconverge almost surely to ξi,j,k
tand Pi∈IPj∈Ji,k ξi,j,k
tre-
spectively, showing that the constraints defining Uare satisfied in the limit, implying
that ξ∈ U.
(ii) Since Uis a convex and a norm-closed subset of L1it follows from the Hahn-
Banach Theorem that Uis the intersection of halfspaces Hξ,c ={X∈ L1|hX, ξi ≤ c}
with ξ∈ L∞and c∈Rsuch that U ⊆ H. Since L∞⊆ L1it holds for each of these
halfspaces Hξ,c that ξ∈ L1. Thus we conclude that Hξ,c ∩L∞={X∈ L∞|hX, ξ i ≤ c}
is closed in (L∞, σ(L∞,L1)). Since by definition it holds that U ⊆ L∞it follows that
Uis given by the intersection of the sets Hξ,c ∩ L∞. Since any intersection of closed
sets is closed we conclude that Uis weakly∗closed in L∞.
Market Designs for Emissions Trading Schemes 17
Proof of Proposition 4.2 Since Uis bounded and weakly∗closed due to Lemma 4.4, it
follows from the Theorem of Banach-Alaoglu that Uis weakly∗compact. Lemma 4.3
concludes the proof since any lower semi-continuous function attains its minimum on
a compact set.
4.2. Relation with the Original Equilibrium Problem . As a consequence
of Assumption 2, for each production policy ξ∈ U , no point masses occur in the
FT−1-conditional distribution of Γ −Π(ξ). Hence, for all t= 0, . . . , T −1 we have:
Pt(Γ + Π(ξ)≥0) = Pt(Γ + Π(ξ)>0).(4.9)
In the next theorem, we show that the value of the conditional probability in (4.9)
characterizes the equilibrium allowance price at time t. To prepare for the proof of
this result, we first prove a technical lemma.
Lemma 4.5. Let ξbe any solution of (4.6) whose existence is guaranteed by
Proposition 4.2, then it follows that:
(i) For fixed t∈ {0, . . . , T −1}and any ξ∈ U with ξs=ξsfor all s= 0, . . . , t −1
Et(G(ξ)) ≥Et(G(ξ)) (4.10)
holds almost surely.
(ii) If Assumption 2 is satisfied, then for each k∈Kand i, i0∈I,j∈Ji,k, j 0∈Ji0,k
it holds that
{ξi,j,k
t∈[0, κi,j,k)}∩{ξi0,j 0,k
t∈(0, κi0,j0,k ]}
⊆ {Ci,j,k
t+ei,j,kAt≥Ci0,j 0,k
t+ei0,j0,k At}(4.11)
for all t= 0, . . . , T −1where At=πPt(Γ + Π(ξ)≥0).
Proof. (i) The assertion (4.10) is seen by the following argumentation: On the
contrary, one uses the Ft–measurable set
O:= {Et(G(ξ)) < Et(G(ξ)}of positive measure P(O)>0,
to outperform ξby ξ0as
ξ0
s= 1Oξs+ 1Ω\Oξsfor all s= 0, . . . , T −1. (4.12)
Note that since ξand ξ0coincide at times 0, . . . , t −1, this definition indeed yields an
adapted process ξ0∈ U . With (4.12), we have the decomposition
G(ξ0) = 1OG(ξ)+1Ω\OG(ξ),
which gives a contradiction to the optimality of ξ:
E(G(ξ0)) = E(Et(1OG(ξ)+1Ω\OG(ξ))
=E(1OEt(G(ξ)) + 1Ω\OEt(G(ξ)))
< E(1OEt(G(ξ)) + 1Ω\O Et(G(ξ))) = E(G(ξ)).
(ii) Introduce a deviation from the global optimal strategy ξ. At time t, consider a
shift in production of htunits of the good k∈K, where the agents i∈Iand i0∈I
increase/decrease their outputs from technologies j∈ J i,k,j0∈ J i0,k respectively.
18 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
This results in the new policy ξ+χwhere χ∈Πi∈IUithe deviation vanishes at all
times with the exception of tand
χi,j,k
t=ht, χi0,j0,k
t=−ht,
Consider
D(ξ, λ) = Et(G(ξ+λχ)) −Et(G(ξ))
λ, λ ∈(0,1].
Approaching 0 by λin the countable set (0,1]∩Qwe obtain by dominated convergence
limits for ei,j,k ≤ei0,j0,k0and ei,j,k ≥ei0,j 0,k0as
lim
λ→0D(ξ, λ) = −(Ci,j,k
t−Ci0,j0,k
t)−πPt(Γ + Π(ξ)>0)(ei,j,k −ei0,j0,k )ht
lim
λ→0D(ξ, λ) = −(Ci,j,k
t−Ci0,j0,k
t)−πPt(Γ + Π(ξ)≥0)(ei,j,k −ei0,j0,k )ht.
That is, with Assumption 2 we obtain the limit as
lim
λ→0D(ξ, λ) = −(Ci,j,k
t−Ci0,j0,k
t)−At(ei,j,k −ei0,j0,k )ht.(4.13)
Further, if the production shift is given by
ht= min{κi,j,k −ξi,j,k
t, ξi0,j 0,k
t},
then ξ+λχ ∈ U for all λ∈(0,1] ∩Qwhich, due to (i), yields
D(ξ, λ)(ω)≤0,for all ω∈˜
Ω with P(˜
Ω) = 1.
Passing through the limit λ↓0, we obtain with (4.10) and (4.13)
−(Ci,j,k
t−Ci0,j0,k
t)−Pt(Γ + Π(ξ)>0)(ei,j,k −ei0,j0,k )ht≤0
almost surely. Hence the inclusion
{ht>0} ⊆ {− (Ci,j,k
t−Ci0,j0,k
t)−At(ei,j,k −ei0,j0,k )≤0}
holds almost surely, which is equivalent to (4.11).
We can now turn to the main result of this section.
Theorem 4.6. Under the above assumptions, the following hold:
(i) If ξ∈ U is a solution of the global optimization problem (4.3), then the processes
(A, S)defined by
At=πPt(Γ + Π(ξ)≥0), t = 0, . . . , T (4.14)
and
Sk
t= max
i∈I, j ∈Ji,k(Ci,j,k
t+ei,j,kAt)1{ξi,j,k
t>0}, t = 0, . . . , T −1k∈K, (4.15)
is a market equilibrium (in the sense of Definition 1), for which the associated pro-
duction strategy is ξ.
(ii) The equilibrium allowance price process is almost surely unique.
Market Designs for Emissions Trading Schemes 19
(iii) For each good k∈K, the price Skis the smallest equilibrium price for good kin
the sense that for any other equilibrium price process S∗k, we have Sk≤S∗kalmost
surely.
Proof. (i) We show that (A, S) so defined forms an equilibrium by an explicit
construction of firm investment strategies θi∈ Vi(A) such that (θi, ξi) satisfies (3.5),
(3.6) and (3.7). Define
θi
t= 0 for all i= 1, . . . , N ,t= 1, . . . , T −1,
θi
T= Γi+ Πi(ξi)−Γ + Π(ξ)
|I|.
Since conditions (3.5) and (3.6) are obviously fulfilled, we focus on (3.7). We first
show that E[LA,S,i(θi(ξi), ξ i)] ≥E[LA,S,i(θi, ξi)] for all (θi, ξi)∈ V i(A)× U i, where
θi(ξi) is constant equal to 0 until time T−1 and
θi
T(ξi) := Γi+ Πi(ξi)−Γ + Π(ξ)
|I|.
We have:
E[LA,S,i (θi, ξi)] = E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t−θi
TAT
−π(Γi+ Πi(ξi)−θi
T)+
since Adefined by (4.14) is a bounded martingale. For all ξi∈ Ui, we show that we
can maximize the above quantity by computing the maximum pointwise in θiinside
the expectation. In view of (4.14), when ω∈ {Γ + Π(ξ)<0}we have ω∈AT(ω) = 0
and the maximum of
z7→ −zAT(ω)−π(Γi(ω)+Πi(ξi)(ω)−z)+(4.16)
is attained on each point z∈[Γi(ω) + Πi(ξi)(ω),∞) showing that θi(ξi)(ω) is a
maximizer. On the other hand, when ω∈ {Γ + Π(ξ)≥0}, we have AT(ω) = π, the
maximum of (4.16) is attained on each point z∈(−∞,Γi(ω) + Πi(ξi)(ω)], and once
again, θi(ξi) is a maximizer. Notice for later reference that in both cases, the value
of the maximum of (4.16) is E[−(Γi+ Πi(ξi))AT].
