A Forward Backward SDEs Approach to Pricing in

Carbon Markets 1st Edition Jean-François

Chassagneux pdf download
https://textbookfull.com/product/a-forward-backward-sdes-approach-to-pricing-in-carbon-markets-1st-
edition-jean-francois-chassagneux/

★★★★★ 4.9/5.0 (35 reviews) ✓ 128 downloads ■ TOP RATED

"Fantastic PDF quality, very satisfied with download!" - Emma W.

DOWNLOAD EBOOK
 A Forward Backward SDEs Approach to Pricing in Carbon
Markets 1st Edition Jean-François Chassagneux pdf download

TEXTBOOK EBOOK TEXTBOOK FULL

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

Financial Modelling with Forward-looking Information: An

Intuitive Approach to Asset Pricing 1st Edition Nadi
Serhan Ayd■n (Auth.)

The Short Run Approach to Long Run Equilibrium in

Competitive Markets A General Theory with Application to
Peak Load Pricing with Storage 1st Edition Anthony Horsley

Financial Decisions and Markets A Course in Asset Pricing

John Y Campbell

An Introduction to Financial Markets A Quantitative

Approach 1st Edition Paolo Brandimarte
Forward Lease Sukuk in Islamic Capital Markets: Structure
and Governing Rules Ahcene Lahsasna

Global Climate Change Policy and Carbon Markets:

Transition to a New Era 1st Edition Richard H. Rosenzweig
(Auth.)

The Pricing Compass: Finding the Solution to Your Pricing

Puzzle 1st Edition Yang

Stochastic Dominance Option Pricing: An Alternative

Approach to Option Market Research Stylianos Perrakis

The Human Rights Based Approach to Carbon Finance First

Paperback Edition Olawuyi
SPRINGER BRIEFS IN MATHEMATICS
OF PLANET EARTH  WEATHER, CLIMATE, OCEANS

Jean-François Chassagneux
Hinesh Chotai
Mirabelle Muûls

A Forward-Backward
SDEs Approach
to Pricing in
Carbon Markets
SpringerBriefs in Mathematics of
Planet Earth • Weather, Climate, Oceans

Managing Series Editors

D. Crisan, London, UK
D. Holm, London, UK

Series Editors
C. Cotter, London, UK
J. Broecker, Reading, UK
T. Shepherd, Reading, UK
S. Reich, Potsdam, Germany
V. Lucarini, Hamburg, Germany
SpringerBriefs present concise summaries of cutting-edge research and practical
applications across a wide spectrum of fields. Featuring compact volumes of 50
to 125 pages, the series covers a range of content from professional to academic.
Briefs are characterized by fast, global electronic dissemination, standard publish-
ing contracts, standardized manuscript preparation and formatting guidelines, and
expedited production schedules.

Typical topics might include:

• A timely report of state-of-the art techniques

• A bridge between new research results, as published in journal articles, and a

contextual literature review

• A snapshot of a hot or emerging topic

• An in-depth case study

SpringerBriefs in the Mathematics of Planet Earth showcase topics of current

relevance to the Mathematics of Planet Earth. Published titles will feature both
academic-inspired work and more practitioner-oriented material, with a focus on the
application of recent mathematical advances from the fields of Stochastic And
Deterministic Evolution Equations, Dynamical Systems, Data Assimilation,
Numerical Analysis, Probability and Statistics, Computational Methods to areas
such as climate prediction, numerical weather forecasting at global and regional
scales, multi-scale modelling of coupled ocean-atmosphere dynamics, adapta-
tion, mitigation and resilience to climate change, etc. This series is intended for
mathematicians and other scientists with interest in the Mathematics of Planet
Earth.

More information about this series at http://www.springer.com/series/15250

Jean-François Chassagneux
Hinesh Chotai Mirabelle Muûls
•

A Forward-Backward SDEs
Approach to Pricing
in Carbon Markets

123
Jean-François Chassagneux Mirabelle Muûls
U.F.R. de Mathématiques Grantham Institute
Université Paris Diderot, LPMA Imperial College
Paris London
France UK

Hinesh Chotai
Department of Mathematics
Imperial College
London
UK

SpringerBriefs in Mathematics of Planet Earth - Weather, Climate, Oceans

ISSN 2509-7326 ISSN 2509-7334 (electronic)
ISBN 978-3-319-63114-1 ISBN 978-3-319-63115-8 (eBook)
DOI 10.1007/978-3-319-63115-8
Library of Congress Control Number: 2017948227

Mathematics Subject Classification (2010): 60H30, 91G80

© The Author(s) 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer International Publishing AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents

1 A Description of the Carbon Markets and Their Role

in Climate Change Mitigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Why Do We Need Emissions Trading Markets? . . . . . . . . . . . . . . 1
1.1.1 The Science of Climate Change . . . . . . . . . . . . . . . . . . . . . 1
1.2 Policy Developments and the Paris Agreement . . . . . . . . . . . . . . . . 2
1.3 Economic Principles Underlying Emissions Trading
as a Policy Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.1 Tax Versus Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Trading Choices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 The European Union Emissions Trading System . . . . . . . . . . . . . . 6
1.4.1 Phases and Caps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.2 EUA Price Evolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Carbon Pricing and the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Introduction to Forward-Backward Stochastic Differential
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Backward Stochastic Differential Equations . . . . . . . . . . . . . . . . . . 12
2.1.1 Well-Posedness of BSDEs . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1.1 Linear BSDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.1.2 The Comparison Theorem . . . . . . . . . . . . . . . . . . 15
2.1.2 Application to Non-linear Pricing . . . . . . . . . . . . . . . . . . . . 16
2.1.2.1 Super-Replication in a Perfect Market . . . . . . . . . 17
2.1.2.2 A Non-linear Market. . . . . . . . . . . . . . . . . . . . . . . 18
2.1.3 Applications to Stochastic Control . . . . . . . . . . . . . . . . . . . 19
2.1.4 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.4.1 Constrained BSDEs . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.4.2 The Non-lipschitz Setting . . . . . . . . . . . . . . . . . . . 22
2.1.4.3 McKean–Vlasov FBSDEs . . . . . . . . . . . . . . . . . . . 23
2.2 Markovian BSDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

v
vi Contents

2.2.1 First Definition and Markov Property . . . . . . . . . . . . . . . . . 24

2.2.2 The Link with PDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Coupled Forward-Backward SDEs . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 The Pontryagin Approach to Stochastic Control
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.2 Well-Posedness of FBSDEs in Small Time Duration . . . . . 31
2.3.2.1 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . 32
2.3.2.2 The Decoupling Field and a Quasilinear PDE . . . . . 34
2.3.3 Existence and Uniqueness for Arbitrary Terminal
Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.3.1 Non-degenerate Diffusion Coefficient . . . . . . . . . . 38
2.3.3.2 FBSDEs with Singular Coefficients . . . . . . . . . . . 38
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3 A Mathematical Model for Carbon Emissions Markets . . . . . . . . . . . 43
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Market Set-Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 The Bid Stack, Emissions Stack and Emissions Rate . . . . . . . . . . . 45
3.4 Risk-Neutral Dynamics of Random Factors . . . . . . . . . . . . . . . . . . 51
3.5 The Single-Period Allowance Pricing FBSDE . . . . . . . . . . . . . . . . 52
3.6 Extension to a Multi-period Emissions Trading System . . . . . . . . . 53
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4 Numerical Approximation of FBSDEs . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1 Decoupled FBSDEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 A Markovian Iteration Scheme for Fully Coupled FBSDEs . . . . . . 64
4.3 Computation of Conditional Expectations Using Regression . . . . . 66
4.4 Numerical Examples of the Scheme’s Convergence . . . . . . . . . . . . 69
4.4.1 Bender and Zhang Test Model . . . . . . . . . . . . . . . . . . . . . . 69
4.4.2 Numerical Investigation of a Simple Singular FBSDE . . . . 71
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5 A Case Study of the UK Energy Market . . . . . . . . . . . . . . . . . . . . . . . 75
5.1 Introduction: An Explicit Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Specifying the Dynamics of the Market Factors . . . . . . . . . . . . . . . 76
5.2.1 Estimating the Seasonal Component . . . . . . . . . . . . . . . . . . 77
5.2.2 Fitting the Diffusion Processes . . . . . . . . . . . . . . . . . . . . . . 84
5.3 Estimating the Bid Stack and Emissions Rate . . . . . . . . . . . . . . . . 89
5.4 Simulation Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Chapter 1
A Description of the Carbon Markets
and Their Role in Climate Change Mitigation