To finish the proof, we prove that E[LA,S,i(θi, ξ i)] ≥E[LA,S,i(θi(ξi), ξi)] for all ξi∈ U i.
According to the above computation, we have:
E[LA,S,i (θi(ξi), ξi)]
=E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t)ξi,j,k
t−(Γi+ Πi(ξi))AT
=E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−ei,j,kAT)ξi,j,k
t−ΓiAT
=E
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−ei,j,kAt)ξi,j,k
t−ΓiAT
.(4.17)
20 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
We now show that the following inclusions hold almost surely:
{Sk
t−Ci,j,k
t−ei,j,kAt>0} ⊆ {ξi,j,k
t=κi,j,k},(4.18)
{Sk
t−Ci,j,k
t−ei,j,kAt<0} ⊆ {ξi,j,k
t= 0}.(4.19)
Inclusion (4.19) is a direct consequence of Definition (4.15) of the price process S.
Using this same Definition (4.15) and Lemma 4.5 we see that:
{Sk
t> Ci,j,k
t+ei,j,kAt}
⊆[
i0∈I,j 0∈Ji0,k
{Ci0,j0,k
t+ei0,j0,k At> Ci,j,k
t+ei,j,kAt}∩{ξi0,j 0,k >0}
⊆[
i0∈I,j 0∈Ji0,k {ξi,j,k
t=κi,j,k}∪{ξi0,j 0,k
t= 0}∩ {ξi0,j0,k >0}
⊆ {ξi,j,k
t=κi,j,k}.
These inclusions allow us to show that E[LA,S,i(θi(ξi), ξi)] ≤E[LA,S ,i (θi, ξi)], thus
completing the proof of (i).
(ii) Proposition 3.1 gives the form of an equilibrium price. Due to Part (i) of
Proposition 3.1 and Proposition 4.1 to prove almost sure uniqueness of the allowance
price process, it is sufficient to prove that for any two solutions ˆ
ξ,˜
ξof the global
optimization problem (4.3) we have:
P{Γ + Π(ˆ
ξ)>0}∩{Γ + Π(˜
ξ)>0}[{Γ + Π(ˆ
ξ)<0}∩{Γ + Π(˜
ξ)<0}= 1
(4.20)
We know that these production strategies are solution of the global problem (4.3),
that we rewrite as a linear programming problem:
inf
ξ∈U, Z ∈L1(FT)
Z≥Γ+Π(ξ)−θ0, Z≥0
E[C(ξ) + πZ].(4.21)
Each solution (ξ?, Z?) of (4.21) satisfies
Z?= (Γ + Π(ξ?))+(4.22)
almost surely. Assume now that there are two optimal solutions (ˆ
ξ, ˆ
Z) and (˜
ξ, ˜
Z) of
the above linear programming problem. Due to the linearity of (4.21) it follows that
any convex linear combination
(λˆ
ξ+ (1 −λ)˜
ξ, λ ˆ
Z+ (1 −λ)˜
Z) (4.23)
is also a solution to (4.21) for all λ∈[0,1]. In view of (4.22), we conclude that for
each λ∈[0,1]
λ(Γ + Π(ˆ
ξ))++ (1 −λ)(Γ + Π(˜
ξ))+
=λ(Γ + Π(ˆ
ξ)) + (1 −λ)(Γ + Π(˜
ξ))+
holds almost surely. Since the above assertion is obviously violated on
{Γ + Π(ˆ
ξ)<0<Γ + Π(˜
ξ)}∪{Γ + Π(ˆ
ξ)>0>Γ + Π(˜
ξ)}
Market Designs for Emissions Trading Schemes 21
this union must have a probability 0, which together with Assumption 2 yields (4.20).
(iii) Assume on the contrary that there exists an equilibrium price process S∗with
S∗k
t(ω)<¯
Sk
t(ω) for all ω∈B(4.24)
for some t∈ {0,1, . . . , T −1},B∈ Ft,P(B)>0 and k∈K. Let ξ∗be the
corresponding equilibrium strategies. Since equilibrium allowance price ¯
Ais unique
it follows from (4.19) that
{S∗k
t−Ci,j,k
t−ei,j,k ¯
At<0} ⊆ {ξ∗i,j,k
t= 0}
up to sets of probability zero. Consequently we obtain
X
i∈IX
j∈Ji,k
ξ∗i,j,k
t=X
i∈IX
j∈Ji,k
ξ∗i,j,k
t1{S∗k
t≥Ci,j,k
t+ei,j,k ¯
At}(4.25)
≤X
i∈IX
j∈Ji,k
κi,j,k1{S∗k
t≥Ci,j,k
t+ei,j,k ¯
At}
almost surely. Moreover it follows from (4.19) and (4.18) that
X
i∈IX
j∈Ji,k
κi,j,k1{¯
St>Ci,j,k
t+ei,j,k ¯
At}(4.26)
=X
i∈IX
j∈Ji,k
¯
ξi,j,k
t−X
i∈IX
j∈Ji,k
¯
ξi,j,k
t1{¯
Sk
t=Ci,j,k
t+ei,j,k ¯
At}
<X
i∈IX
j∈Ji,k
¯
ξi,j,k
t
holds almost surely. In the last equality we used
X
i∈IX
j∈Ji,k
¯
ξi,j,k
t1{¯
Sk
t=Ci,j,k
t+ei,j,k ¯
At}(ω)>0 for all ω∈Ω
which follows from the definition of ¯
S. Further due to (4.24) it holds that
X
i∈IX
j∈Ji,k
κi,j,k1{S∗k
t≥Ci,j,k
t+ei,j,k ¯
At}(ω) (4.27)
≤X
i∈IX
j∈Ji,k
κi,j,k1{¯
Sk
t>Ci,j,k
t+ei,j,k ¯
At}(ω) for all ω∈B.
From (4.25), (4.26) and (4.27) we conclude that there exists a C⊆Bwith
X
i∈IX
j∈Ji,k
ξ∗i,j,k
t(ω)<X
i∈IX
j∈Ji,k
¯
ξi,j,k
t(ω) = Dt(ω)
for all ω∈C, which implies that S∗is no equilibrium product price.
Remark 3. On the basis of what is known for merit-order equilibria with discon-
tinuous cost functions, we do not expect uniqueness of the price process S∗k.
Remark 4. In the introduction, we referred to social costs as the costs of regu-
lation, i.e. the pollution reduction costs. We now give a formal definition of what we
mean by social costs. For each regulatory allocation ((Λi
t)T−1
t=0 )i∈I, and for any choice
22 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
of an equilibrium production schedule ξ∗∈ U, we define the social costs SC as the
random variable given by the difference between the production costs G(ξ∗)under this
production schedule and the production costs incurred in the same random scenarios
had we used the BAU equilibrium production schedule. In other words, the social costs
are given by the random variable:
SC =C(ξ∗)−C(ξ∗
BAU ).(4.28)
Notice also that as defined, the social costs do not depend upon the trading strategies
of the individual firms in the emissions market.
Remark 5. The results of this section were derived under the assumption that
the emission coefficients ei,j,k were constant. However, by mere inspections of the
proofs, the reader will easily convince herself that all the results remain true if these
emission coefficients are instead adapted stochastic processes in L1
T−1(R).
5. Prices and Windfall Profits in the Standard Scheme. The previous
sections were dedicated to the introduction and the mathematical analysis of what we
called the standard emission trading scheme. This cap-and-trade scheme was chosen
because it is representative of the EU-ETS implementation.
In this section, we focus on an economy where one single good is produced. We
choose the example of electricity because the power sector is worldwide one of the
most important sources of green house gases. We study the impact of regulation on
spot prices and producers’ profits. In order to provide insight on the effects of cap-
and-trade legislations, we performed numerical simulations of equilibrium prices and
optimal production schedules by solving the global optimization problem (4.3) using
data from the Texas electricity market. Specifics about the numerical implementation
are given in the Appendix at the end of the paper. We shall report numerical findings
from this case study throughout the remainder of the paper.
5.1. A Model for Electricity and Carbon Trading in Texas . To perform
numerical simulations, we chose to focus on the electricity sector in Texas. Texas has
an installed capacity of 81855 MW, mainly split into gas-fired (51489 MW), coal-fired
(23321 MW), and nuclear (9019 MW) power plants. These figures are based on the
installed capacity in 2007, including also additional nuclear and coal fired power plants
that are planned to come online for the next 7 years. Including upcoming capacity
slightly changes the production stack and leads to more interesting results than using
the actual 2007 installed capacity. Nuclear technology has close to zero emissions, and
it is always running in base-load. The source of emission reduction thus essentially
comes from fuel switching between gas and coal.