1.1 Why Do We Need Emissions Trading Markets?

Climate change is a term widely used in the academic, policy and business spheres.
Here we briefly explain its significance by describing the current scientific under-
standing of the earth’s atmosphere and why its recent evolution is a source of concern.

1.1.1 The Science of Climate Change

The climate system is driven by energy brought to the planet by the sun’s radiation.
Some of the gases in the atmosphere, the so-called greenhouse gases, allow sunlight
to reach the surface of the earth, but trap the heat that is embedded in infrared radia-
tion. The main greenhouse gases (GHGs) are carbon dioxide (CO2 ) and water vapour
[9]. Other GHGs are nitrous oxide and methane. Thanks to this natural greenhouse
effect, the average surface temperature is 33 ◦ C warmer than what it would be without
it [2]. However, since the industrial revolution, human activity has been releasing
CO2 and other GHGs into the atmosphere through the burning of fossil fuels (such
as coal, oil and gas), manufacturing, agriculture and land use changes such as defor-
estation [9]. The Intergovernmental Panel on Climate Change (IPCC) has reported
that the atmospheric carbon dioxide levels increased in the last 150 years from 280
parts per million to 400 parts per million [5]. The IPCC, which includes more than
1300 scientists from around the globe, regularly produces reports based on scientific
peer-reviewed research on the causes, evidence and effects of climate change, on
mitigation and adaptation. In addition to the evidence on GHG concentrations, the
earth’s average temperature is increasing at a rate that is unmatched in the past 1300
years [5]. The IPCC [5] states that:
It is extremely likely that human influence has been the dominant cause of the observed
warming since the mid-20th century.

© The Author(s) 2017 1

J.-F. Chassagneux et al., A Forward-Backward SDEs Approach to Pricing
in Carbon Markets, SpringerBriefs in Mathematics of Planet Earth,
DOI 10.1007/978-3-319-63115-8_1
2 1 A Description of the Carbon Markets and Their Role in Climate Change Mitigation

Research shows that 97.2% of researchers agree with this assertion, according to
[3]. There is high confidence among scientists that temperatures will continue to rise
in the coming decades, with the effects of such warming likely to be different in
different regions of the world and at different times according to the IPCC, as the
different economic and environmental systems will react and adapt differently. Even
if some regions could benefit from increases in global mean temperature, the net
annual costs will increase over time, see [6], due to a long list of negative impacts
for humans, economies and ecosystems. More details are summarised here: https://
climate.nasa.gov/effects/. These conclusions lead to two necessary areas for action.
First, society will need to adapt in many dimensions so as to minimise the negative
consequences of climate change. Second, the economy needs to change so as to
ensure that GHG emissions are significantly reduced at a rate that is sufficient to
reduce the risk of dangerous climate change. Mitigation of GHGs requires technical
and behavioural change, innovation and effective policy. This brief focuses on one
particular policy instrument that seeks to lead to effective mitigation. The difficulty of
achieving meaningful and rapid mitigation is due to the so-called “public bad” nature
of GHG emissions: the cost of decreasing GHG emissions is borne by individuals
or countries who are unlikely to reap all the benefits this might lead to. Indeed, not
only will the reduced impacts benefit future generations, but also societies in many
countries, regardless of whether they mitigated strongly or not. This is why policy
makers and governments have been discussing the problem of climate change for
several decades.

1.2 Policy Developments and the Paris Agreement

Although the potential problem of GHG emissions and climate change had long
been put forward by scientists, it was only in 1992 that governments responded with
the creation of the United Nations Framework Convention on Climate Change. A
few years later, in December 1997, the UNFCCC established the Kyoto protocol,
an international agreement requiring worldwide reductions in emissions by 2012 of
about 5% on average compared with 1990 levels. Developed countries were each
allocated a target on emissions abatement while developing countries, including
emerging economies such as China, were given no targets. Although the protocol
came into force in 2005, it never met its objectives because the US never ratified it and
it did not include a more global effort agreement. As a consequence, countries started
negotiating through annual meetings, or Conference of Parties (COP), an agreement
to replace the Kyoto Protocol and achieve a more ambitious plan for dealing with
the risk of dangerous climate change. Despite a failed attempt at the Copenhagen
based COP, a momentous and significant agreement was achieved in Paris at the
21st COP in December 2015. 195 countries adopted a universal and legally binding
global climate deal to keep the increase in global average temperature to well below
2 ◦ C above pre-industrial levels, by ensuring that global emissions peak as soon as
1.2 Policy Developments and the Paris Agreement 3

possible and are then rapidly reduced. It came into force in November 2016, once
the threshold number of ratifications was achieved. Each country now has to
prepare, communicate and maintain successive nationally determined contributions (NDCs)
that it intends to achieve [and it will] pursue domestic mitigation measures, with the aim of
achieving the objectives of such contributions.

[11]. NDCs are the reductions of GHG emissions that each country agrees to cut,
as a contribution to the global mitigation levels needed to limit the risk of climate
change. As an example, the EU’s NDC proposes an overall 40% reduction in GHG
emissions by 2030 from 1990 levels. When adding all NDCs put forward so far around
the world, these proposal are unfortunately not consistent with limiting warming to
below 2 ◦ C. Even so, fixing and meeting these targets will involve different costs
and commitments by each country. As per the agreement, in order to achieve the
reductions agreed to in their NDCs, governments need to put in place a consistent
list of measures. Carbon markets are one of the domestic mitigation measures that can
be implemented nationally, regionally or internationally. The next section explains
its principles.

1.3 Economic Principles Underlying Emissions Trading

as a Policy Tool

Economics helps us to understand why emissions trading can be an efficient policy

tool for dealing with the externality1 created by the emissions of GHGs by a large
number of emitters. The first to put forward this idea was Ronald Coase in 1960
[1]. The main idea is that a regulator will set a collective target on total emissions,
or cap. Each emitter, whether a country or a company, is given a part of this total
in the form of emission rights. These are called “allowances”. For example, in the
case of carbon, one allowance means that the person that holds it is allowed, during
a specific time period, to emit one tonne of CO2 . If that target is set lower than what
the total emissions would have been without the cap, then the system yields emission
reductions.

1.3.1 Tax Versus Market

An alternative policy instrument to carbon markets that governments can use are
carbon taxes. Figure 1.1 illustrates how, with perfect information, a tax and a carbon
market are equivalent. A given country or jurisdiction will face damages per tonne

1 An externality refers to the consequence that an economic activity has on others that is not reflected

in market prices. In this context, emissions negatively affect future generations by causing climate
change, but emitters do not pay for the harm they generate and it is therefore not integrated into
prices.
4 1 A Description of the Carbon Markets and Their Role in Climate Change Mitigation

Fig. 1.1 Tax versus cap and trade

of emissions that can be assumed to be increasing in the total level of emissions:

damages are exponential as a function of emissions levels. We can therefore assume
that marginal damages are linearly increasing in emissions. This can also be seen
as the marginal benefits from abating emissions. On the other hand, the aggregation
of companies form that country’s marginal cost of abating emissions. Departing
from the “business as usual” point towards the left has a cost. Each extra ton of
abatement is more costly: for example, changing light bulbs is cheap, but changing
your production machine is costly. Aggregating this at the economy level yields the
downward sloping curve in Fig. 1.1. The point at which the regulator fixes the total
level of emissions that will maximise total welfare is that where the marginal cost of
abating emissions is equal to its marginal benefit. This is point A. If the regulator was
to move to the left, to a larger level of abatement, the cost of abating that extra ton of
carbon would be larger than the benefit it would bring to society. In order to reach
point A, the regulator can either distribute a certain number of permits, depicted
below the x-axis, so that any emissions made would need to be covered by a permit.
With an enforceable and binding penalty, emissions would therefore be at the optimal
level, corresponding to the cap of point A. Alternatively, the regulator can fix a tax
per ton of carbon emitted at the level of point A on the y-axis. Choosing a point to
the right of A would imply paying a tax per ton that would be larger than the cost of
reducing emissions by a ton, as shown by the Marginal Cost of Abating emissions
line. This would lead to choosing point A. Similarly, if emitting at levels to the left
of A, it would have been cheaper to pay the tax than abate by an additional ton. This
shows that whether with a tax or a cap (and trade), the equilibrium is at point A:
under perfect information and enforcement, the tax and the market are equivalent.
1.3 Economic Principles Underlying Emissions Trading as a Policy Tool 5

1.3.2 Trading Choices

The second important element is that market participants can trade their allowances.
There are likely to be some companies for which abating is very costly, and others
where there are “low-hanging fruits”: ways to reduce their emissions at zero or
negative cost. The carbon market means that those who can reduce at low cost will
do so, and therefore have an excess of allowances that they will be willing to sell
on the market at a given price. Participants will reduce their emissions up until the
point at which they are indifferent between buying one permit at the market price
and paying the cost of reducing emissions by one additional ton of carbon. This cost
is called the marginal abatement cost. This can be illustrated with a simple model of
the firm’s decision making process. The equation

Π1 (E 1 ) = a + bE 1 − E 12 − θ E 1 + A1 (1.1)

defines firm 1’s profit. The first linear term is increasing in emissions E 1 because
a higher production, which we assume will generate emissions, will increase its
revenue. The third term represents the cost of production, which we assume to be
quadratic in production levels which are directly correlated with emissions. Theta is
the carbon price and market participants will need to buy permits for their level of
emissions. Finally, A1 is the allocation of permits which is added as a lump sum to
the profit. When maximising profit, the firm chooses its emissions as a function of
its production parameters and the price of EUAs. The maximisation problem can be
written as
∂Π1 (E 1 )
= b − 2E 1 − θ = 0. (1.2)
∂ E1

If there are more producers on the market, they each face similar profit constraints.
In equilibrium, the total level of emissions, or cap, is determined by the regulator.
Assume there are only two market players. In this case,

E1 + E2 = E ∗, (1.3)

where E ∗ is the cap. In this case, if both players are similar except that the b that
enters the production function of company 1 will be c for company 2, then

b−θ c−θ
+ = E ∗, (1.4)
2 2
b+c
θ= − E ∗. (1.5)
2
Equation (1.5) determines the carbon price at a point where the market will be in
equilibrium and the marginal cost and marginal damages from emissions are equated.
6 1 A Description of the Carbon Markets and Their Role in Climate Change Mitigation

1.4 The European Union Emissions Trading System

From the signing of the Kyoto Protocol in 1996 the European Union took a strong
stance on climate policy and started running in 2005 a carbon market—the EU
Emissions Trading System (EU ETS)—governing the greenhouse gas emissions from
12,000 power and manufacturing plants in 31 countries. These account for around
45% of the EU’s greenhouse gas emissions or 5% of global emissions. The EU ETS
is administrated by the European Commission. The EU ETS Directive (2003/87/EC)
was further amended in 2009 by the amending EU ETS Directive (2009/29/EC).
These are detailed and explained in [12].