So for all practical purposes, our model for Texas can be assume to involve one
good, electricity, produced from two different technologies, gas and coal. Stochastic
costs of production are equal to Ci,j,k
t=HjPj
t, where j∈ {g, c},Hjis the heat rate
of technology jand Pj
tis the corresponding fuel price. Dtstands for the electricity
demand from which nuclear capacity has already been subtracted. We set the emission
rates to 0.42 ton/MWh for gas technology (CCGT-like) and 0.95 ton/MWh for coal
technology respectively. These average emission rates have been chosen to give a
faithful representation of Texas’ park of power plants.
The global optimization problem reads:
inf
ξ∈U E
T−1
X
t=0
(Cg
tξg
t+Cc
tξc
t) + π T−1
X
t=0
(egξg
t+ecξc
t)−Γ0!+
Market Designs for Emissions Trading Schemes 23
under the constraint: ξg
t+ξc
t=Dtfor every time t. In the particular case of two
technologies, we can proceed to the change of variable (ξg
t, ξc
t)7−→ (Et,Ct), where
Et=ecξc
t+egξg
tand Ct=Cc
tξc
t+Cg
tξg
t
are respectively the total emission and the cost of production for the period [t, t +
1). Using the constraint that the demand has to be met, we obtain an equivalent
formulation in terms of an emission abatement problem:
min
E≤E ≤E
E
T−1
X
t=0
(Dt(ecFt+Cc
t)−FtEt) + π T−1
X
t=0
Et−Γ0!+
(5.1)
where:
Et=egmin(Dt, κg) + ec(Dt−κg)+
Et=ecmin(Dt, κc) + eg(Dt−κc)+
are respectively the maximal and emissions possible at time t, and
Ft:= Cg
t−Cc
t
ec−eg(5.2)
is the fuel spread per ton of CO2(or abatement cost). The fuel spread Frepresents
the marginal switching cost necessary to decrease emissions by 1 unit. We observe
that the above formulation (5.1) only involves 2 exogenous stochastic processes: D
and F. Finally, we set the aggregated uncontrolled emissions Pi∈I∆iinfinitesimally
small to stay in the realm of the assumptions of Theorem 4.6, and solve the global op-
timization problem by stochastic dynamic programming on a 2-dimensional trinomial
tree. Details are given in the appendix at the end of the paper.
5.2. Electricity Prices Under the Standard Scheme. In this subsection, we
discuss the impact of the regulation on electricity prices. We already emphasized that
uniqueness of equilibrium electricity prices was not granted. However, we identified
the minimal price among all the possible equilibrium prices in Equation (4.15). In
what follows, we focus on this price.
!
!""!
##
Figure 5.1.Histograms of the consumer costs, social costs, windfall profits and penalty pay-
ments under a standard trading scheme scenario.
24 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Equation (4.15) shows two sources of change in the spot price compared to busi-
ness as usual. First, the marginal technology may be different: this induces a varia-
tion in marginal cost. This variation is likely to be positive but a negative variation
is possible. Suppose for example that in a BAU scenario, coal is started first but
that demand is high enough so that gas is the marginal technology. Suppose that in
the presence of the trading scheme, allowance price is high enough to induce a fuel
switch, so that gas is started first. Assume also that demand is high so that coal is
the marginal technology. In this case, the variation in marginal cost can be negative.
The second source of variation is the price of pollution ei,j,kA∗
tfor the marginal tech-
nology. The producers pass through the cost of expected penalties to end-consumers.
This second contribution is always positive and is such that the spot price under the
trading scheme is always greater than the spot price in BAU.
A possible interpretation of formula (4.15) is that the allowance price enters the
electricity price as the price of an additional commodity that is used for power genera-
tion besides fuels. Producing the last infinitesimal unit of electricity at time tinduces
not only costs due to extra fuel consumption, but also increases the emissions by ei,j,k
and hence also the expected penalty at time Tby ei,j,kA∗
t. Consequently these costs
have to be covered by the end-consumers, for the marginal production of product k
to be profitable. Since this amount is passed on to the endconsumer in each timestep
the consumer cost PT−1
t=0 (S∗
t−SBAU
t)Dtare much bigger than the penalty that is
actually paid. As we will see in the following the consumer costs exceed also by far
the social cost of the scheme.
Figure 5.1 quantifies both the penalty payments and the consumer cost and com-
pares them to social costs and windfall profits (as defined in the next section) under
a standard trading scheme for the Texas electricity sector. The penalty and initial
allocation for this example are π= 100$ and θ0= 1.826 ×108allowances respectively.
This allocation corresponds to a reduction target of 10%, i.e. 1.827 ×108t Carbon, to
be reached with 95% probability.
The results depicted in Figure 5.1 illustrate the major critic articulated by some
of the opponents of the cap-and-trade systems. We observe that end consumer costs,
are approximately more than 10 times higher than social costs due to the trading
scheme. Hence the consumers’ burden exceeds by far the the overall reduction costs,
which gives rise for significant extra profits for the producers.
5.3. Windfall Profits and Penalty Under the Standard Scheme. As ex-
plained above, the pricing mechanism of the standard emissions trading scheme in-
duces a significant wealth transfer from consumers to producers.
Another way of understanding the extra profits made by the producers is to con-
sider the windfall profits defined as follows. In the general framework of a standard
cap-and-trade system with multiple goods introduced earlier, if ξ∗is an optimal pro-
duction strategy associated with the equilibrium (A∗, S∗), we define the target price
ˆ
Sk
tof good kas:
ˆ
Sk
t:= max
i∈I,j ∈Ji,k Ci,j,k
t1{ξ∗i,j,k
t>0}.(5.3)
This price is the marginal cost under the optimal production schedule without taking
into account the cost of pollution. We then define the windfall profits of firm ias:
T−1
X
t=0 X
(j,k)∈Mi
(S∗k
t−ˆ
Sk
t)ξ∗i,j,k
t,
Market Designs for Emissions Trading Schemes 25
and the overall windfall profits as
W P =
T−1
X
t=0 X
k∈K
(S∗k
t−ˆ
Sk
t)Dk
t.(5.4)
These windfall profits measure the profits for the production of goods in excess over
what the profits would have been, had the same dispatching schedule been used, and
the target prices (e.g. the marginal fuel costs) be charged to the end consumers
without the cost of pollution.
Remark 6. Another reasonable definition of the windfall profits of firm iwould
be
T−1
X
t=0 X
(j,k)∈Mi
(S∗k
t−ˆ
Sk
t)ξ∗i,j,k
t−πΓi−Πi(ξ∗i)+(5.5)
meaning that the penalty payments due to the scheme are withdrawn from the extra
profits. Since producers decide upon their production strategy and therewith the risk to
pay the penalty, we take the point of view that they should pay the penalty and not the
endconsumer. However as can be seen in Figure 5.1 the penalty payments vanish in
comparison to the windfall profits as defined in (5.4). Hence in practical applications,
both definitions should give similar results.
Figure 5.1 shows the distribution of windfall profits as computed in the example
of the Texas electricity market chosen for illustration purposes. We observe that the
windfall profits are in average almost 10 times higher than actual abatement costs.
Furthermore it also shows that the costs of expected future penalty passed to the
customers are much higher (4637 times) than the penalty actually paid. This is
consistent with the deterministic example presented in the introduction.
5.4. Incentives for Cleaner Technologies. Using (4.17) we see that the ex-
pected profits and losses of firm i∈Iin an equilibrium (A∗, S∗) with associated
production schedules ξ∗are given by
E[LA∗,S∗,i (θ∗i, ξ∗i)] = E"(−∆i+
T−1
X
t=0
Λi
t)A∗
T#
+E
T−1
X
t=0 X
(j,k)∈Mi
(S∗k
t−Ci,j,k
t−ei,j,kA∗
t)ξ∗i,j,k
t
.(5.6)
As will be shown in Proposition 7.1, both the equilibrium price processes (A∗, S∗) and
the production strategies ξ∗are preserved under a change of the regulatory allocation
from ((Λi
t)T−1
t=0 )i∈Ito ((˜
Λi
0)T−1
t=0 )i∈Ias long as
X
i∈I
T−1
X
t=0
Λi
t=X
i∈I
T−1
X
t=0
˜
Λi
t
holds almost surely. However, such an adjustment of the allocation changes the ex-
pected profits and losses of producer i∈Iby the amount:
Eh(˜
Λi
0−Λi
0)A∗
Ti.(5.7)
26 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Obviously this gives a relative (relative to Λi
0) expected money transfer of
E"X
i∈I
(˜
Λi
0−Λi
0)+A∗
T#(5.8)
from producers with ˜
Λi
0−Λi
0<0 to producers with ˜
Λi
0−Λi
0>0. If the initial allo-
cation is given depending on the type of production plant it is possible to utilize this
mechanism to increase or decrease the incomes of clean and dirty plants respectively,
i.e. the initial allocation can be used to adjust the incentives to build cleaner plants.