1.4.1 Phases and Caps

The EU ETS has been designed in three phases so far. Phase 1 was three years
long, from 2005 to 2008, and can be considered as a pilot period. Phase 2 ran from
2008–2012, with an increase in the number of countries and sectors covered. Phase
3 is scheduled to end in 2020 and negotiations are under way for the design of the
next phase. At the start of Phase 3, the upper limit on total emissions was set and
is now declining at a rate of 1.74% per year up until 2020 and 2.2% per year until
2030. As a result, EU emissions will be 43% less in 2030 than they were in 2005.
Improvements to the policy have been proposed by a variety of stakeholders, see
[7]. Phase 1 of the EU ETS was centred on power generation and energy-intensive
manufacturing industries, with an emphasis on CO2 emissions. All plants exceed-
ing 20MWh of energy use, including conventional power plants, had to participate.
Large emissions-intensive plants such as mineral oil refineries, coke ovens, iron and
steel, and factories producing cement, glass, lime, bricks, ceramics, and pulp and
paper were included. The following two phases extended the sectoral coverage to
include airlines, aluminium and ammonia manufacturing plants. Other greenhouse
gas emissions such as nitrous oxide and perfluorocarbons were also added to the
system. In practice, plants included in the EU ETS need to surrender one GHG per-
mit, known as an EU allowance (EUA), for each metric ton of CO2 (tCO2 e) emitted.
Permits can either be distributed to companies for free, or through an auction. By
imposing an EU-wide carbon price, the EU ETS sets for all participants in the market
the opportunity cost of emitting CO2 . As described above, if permits are scarce their
price increases and vice versa.
1.4 The European Union Emissions Trading System 7

Fig. 1.2 Evolution of the EU ETS prices between 2005 and 2016

1.4.2 EUA Price Evolutions

Looking back at the price on the EU ETS since its inception as in Fig. 1.2, there have
been considerable variations in it. The initial expectation was for the price of EUAs
to be in the range of e5−10/tCO2 e, and the trades in early 2005 on the new market is
consistent with this. The price rose at first. When a number of member states reported
their levels of emissions in April 2006, they were lower than expected. The price fell
to less than a Euro as it became increasingly apparent that Phase 1 emissions would
be lower than the cap [4].
The reasons for the price collapse are twofold. First, market participants were not
allowed to bank their permits from Phase 1 for use in Phase 2. This means that if
they had excess permits in Phase 1, they would be worth nothing in Phase 2 and they
would therefore be keen to sell them during Phase 1. Second, all plants had enough
permits to produce their desired level with no constraint, so that there was no demand
for additional permits. Phase 2 saw a modification of the rules, such that the price of
EUAs increased to over e20, reaching almost e30. The great recession in 2008 led
to a significant reduction of the EUA price to around e15. This price reduction, in
contrast to the 2005 movement, was not due to any flaws in the system’s allocation
of permits, but rather to the reduced economic activity and, hence, emissions due to
the recession. The price recovered in 2009, followed by a two-year period of stability
with a price close to e15. In the summer of 2011, the EUA price fell to e7–8 before
reaching around e4 at the start of Phase 3. An interesting observation is that, despite
concerns that it could reach zero, the price has stayed positive since. Comparing the
price of EUAs and the numbers of surplus permits at the ends of Phases 1 and 2,
as implied in Fig. 1.2, shows how important the possibility to bank allowances is.
8 1 A Description of the Carbon Markets and Their Role in Climate Change Mitigation

Looking at the evolution of the price in recent years also reflects how sensitive it is
to policy-related events rather than the underlying mechanisms. In the future, with
more stable policy and the possibility of linking different markets around the world,
the price is likely to reflect more clearly the economic elements behind it and display
features of a well-functioning and efficient market. In such conditions, an improved
knowledge of carbon markets and price formation is needed. This is the subject of
this brief.

1.5 Carbon Pricing and the Future

There is increasing evidence that the EU ETS led to significant emission reductions
and increases in clean innovation, and that this was not at the expense of a loss in
competitiveness. See [8] for a review of the evidence in the economics literature.
This is particularly interesting given that critics will argue that having more stringent
and unilateral climate change policies in the EU will lead to competitiveness losses
in global markets and no real impact on global emissions.
New carbon markets are being implemented worldwide: so far, 39 national and
23 sub-national jurisdictions have or are about to implement pricing instruments,
either cap-and-trade or carbon taxes. Together, such schemes are valued at around
$50 billion globally and represent close to 12% of global GHG emissions. This is a
perpetually evolving figure, and the latest numbers are regularly updated by the World
Bank. For example, in China, the world’s biggest GHG emitter, a national carbon
market is due to start by 2020. Pilots are already being implemented in seven different
pilot trading schemes: Chongqing, Shenzhen, Shanghai, Beijing, Guangdong, Hubei
and Tianjin. When the national market is started, there will be a significant increase
in the percentage of total global emissions covered by ETSs. Given the urgency of
the problem described at the start of this chapter, in the medium term one could
expect that such instruments will become increasingly common. As advocated by
many [10], a global market would lead to the most gains in efficiency. There are
many hurdles to pass before reaching such a result, and a global carbon market
was neither included nor discussed at the Paris COP21. A better understanding of
the pricing mechanisms of carbon permits is however needed, whether for local,
national or global markets, as it will ensure that both corporations and regulators can
act as rational market participants, so that it generates the highest efficiency gains
and emissions reductions. The complexity of determining the price of an emission
allowance on a carbon market, taking the EU ETS as an example, is presented and
approached in the following chapters.
References 9

References

1. Coase, Ronald H. 1960. The problem of social cost. The Journal of Law and Economics 3:
1–44.
2. Committee on Climate Change. 2010. The fourth carbon budget – reducing emissions through
the 2020s.
3. Cook, John, Naomi Oreskes, Peter T. Doran, William R.L. Anderegg, Bart Verheggen, Ed W.
Maibach, J. Stuart Carlton, Stephan Lewandowsky, Andrew G. Skuce, Sarah A. Green, Dana
Nuccitelli, Peter Jacobs, Mark Richardson, Bärbel Winkler, Rob Painting, and Ken Rice. 2016.
Consensus on consensus: a synthesis of consensus estimates on human-caused global warming.
Environmental Research Letters 11 (4): 048002.
4. Ellerman, A.Denny, and Barbara K. Buchner. 2008. Over-allocation or abatement? A prelimi-
nary analysis of the EU ETS based on the 2005–06 emissions data. Environmental and Resource
Economics 41 (2): 267–287.
5. IPCC. 2013. The physical science basis. Contribution of working group I to the fifth assessment
report of the intergovernmental panel on climate change.
6. IPCC. 2014. Summary for policymakers. In Climate change 2014: impacts, adaptation, and
vulnerability.
7. Martin, Ralf, Mirabelle Muûls, Laure B. de Preux, and Ulrich J. Wagner. 2014. On the empirical
content of carbon leakage criteria in the EU emissions trading scheme. Ecological Economics
105: 78–88.
8. Martin, Ralf, Mirabelle Muûls, and Ulrich J. Wagner. 2016. The impact of the European Union
Emissions Trading Scheme on regulated firms: What is the evidence after ten years? Review of
Environmental Economics and Policy 10 (1): 129–148.
9. Royal Society. 2010. Climate change: a summary of the science.
10. Stern, Nicholas. 2008. Key elements of a global deal on climate change. London, UK: London
School of Economics and Political Science.
11. United Nations. Framework convention on climate change. 2015. Adoption of the Paris agree-
ment. In 21st conference of the parties.
12. Woerdman, E. 2015. The EU greenhouse gas emissions trading scheme. In Essential EU climate
law. Cheltenham: Edward Elgar.
Chapter 2
Introduction to Forward-Backward
Stochastic Differential Equations

Forward-Backward Stochastic Differential Equations (FBSDEs) provide a powerful

modelling tool that has been intensively used in various areas of stochastic control
and, in particular, in mathematical finance. They were first introduced by Bismut
[3, 4] and then studied in a general way by Pardoux and Peng [41]. Since then,
FBSDEs have attracted a lot of interest.1 Although the basic theory is now well
understood, new questions or applications arise every day, making it a very active
field of research. Their link with a class of non-linear PDEs is also very fruitful and
has led to the design of probabilistic methods for solving such PDEs, as discussed
in the next chapter.
We present here a short self-contained introduction to FBSDEs which should be
enough to grasp the main concepts presented in the subsequent chapters of this brief.
Suggested lectures, on top of the main research articles, are [25, 38, 43].
Throughout this chapter, we let (Ω, A , P) be a complete probability space sup-
porting a d-dimensional Brownian Motion W . We shall denote by F = {Ft }t≥0 the
natural (augmented) filtration of W. Adaptedness and other measurability properties
of processes have to be understood with respect to F. When this is not the case (and
when it matters), it will be clearly pointed out in the text.