Depending on the specific market this will often be the main incentive to build clean
plants.
This mechanism is one of the main features of cap and trade schemes and will in
general fail if auctioning is used to abolish windfall profits. First notice that even a
100% auction can not always reduce windfall profits to zero. This becomes obvious
in a market with a lot of nuclear power plants where coal is marginal the whole time.
In such a market the producers of nuclear power make huge windfall profits but since
their emissions are zero they do not need any allowances. Hence the auction can only
cover the windfall profits due to the coal fired plants. Therefore using auctioning to
cut windfall profits a huge amount if not all allowances of the initial allocation should
be auctioned. However in such a case, the regulator looses the instrument to control
above incentives. Therefore in the next section we propose alternative cap and trade
schemes that not only reduce windfall profits to zero in average, but also provide
a considerable amount of allowances that can be used to adjust incentives to build
cleaner plants.
6. Alternative Designs of Emission Trading Schemes. The main objec-
tives of emission trading schemes are both to force the market to reach a certain reduc-
tion target, and at the same time, to give incentives to develop and build cleaner pro-
duction facilities. In view of the shortcomings of the standard cap-and-trade scheme
demonstrated in the last section, we propose alternative designs which fulfil both ob-
jectives at low social costs, low windfall profits and hence low costs transfered to the
consumer.
This is possible because the mathematical theory developed in the previous sec-
tions allows us to study emissions reduction policies that are different from the stan-
dard EU-ETS scheme.
In the first Subsection 6.1 below, we introduce a general (and fairly complex)
cap-and-trade scheme including taxes and subsidies. We argue that the theoretical
results derived earlier in the paper for standard schemes, still hold in this more gen-
eral situation. The remaining of the section is devoted to the identification and the
calibration of two of the simplest particular cases of interest. A relative scheme is
introduced in Subsection 6.2 and a tax scheme is introduced in Subsection 6.3. The
final Subsection 6.4 provides comparative statics highlighting the differences between
these schemes on the case study of the Texas electricity market.
6.1. General Market Designs for Emission Trading Schemes. We de-
scribe the new regulator policies by first generalizing the allocation procedure. Be-
yond the static allocation Λi
tfor firm iat time t, the regulator is now allowed to
distribute credits dynamically and proportionally to production. To be more specific,
at each time 0 ≤t < T , firm iis provided with an allocation
Λi
t=Xi
t+X
(j,k)∈M(i)
Yi,j,k
tξi,j,k
t,(6.1)
Market Designs for Emissions Trading Schemes 27
where Xiand Yi,j,k are adapted processes in L1
T−1(R). For the sake of generality we
let Yi,j,k
tdepend upon j. However in this case the opportunity to relate the number
of allowances to real emissions is lost.
In addition, the regulator can also tax or subsidize the various firms by means of
financial incentives or disincentives similar to the credit endowments described above.
In this case, the firms’ profits are lowered at time tby an amount
T Si=Vi
t+X
(j,k)∈M(i)
Zi,j,k
tξi,j,k
t,(6.2)
where Viand Zi,j,k are as before, adapted processes in L1
T−1(R). Remark that Vi
and Zi,j,k stand for a tax when positive and a subsidy when negative. Examples of
positive Zi,j,k include fuel and CO2taxes. The combination of Viand Zi,j,k allows
for the introduction of alternative regulation such as a system of reward/penalty with
respect to a given production (or equivalently emission) target ξi,j,k. By charging the
quantity
T−1
X
t=0 X
(j,k)∈Mi
Zi,j,k
t(ξi,j,k
t−ξi,j,k
t)
corresponding to Vi
t=−P(j,k)∈MiZi,j,k
tξi,j,k
t, the regulator can provide incentives
for firm ito stay close to a given production or emission strategy.
Under such a generalized cap-and-trade scheme, the terminal wealth (or profits
and losses) of firm i∈Ireads:
LA,S,i(θi, ξ i) := −
T−1
X
t=0
Vi
t+
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−Zi,j,k
t)ξi,j,k
t
+
T−1
X
t=0
θi
t(At+1 −At)−θi
TAT
−π
∆i+ Πi(ξi)−
T−1
X
t=0
Xi
t+X
(j,k)∈M(i)
Yi,j,k
tξi,j,k
t
−θi
T
+
.(6.3)
Despite the obviously greater generality of the present framework, the proofs of the
results of Theorem 4.6 are sufficient to cover the analysis of this broader class of
trading schemes:
Proposition 6.1. If we set
ˆ
Γi:= ∆i−
T−1
X
t=0
Xi
t,ˆei,j,k
t:= ei,j,k −Yi,j,k
t,and ˆ
Ci,j,k
t:= Ci,j,k
t+Zi,j,k
t(6.4)
for a set of adjusted parameters, then the results of Theorem 4.6 hold true in the case
of the the generalized cap-and-trade scheme of this subsection provided we replace the
parameters of Theorem 4.6 by the adjusted parameters so-defined.
Proof. The proof of this proposition follows a straightforward adaptation of the
arguments used in the previous sections and Remark 5 about stochastic emission
factors.
28 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
The present formulation gives a general framework for the analysis of a broader
class of cap-and-trade schemes. We mostly focus on two important particular cases:
1) the case where Zi,j,k
t≥0 varies with iand jwhich represents a fuel or emission
tax scheme, and 2) the case where Zi,j,k
t≤0 only depends on k, which corresponds
to a subsidy for the production of good k.
For an equilibrium (A∗, S∗) of the generalized scheme with associated strategies
(θ∗, ξ∗) it is straightforward to extend the definition of windfall profits of firm ias:
GW P i=
T−1
X
t=0 X
(j,k)∈Mi
(S∗k
t−ˆ
Sk
t)ξ∗i,j,k
t−
T−1
X
t=0 Vi
t+X
(j,k)∈Mi
Zi,j,k
tξ∗i,j,k
t,
the overall windfall profits being then defined as
GW P =
T−1
X
t=0 X
k∈K
(S∗k
t−ˆ
Sk
t)Dk
t−
T−1
X
t=0 Vi
t+Dk
tX
i∈I,j ∈Ji,k
Zi,j,k
t.(6.5)
The above discussion suggests that windfall profits could be reduced with a relative
allocation rule constant over time. This motivates the following analysis.
6.2. Cap-and-Trade Schemes with Relative Allowance Allocation. A
positive relative allocation for some product Yi,j,k
t=yk<0 for some k∈Kall i∈I
and those j∈Ji,k can be seen as a subsidy for good kthat is given in the form of
allowances rather than in cash. When producing one unit of good k, the marginal
penalty increases only by (ei,j,k −yk)A∗
trather than by ei,j,kA∗
tas in a standard
scheme. Thus the net marginal overall production costs of the firms are lower when
compared to the standard scheme. This should result in a decrease of the price of good
k. In the present subsection, we study the simplest generalized cap-and-trade scheme
taking advantage of this mechanism. It applies this mechanism only for production
means Ji,k
marg ⊆Ji,k which can be marginal (e.g. in the case of the electricity markets,
this excludes nuclear plants) and is obtained by setting:
Yi,j,k
t=yk1{j∈Ji,k
marg}∈Rfor all t= 0, . . . , T −1
Xi
0=xi∈R, Xi
t= 0 for all t= 1, . . . , T −1
Vi
t= 0, Zi,j,k
t= 0 for all t= 0, . . . , T −1
for all (i, j, k)∈M.
In what follows, not only do we discuss this relative cap-and-trade scheme, but
we also gain new insight into the standard cap-and-trade scheme by treating it as a
relative cap-and-trade scheme with yk= 0 for all k∈K.
But for any comparison of different cap-and-trade schemes to be meaningful, we
need to calibrate their respective parameters to common characteristics. We proceed
to the discussion of such a calibration.
6.2.1. Calibration of the Parameters. The relative scheme has three regula-
tory parameters. Using the notation of this section, they are: 1) the penalty π, 2) the
relative allocation coefficients (yk)k∈K, 3) the total initial allocations x=Pi∈Ixi
given to the firms i∈I. In this subsection we show, using again the example of
the Texas electricity market, how one should choose these parameters in order to
guarantee an emissions reduction target with given probability while keeping the ex-
pected windfall profits near zero and controlling the social costs to keep them as low
Market Designs for Emissions Trading Schemes 29
0.0
0.1
0.2
0.3
0.4
0.5
0.6
ye
1.7⋅1081.8⋅1081.9⋅1082.0⋅1082.1⋅108
x+E(yeD)
-10
-5
0
5
10
15
20
25
$/MWh
0.0
0.1
0.2
0.3
0.4
0.5
0.6
ye
1.7⋅1081.8⋅1081.9⋅1082.0⋅1082.1⋅108
x+E(yeD)
-10
-5
0
5
10
15
20
25
$/MWh
0.0
0.1
0.2
0.3
0.4
0.5
0.6
ye
1.7⋅1081.8⋅1081.9⋅1082.0⋅1082.1⋅108
x+E(yeD)
1.7⋅108
1.8⋅108
1.8⋅108
1.8⋅108
1.9⋅108
2.0⋅108
2.0⋅108
t CO2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
ye
1.7⋅1081.8⋅1081.9⋅1082.0⋅1082.1⋅108
x+E(yeD)
1.7⋅108
1.8⋅108
1.8⋅108
1.8⋅108
1.9⋅108
2.0⋅108
2.0⋅108
t CO2
Figure 6.1.Windfall profits (left) and 95% percentile of total emissions (right) as functions of
the relative allocation parameter and the expected allocation. Here D=PT−1
t=0 De
tdenotes the total
electricity demand over one compliance period.
as possible. In the particular simulation used to illustrate the strategy, we choose an
emissions reduction target of 1.827 ×108to be reached with probability 95%.
To gain a first insight into the numerics, we fix the penalty πat 100$. The left
pane of Figure 6.1 gives the expected windfall profits while the right pane gives the
95% percentile of the total emissions for different values of the relative allocation coef-
ficient (ye) and the expected total allocation. It appears that the expected allocation
controls the amount by which carbon emissions are reduced, while the relative alloca-
tion coefficient yecontrols the windfall profits. Designing a cap-and-trade scheme with
zero windfall profits and pre-decided emissions target levels can be done by choosing
the parameters of our relative scheme at the intersection of the zero windfall profit
level set with the 1.827 ×108emission percentile level set. This procedure is depicted
in Figure 6.2. We find ye= 0.54 and Pi∈Ixi= 5.4×107.
30 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Figure 6.2.The left pane shows the level sets of the two plots of Figure 6.1. The blue and the
red lines indicate level sets of the windfall profits and the 95% quantile of emissions respectively.
The right pane gives the plots of the overall production costs for electricity for one year as function
of the penalty level for both the absolute and relative schemes. The free regulatory parameters are
chosen to guarantee the desired emissions percentile, and in the case of the relative scheme, such
that the windfall profits are zero.
Since for the standard cap-and-trade scheme the parameter ye= 0 is fixed, we
have one less regulatory parameter. Thus controlling the emissions level by the initial
allocation, we are not able to control the windfall profits. Hence the desired parameter
values are obtained at the intersection of the 1.827 ×108emission percentile level set
with ye= 0. Giving the initial allocation Pi∈Ixi= 1.826 ×108.
Repeating the above procedure for different penalties levels gives regulatory set-
tings with different production costs for the relative and the standard scheme in Figure
6.2. Obviously for both schemes the social costs are reduced by increasing penalty.
As shown in the right pane of Figure 6.2, this decrease in social costs is significant
until the penalty reaches the level π=100$. after that, the social costs stay nearly
the same, becoming independent of πfor larger values of π. Hence, we conclude that
in this setting a penalty of 100$ is a reasonable choice for both the relative and the
standard scheme.
6.3. Emission Taxes. A static tax scheme is a regulation that penalizes the
emission of each ton of carbon by a fixed amount, say z > 0. Formally, it can be
viewed as a generalized scheme for which
Zi,j,k
t=ei,j,kz , V i
t= 0
Yk
t= 0 , Xi
t= 0
for all (i, j, k)∈Mand t= 0, . . . , T −1. Using the results of Proposition 6.1,
we see that in such a tax scheme the prices of goods follow a merit order pricing
rule with effective production costs given by Ci,j,k
t+ei,j,kzfor all (i, j, k )∈Mand
t= 0, . . . , T −1. The earnings under a tax scheme are based on the spread of these
effective production costs. Since this spread does not depend solely on the original
fuel spread, it is in general not clear what windfall profits will be. It is not even clear
if they are negative or positive. To gain some insight on this issue consider a tax of
Market Designs for Emissions Trading Schemes 31
z= 60$ (which is realistic for a 10% reduction target as will be seen below), and
assume that at some point in time, the marginal production costs of coal and gas are
the same while all plants have to run to satisfy the demand. In this case the spread
in effective production cost is (ec−eg)z= 31.8$ and will be earned for each MWh
that is produced with gas. However, in the case of BAU, the earnings are zero. Hence
the windfall profits are 31.8$ per MWh of electricity produced with gas.
In tax schemes the only regulatory control parameter zshould be adjusted in order
to guarantee a specified reduction target. Thus the windfall profits are automatically
given by the reduction target and can not be adjusted.
6.4. Comparison of the Various Abatement Schemes. We now compare
the characteristics of the standard and the relative cap-and-trade schemes with the
regulatory parameters chosen in the previous subsection.
We first consider the windfall profits and the consumer costs. The results are
given in Figure 6.3.
Figure 6.3.Histograms (computed from 500000 simulation scenarios) of the yearly distribution
of windfall profits (left) and consumer costs (right) for the Standard Scheme, a Relative Scheme and
a Tax Scheme.
As expected the relative scheme gives much lower consumer costs than the stan-
dard scheme. This is related to the fact that the windfall profits have a narrow
distribution around zero in the case of the relative scheme, while the windfall profits
of the standard scheme are 10 times higher than the social costs. When compared
to the standard scheme, the only drawback of the relative scheme seems to be the
slightly higher level of social costs which can be observed on the right pane of Figure
6.4. However since this cost increase corresponds to approximately 0.4$ per MWh
it is small in comparison to production costs and thus can be neglected in practice.
Moreover those higher production costs are not just wasted money, they are paid for
higher emission reduction in many scenarios as can be seen on the left pane of Figure
6.4. In particular the relative scheme takes advantage of cheap fuel switches when the
standard scheme cannot reduce emissions anymore. Moreover relative scheme is less
sensitive to weather, since in warm winters less allowances are allocated pushing the
price up. This in turn is responsible for higher emission abatements and consequently
32 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
!
Figure 6.4.Yearly emissions from electricity production (left) for the Standard Scheme, the
Relative Scheme, a Tax Scheme and BAU, and yearly abatement costs (right).
higher abatement costs.
In this example approximately 30% of the allowances are given as initial allocation,
by allocating these to clean plants, further incentives can be set to build cleaner plants.
This seems to be an important advantage of the relative scheme over other mechanisms
such as auctioning and tax.
Next, we study the effect of an emission tax on the Texas electricity market.
Figure 6.4 shows that a pure tax scheme that fulfills the above reduction target of
1.827×108tCO2with 95% probability, is on average, more than twice (2.4×) as expen-
sive as the standard cap and-trade-scheme. In other words, it has a poor emissions
reduction performance. These extra costs are paid for extra emission reductions.
However in contrast with the results in the case of the relative scheme, the aver-
age cost increase per reduced ton of carbon is considerable when we compare it to
the case of the standard scheme. The reason is that a tax is not flexible enough to
control emissions when abatement costs are stochastic. This results in an emission
uncertainty that exceeds even the BAU uncertainty with several orders of magnitude.
Notice moreover that it carries a significant risk to reduce nearly no emissions. In
such a scenario the tax corrections for upcoming years will be extremely expensive.
Needless to say a tax scheme induces a huge money transfer from consumers to the
regulator, which as can be seen in Figure 6.4 is even bigger than the costs transferred
to the consumer in a standard cap and trade scheme.