1 Pardoux and Peng counts more than 2000 citations as of March 2017.

© The Author(s) 2017 11

J.-F. Chassagneux et al., A Forward-Backward SDEs Approach to Pricing
in Carbon Markets, SpringerBriefs in Mathematics of Planet Earth,
DOI 10.1007/978-3-319-63115-8_2
12 2 Introduction to Forward-Backward Stochastic Differential Equations

2.1 Backward Stochastic Differential Equations

For a prescribed terminal time T > 0, the solution of a backward stochastic differ-
ential equation is a pair (Y, Z ) satisfying on [0, T ]

dYt = − f (t, Yt , Z t )dt + Z t dWt ,
YT = ξ,

for some progressively measurable random function f , called the driver, and a termi-
nal condition ξ which is a FT -measurable random variable. It is reasonable to assume
that a solution satisfies some conditions so that the various integrals appearing above
make sense, and this will be discussed below.
The first peculiarity of BSDEs is that contrary to (forward) SDEs, the solution
is not known at the initial time 0 but at the terminal time T . The second difference
with forward SDEs comes from the fact that the solution is a pair (Y, Z ). Before
giving some general existence and uniqueness results and stating precisely some
assumptions on the coefficients, we will comment on the shape of the equation and
give some hints on the extra process Z .
The simplest example is f ≡ 0 and ξ ∈ L 2 (FT ), where for t ∈ [0, T ], L 2 (Ft )
stands for the set of square integrable Ft measurable random variables. Then, the
natural solution to the differential equation dY
dt
t
= 0 and YT = ξ is Yt = ξ , which is
generally not adapted (unless ξ is deterministic). The best approximation—say in
L 2 —is given by the martingale Yt = E[ξ | Ft ]. Using the martingale representation
theorem, we introduce a Z -process which is square integrable
 t
Yt = E[ξ | Ft ] = E[ξ ] + Z s dWs ,
0

leading to
 T
Yt = ξ − Z s dWs , i.e. − dYt = −Z t dWt , with YT = ξ.
t

We observe in this example that the role of Z is to guarantee the adaptedness of Y . In

particular, a deterministic terminal condition will lead to Z ≡ 0. This basic example
can already be linked to pricing in complete financial markets: The Y represents the
price of a contingent claim with random terminal payoff ξ and the Z is linked to the
replication portfolio. We discuss this financial application in detail in Sect. 2.1.2.
We now present the basic well-posedness results for BSDEs.
2.1 Backward Stochastic Differential Equations 13

2.1.1 Well-Posedness of BSDEs

Here we present results in the Lipschitz setting, which were first studied in [41],
see also [25]. This setting allows us to present the theory in a quite advanced and
useful form without encountering too many complications. Moreover, this is the main
framework generally adopted for numerical studies.
In order to state precisely the main existence and uniqueness result for BSDEs in
the Lipschitz framework, we have to introduce some notation and assumptions.
• We denote by S 2 (Rk ) the vector space of RCLL2 adapted processes Y , with values
in Rk , and such that:
 
Y S 2 := E sup |Yt | < ∞,
2 2
0≤t≤T

and Sc2 (Rk ) is the subspace of continuous processes.

• The set H 2 (Rk×d ) is the set of Rk×d -valued progressively measurable Z -processes
such that
 T 
Z H 2 := E
2
|Z t | dt < ∞,
2
0

where for z ∈ Rk×d , |z|2 = Tr(zz † ).

We shall often omit Rk and