7. More Financial Incentives. One of the main arguments in favor of the
relative schemes studied in the previous section is the fact that they reduce windfall
profits. However, this reduction comes with slightly higher reduction costs than in
the case of the absolute scheme. While this cost increase is negligible in practice, it is
of great theoretical interest to understand how and why one can design schemes that
give exactly zero windfall profits at exactly the same reduction costs as the standard
cap-and-trade scheme. In order to do so, we need to identify the generalized schemes
which are in a one-to-one correspondence with the production policies of the standard
Market Designs for Emissions Trading Schemes 33
scheme. The latter are given by a subclass of generalized schemes for which Zi,j,k
t
and Yi,j,k
tdepend only on k. The terminal wealth of firm i∈Iunder such a scheme
reads:
LA,S,i(θi, ξ i):=−
T−1
X
t=0
Vi
t+
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−Ci,j,k
t−Zk
t)ξi,j,k
t
+
T−1
X
t=0
θi
t(At+1 −At)−θi
TAT
−π∆i+ Πi(ξi)−
T−1
X
t=0 Xi
t+X
(j,k)∈M(i)
Yk
tξi,j,k
t−θi
T+
.(7.1)
The results of this section will demonstrate the versatility and the flexibility of the
generalized framework introduced in this paper. However, because of the level of
complexity of their implementations, and despite the high degree of control they
provide the regulator with, it is unlikely that the schemes identified here will be used
by policy makers or regulators.
7.1. Equilibria Equivalence. Our first result exhibits a one-to-one correspon-
dence between the equilibria of standard schemes and generalized schemes leading to
profits and losses for firm iof the type (7.1).
Proposition 7.1. If (A∗, S ∗)is an equilibrium with production strategies ξ∗for
a standard cap-and-trade scheme with adjusted uncontrolled emissions given by
Γi= ∆i−
T−1
X
s=0
(Xi
s+ Ξi
s)for all i∈I , (7.2)
where Ξis a stochastic process in L1
T−1(R)such that
T−1
X
t=0 X
i∈I
Ξi
t=
T−1
X
t=0 X
k∈K
Yk
tDk
t,(7.3)
then the prices (A∗, S†)where
S†k
t=S∗k
t+Zk
t−Yk
tA∗
tfor all k∈K, t = 0, . . . , T −1 (7.4)
define an equilibrium of the generalized cap-and-trade scheme with the same production
strategies ξ∗. The converse statement also holds.
In particular if
T−1
X
t=0 X
i∈I
Xi
t+X
k∈K
Yk
tDk
t=X
i∈I
Λi
0
where Λi
0is the allocation in standard scheme, there is a one-to-one correspondence
between the generalized scheme and the standard scheme with initial allocation Λi
0.
In particular, the lowest equilibrium product price is given by (7.4) where S∗is
given by Theorem 4.6 for the standard scheme with adjusted uncontrolled emissions
(ˆ
Γi)i∈I. To prove Proposition 7.1, we shall need the following lemma.
34 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Lemma 7.2. Let Abe an integrable martingale, θ, θ0∈ V i(A), ξ i∈ Ui, and S,S0
be two integrable price processes, such that:
θ0i
t=θi
tfor all t≤T−1 (7.5)
θ0i
T=θi
T−
T−1
X
t=0 Ξi
t−X
(j,k)∈M(i)
Yk
tξi,j,k
t(7.6)
S0k
t=Sk
t+Zk
t−Yk
tAtfor all k∈K, t = 0, . . . , T −1
where Zk≤0is a subsidy, then we have:
E[LA,S,i(θi, ξ i)] = E[HA,S0,i(θ0i, ξi)] + E"T−1
X
t=0
(Vi
t−Ξi
tAt)#.(7.7)
Proof. The martingale property of Ayields:
E
θ0i
TAT+π
∆i+ Πi(ξi)−
T−1
X
t=0
(Xi
t+X
(j,k)∈M(i)
Yk
tξi,j,k
t)−θ0i
T
+
=E
θi
TAT+
T−1
X
t=0
−Ξi
t+X
(j,k)∈M(i)
ξi,j,k
tYk
t
E(AT|Ft)
+π ∆i+ Πi(ξi)−
T−1
X
t=0
(Xi
t+ Ξi
t)!+
=E
θi
TAT+
T−1
X
t=0
X
(j,k)∈M(i)
ξi,j,k
tYk
t
At+πΓi+ Πi(ξi)−θi
T+
−E(
T−1
X
t=0
Ξi
tAt))
which proves the desired result.
We can now turn to the proof of Proposition 7.1.
Proof. Let (A∗, S ∗) be equilibrium price processes of a standard scheme with
strategies (ξ∗, θ∗). Let θ0∗be the adjusted optimal strategy as in Lemma 7.2. The
assertion follows by checking that conditions (i) to (iii) of Definition 1 are fulfilled
by the pair of price processes (A∗, S†) and strategies (ξ∗, θ0∗). Since θ∗satisfies
the market clearing condition (3.5), so does θ0∗. This proves (i) while condition (ii)
follows directly from (7.3). Moreover, given (θ0, ξi)∈ Vi(A∗)× U i, we define strategies
θi∈ Vi(A∗) such that (7.5) and (7.6) hold. According to Proposition 3.1 A∗is an
integrable martingale and the result of Lemma 7.2 yields
E[HA∗,S†,i (θ0i, ξi)] = E[LA∗,S ∗,i(θi, ξ i)] −E"T−1
X
t=0
(Vi
t−Ξi
tA∗
t)#
≤E[LA∗,S∗,i (θ∗i, ξ∗i)] −E"T−1
X
t=0
(Vi
t−Ξi
tA∗
t)#(7.8)
=E[HA∗,S†,i (θ0∗i, ξ∗i)]
Market Designs for Emissions Trading Schemes 35
where we used the optimality of the equilibrium strategies (ξ∗, θ∗) of the standard
scheme in (7.8). This holds for all (θ0, ξi)∈ V i(A∗)× U iwhich proves condition (iii).
The converse can be proved in exactly the same way.
Note that not only do allocation prices coincide, but also equilibrium production
strategies: ξ†=ξ∗. Thus the switching costs of the generalized cap-and-trade schemes
are the same as for the standard cap-and-trade schemes with adjusted uncontrolled
emissions (Γi)i∈I.
7.2. Design of Financial Incentives. In this paragraph we discuss the design
of financial incentives that adjust the financial positions of each firm i∈Iby
−
T−1
X
t=0
(Vi
t+X
(j,k)∈Mi
ξi,j,k
tZk
t) (7.9)
depending on his production strategy ξi. Obviously, the results of Proposition 7.1
hold in this case. Allowance prices, production strategies, and penalty are identical
in equilibrium, to those of the standard scheme. Electricity price is increased by the
quantity Zk
tat t= 0, . . . , T −1. Hence, as depicted in Figure 7.1 the scheme induces
a money transfer
T−1
X
t=0
(Vi
t+X
(j,k)∈M(i)
ξi,j,k
tZk
t)
from producers to the regulator. In the meantime, the quantity
T−1
X
t=0 X
(j,k)∈M(i)
ξi,j,k
tZk
t
is entirely passed on to the end consumer, so that (Zk
t)T−1
t=0 ≥0 results in a money
transfer from consumers to the regulator.
Remark 7. This explains why it is not trivial to reduce windfall profits (and this
precise discussion of incentives is needed). At first sight one could e.g. think that
windfall profits could be reduced by keeping a book on any traded MWh , marginal
technology and carbon price with the objective to charge the producers the windfall
profits and reimburse consumers later. The amount that would be charged from each
producer i∈Iwould then depend on his strategy ξiand the markets product price
process Swould be given by
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−ˆ
Sk
t)ξi,j,k
t.(7.10)
Obviously under such a regulation the demand will never be satisfied, and hence there
exists no equilibrium. On the other side if we change the amount to be charged to
T−1
X
t=0 X
(j,k)∈Mi
(Sk
t−ˆ
Sk
t)ξi,j,k
t.(7.11)
it follows from above discussion that the entire amount will be passed on to the end-
consumer. Thus the windfall profits will not be reduced.
36 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
Figure 7.1.V > 0gives a money transfer from producers to the regulator, while Z > 0gives a
money transfer from consumer to the regulator. By choosing Vand Zin an appropriate way it is
possible to avoid a money transfer to/from the regulator.
Proposition 7.1 may look esoteric at first. However, it happens to be very versatile
a tool when it comes to designing new schemes with required properties. As corollaries
to this proposition, two appropriate adjustments with zero windfall profits are given
in Sections 7.3 and 7.4.
7.3. Zero Windfall Profit Scheme with Tax and Subsidy. In this section
we consider the generalized allocation scheme given by
Xi
t= Λi
01{t=0}
Yk
t= 0
for all i∈I,k∈Kand t= 0, . . . , T −1. We show here how the tax/subsidy system
comprised into the generalized scheme can theoretically lead to zero windfall profits
at equilibrium. This result is a direct corollary of Proposition 7.1.
Corollary 7.3. Consider a generalized cap-and-trade scheme such that for all
i∈Iand k∈K
Vi
t=X
(j,k)∈M(i)
ξi,j,k
t(St−ˆ
Sk
t)for all t= 0, . . . , T −1
Zk
t= 0 for all k∈K, t = 0, . . . , T −1.(7.12)
Then each equilibrium (A∗, S∗)of the standard cap and trade scheme is also an equi-
librium of this generalized scheme. In particular the equilibrium with lowest product
prices is given by (A, S)from Theorem 4.6. For this scheme the windfall profits for
the aggregated producing sector are zero.
Remark 8. The processes S,ˆ
Sand ξthat occur in (7.12) are given by Theo-
rem 4.6 and hence independent of the actually realized equilibrium (A†, S†)and their
strategies (θ†, ξ†). Hence their computation involves necessarily solving the global opti-
mal control problem. However generalizing Proposition 7.1 easier ways could be found
to compute the amount to charge.
To adjust end-consumer costs to a reasonable level, the amount Pi∈IPT−1
t=0 Vi
t
has to be redistributed from the regulator to the end consumers in an appropriate
way. As can be seen in Figure 7.1 this can be omitted for the financial incentives
fulfilling
T−1
X
t=0 X
i∈I
Vi
t=−
T−1
X
t=0 X
k∈K
Dk
tZk
t.(7.13)
Market Designs for Emissions Trading Schemes 37
Here there is no money transfer from producers to regulators. Such a setting is
discussed in the following Corrolary.
Corollary 7.4. Consider a generalized cap-and-trade scheme such that for all
i∈Iand k∈K
Vi
t=X
(j,k)∈Mi
ξi,j,k
t(Sk
t−ˆ
Sk
t)for all t= 0, . . . , T −1
Zk
t=−(Sk
t−ˆ
Sk
t)for all t= 0, . . . , T −1
where Sdenotes the equilibrium electricity price in the standard scheme, recall Theo-
rem 4.6, and ˆ
Sdenotes the pure merit order price as defined in (5.3). In this setting
each equilibrium (A∗, S∗)of the standard cap-and-trade scheme corresponds to an
equilibrium of the generalized cap-and-trade scheme given by (A∗, S∗−(S−ˆ
S)). In
particular the equilibrium with lowest product price is given by (A, ˆ
S). In this equilib-
rium, each producer realizes null windfall profits.
If a firm ifollows strategy ξiits net money transfer to the regulator is given by
X
(j,k)∈Mi
(ξi,j,k
t−ξi,j,k
t)(Sk
t−ˆ
Sk
t).(7.14)
Notice that this is zero if a firm follows strategy ξ. If it decides to produce more or less
than in equilibrium, it is rewarded or respectively penalized by (7.14). This increases
the incentives to produce and hence results in lower product prices. As shown by
(7.13), the only money transfer in this scheme is the transfer of penalty payments
from penalized firms to rewarded firms.
7.4. Zero Windfall Profit Scheme with Dynamic Allocation. Here we set
V≡Z≡0, and show how dynamic stochastic allocation of allowances can also lead
to null windfall profits for the producers.
Corollary 7.5. Consider a cap and trade scheme with dynamic allowance al-
location where
Xi
t=−X
(j,k)∈M(i)
ξi,j,k
t
Sk
t−ˆ
Sk
t
At
+Λ0
T
Yk
t=Sk
t−ˆ
Sk
t
At
at each time point t= 0, . . . , T −1. Here, Aand Sdenote the equilibrium allowance
and lowest electricity price respectively in the standard scheme from Theorem 4.6.
Further let ˆ
Sbe the pure merit order price defined in (5.3). In this setting it holds
that (Xi
t)T−1
t=0 ∈ L∞
T−1and (Yk
t)T−1
t=0 ∈ L∞
T−1for all i∈Iand k∈K. Each equilib-
rium (A∗, S∗)of the standard cap-and-trade scheme corresponds to an equilibrium of
the generalized cap-and-trade scheme given by (A∗, S∗−(S−ˆ
S)). In particular the
equilibrium with lowest prices for goods is given by (A∗,ˆ
S). In this equilibrium, the
windfall profits of each firm are zero.
Proof. The equilibrium result is a direct consequence of Proposition 7.1. Hence
it remains only to prove that (Xi
t)T−1
t=0 ∈ L∞
T−1and (Yi
t)T−1
t=0 ∈ L∞
T−1for all i∈I. Let
38 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
us first prove that Sk
t−ˆ
Sk
t
Atis bounded from above. Since
Sk
t= max
i∈I,j ∈Ji,k (Ci,j,k
t+ei,j,kAt)1{ξi,j,k
t>0}
≤max
i∈I,j ∈Ji,k Ci,j,k
t1{ξi,j,k
t>0}+ max
i∈I,j ∈Ji,k ei,j,k At1{ξi,j,k
t>0}
≤ˆ
Sk
t+Atmax
i∈I,j ∈Ji,k ei,j,k
it follows that
Sk
t−ˆ
Sk
t
At
≤max
i∈I,j ∈Ji,k ei,j,k for all k∈K, t = 0, . . . , T −1 (7.15)
which gives the upper bound. To prove the lower bound, we notice that for all k∈K
and t= 0, . . . , T −1 it holds that Sk
t−ˆ
Sk
t≥0 and At≥0. Thus
Sk
t−ˆ
Sk
t
At
≥0 for all k∈K, t = 0, . . . , T −1 (7.16)
which concludes the proof.
This scheme ensures zero windfall profits at exactly the same reduction costs
as for the standard cap and trade scheme, without any extra money transfer from
producer to regulator, as e.g. would be the case for auctioning. A further advantage
compared to auctioning is that it allows to distribute the amount Pi∈IXi
tdifferently
to control the incentives to build cleaner plants.
The drawback of this scheme is that it requires a random allocation (following
a prespecified rule) of allowances dynamically through time. Notice that the daily
allocation can be negative for some producers some of the days, this not a problem
as can be seen in Green Certificate Schemes.
Notice further that it suffices to give a relative allocation only to those plants that
are usually marginal. Therefore in electricity markets the relative allocation does not
need to be given to nuclear plants.
The theoretical results of this section suppose market rules that may be uneasy
to enforce in a real market. However they show how the different levers brought by
the generalized schemes can be used to keep the prices of goods at a low level.
8. Conclusions. In this paper we introduced a new mathematical framework for
competitive equilibrium, in which emissions trading schemes can be analyzed. This
framework is general enough to accommodate tax based abatement policies, existing
cap-and-trade schemes such as those implemented in the first phase of the European
Union ETS, as well as new market designs.
The main thrust of the paper is to provide policy makers and regulators with the
tools necessary to design and implement cap-and-trade schemes capable of reaching
reasonable pollution targets at low social costs while controlling windfall profits and
incentives for cleaner production technologies.
On one hand, we develop a rigorous mathematical theory for competitive economic
models in which stochastic demand and production costs are given exogenously. We
prove existence and uniqueness, of an equilibrium in which price processes for goods
and pollution appear endogenously in equilibrium.
On the other hand, we provide analytic and computational tools to analyze and
compare the various emissions trading schemes. Regulators and policy makers need
Market Designs for Emissions Trading Schemes 39
to understand the structure and the role of these new markets vis-a-vis pollution
control, and we view these tools as crucial in the design and the implementation of
sound environmental economic policy.
The computational tools that we developed provide, for each market design
•Monte Carlo scenarios generators for equilibrium prices of goods and pollution
allowance certificates
•Computation for each scenario of
–pollution levels
–end-consumers costs
–producers windfall profits
–social costs
Finally, as illustration of the versatility of the tools developed for the purpose of
the qualitative analysis of cap-and-trade schemes considered in the paper, we imple-
mented them in a case study of the Texas electricity market.
As observed in the SOx and NOx California RECLAIM program and at the end of
the first implementation phase of the EU ETS, cap-and-trade systems can fail as too
generous an allocation of pollution permits will serve as a disincentive for emissions
reductions and will deflate pollution prices. Numerical experiments prove that cap-
and-trade schemes can work in the sense that emissions targets properly chosen
can be reached at low costs.
Moreover we use our computational tools to provide a thorough comparison of a
sample of alternative schemes: the standard scheme inspired by the first EU imple-
mentation phase, an emission tax scheme, and a relative scheme in which allowances
distributions are driven by instantaneous (as opposed to historical) production levels.
Thereby show that the new allowance allocation scheme which we propose can reduce
average windfall profits to zero while keeping reduction cost nearly at the same level
as the standards cap and trade scheme. Further this relative scheme allows to control
the incentives to build new plants.
The following table summarizes the results of our comparative analysis of standard
cap-and-trade schemes (whether or not they including auctioning of allowances) with
tax schemes and the relative scheme which we identified in this paper.
Comparison of Schemes
Red. Target Incentives Windfall Social Cost Cons. Cost
Standard + + −+−
Std&Auct + − − +−
Relative + + + + +
Tax − − +− −
9. Appendix. This final section contains the technical details of the implemen-
tations used to produce the numerical results presented in the paper.
9.1. Model Calibration. We chose to run the numerical experiments with data
coming from the Texan electricity market, because it forms an independent grid, with
few interconnection with other grids. We can thus address the question of the impact
of introducing an emission reduction policy in this state without taking into account
emission leakage. In subsections 9.1.1 and 9.1.2 electricity demand and fuel switch
price process are specified using continuous-time pendants
(D(t))t∈[0,T ],(F(t))t∈[0,T ](9.1)
40 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
for both processes. Note that we write the time parameter in parenthesis instead using
subscript, to indicate continuous-time processes. Moreover, the horizon for continuous
time is [0, T ], where we suppose that the time unit equals to one year. By sampling
(9.1) at discrete times, we obtain discrete-time versions of the processes on a daily
time step resolution, which is used for numerical computations.
!
!"
Figure 9.1.Depicted are the historical daily electricity demand for the Ercot supply area from
18/9/04 to 17/9/05 and the corresponding deterministic part (P(t))t=[0,T ].
9.1.1. Electricity Demand Process. The continuous–time demand process is
modeled by
D(t) = min{(PD(t) + XD(t))+, κn+κc+κg}t∈[0, T ]
where the deterministic part
PD(t) = aD+bDt+
6
X
j=0
cjcos(2πϕjt+lj)t∈[0, T ] (9.2)
accounts for a linear demand increase superimposed by seasonal and weekly demand
fluctuations. The stochastic part (X(t))t∈[0,T]is modeled by an Ornstein-Uhlenbeck
process whose evolution follows the stochastic differential equation
dXD(t) = γD(αD−XD(t))dt +σDdW (t) (9.3)
driven by Brownian motion (W(t))t∈[0,T ]with parameters γD, αD, σD∈R. The
parameter estimation is based on historical load data for the time period 12/23/03-
12/23/06 available on ERCOT’s website, depicted in the Figure 9.1.1, the parameters
of the process were identified in two steps. First the deterministic harmonics (9.2) in
the demand process are identified with parameters obtained from peaks in the Fourier
transform. Secondly after removing the deterministic part (PD(t))t∈[0,T ](red line in
this figure) the residual component (XD(t))t∈[0,T ]is estimated by the standard linear
regression described in Subsection 9.1.3. The resulting parameters are:
Stochastic Part (XD(t))t∈[0,T ]
γDαDσD
102 0 819340
Market Designs for Emissions Trading Schemes 41
Deterministic Part (PD(t))t∈[0,T ]
aDbDi= 0 i= 1 i= 2 i= 3 i= 4 i= 5 i= 6
766381 38027 ci157526 71355 19047 15056 30608 16601 6364
ϕi1 2 3 4 52.14 104.29 208.57
li−3.56 −0.82 −0.83 −4.36 0.77 −1.55 −2.15
Notice from (ϕi)i=06 that long term periodicities where computed with a yearly pe-
riodic Fourier transform while short term periodicities were computed with a weekly
periodic fourier transform.
9.1.2. Fuel Switch Price Process. The continuous–time fuel switch price pro-
cess is modeled by
F(t) = aF+XF(t)t∈[0, T ]
where the stochastic part (XF(t))t∈[0,T ]is again modeled by an Ornstein-Uhlenbeck
process whose evolution follows the stochastic differential equation
dXF(t) = γF(αF−XF(t))dt +σFdW (t) (9.4)
driven by Brownian motion (W(t))t∈[0,T ]with parameters γF, αF, σF∈R.
For estimation of the parameters of the fuel spread we used Katy gas spot prices
(ICE Katy Exxon Plant Tailgate East Texas) and Platts coal prices (PRB 8400B
.35S Dly 1-Mo) from January 2004 to February 2007. Taking into account expected
long time gas and coal price we fixed a= 60$ neglecting the recent fuel switch price
increase. As for the electricity demand process the parameters of the stochastic com-
ponent (XF(t))t∈[0,T ]were calibrated using the procedure described in Subsection
9.1.3. The parameters of the Fuel Switch Price Process are:
Fuel Switch Price Process (F(t))t∈[0,T ]
aFγFαFσF
60 15.26 0 77.29
Figure 9.2.Scatter plot of (X(t∆), Y (t∆)) calculated by (9.5) based once historical fuel switch
prices for the ercot region. The straight lines depicts the respective estimated linear regressions.
42 R. CARMONA, M. FEHR, J. HINZ AND A. PORCHET
9.1.3. Linear Regression. The parameters γi, αi, σifor i∈ {D, F }of the
Ornstein-Uhlenbeck processes (9.3) and (9.4) are estimated by a standard linear re-
gression method applied as follows: From the formulas for conditional mean and
variance
E(X(t)|Fs) = X(s)e−γ(t−s)+α(1 −e−γ(t−s))s≤t
Var(X(t)|Fs) = σ2
2γ(1 −e−2γ(t−s))s≤t
we obtain the regression
Y(t∆) := X((t+ 1)∆) −X(t∆) = β0+β1X(t∆) + β2tt= 1, . . . , n −1 (9.5)
where (t)n−1
t=1 are independent, standard Gaussian random variables and β0, β1, β2
are connected to α, γ, σ by
α=−β0
β1
γ=−1
∆ln(1 + β1)
σ=r2γβ2
1−e−2γ∆.
REFERENCES
[1] B. Bueler. Solving an equilibrium model for trade of co2emission permits. European Journal
of Operational Research, 102(2):393–403, 1997.
[2] M. Cronshaw and J. B. Kruse. Regulated firms in pollution permit markets with banking.
Journal of Regulatory Economics, 9(2):179–89, 1996.
[3] J. H. Dales. Pollution, Property and Prices. University of Toronto Press, Toronto, 1968.
[4] M. Fehr and J. Hinz. A quantitative approach to carbon price risk modeling. Preprint, 2006.
[5] H. L. Feng and J. Zhao. Alternative intertemporal permit trading regimes with stochastic
abatement costs. Resource and Energy Economics, 28(1):24–40, 2006.
[6] P. L. Joskow, R. B. Schmalensee, and E. Bailey. The market for sulfur dioxide emissions. The
American Economic Review, 88(4):669–685, 1998.
[7] P. Leiby and J. Rubin. Intertemporal permit trading for the control of greenhouse gas emissions.
Environmental and Resource Economics, 19(3):229–256, 2001.
[8] J. Lemming. Financial risks for green electricity investors and producers in a tradable green
certificate market. Energy Policy, 31(1):21–32, 2003.
[9] W. D. Montgomery. Markets in licenses and efficient pollution control programs. Journal of
Economic Theory, 5(3):395–418, 1972.
[10] A.P. Robertson and W.S. Robertson. Topological vector spaces. Cambridge Univ. Press, 1973.
[11] J. Rubin. A model of intertemporal emission trading, banking and borrowing. Journal of
Environmental Economics and Management, 31(3):269–286, 1996.
[12] S. M. Schennach. The economics of pollution permit banking in the context of title iv of the
1990 clean air act amendments. Journal of Environmental Economics and Management,
40(3):189–21, 2000.
[13] J. Seifert, M. Uhrig-Homburg, and M. Wagner. Dynamic behavior of carbon spot prices .theory
and empirical evidence. Preprint, 2006.
[14] J. Sijm, K. Neuhoff, and Y. Chen. Co2 cost pass-through and windfall profits in the power.
Climate Policy, 6:4972, 2006.
[15] T. Tietenberg. Environmental Natural and Resource Economics. Addison Wesley, Boston,
2003.
[16] F. Wirl. Consequences of irreversibilities on optimal intertemporal co2emission policies under
uncertainty. Resource and Energy Economics, 28(2):105–123, 2006.
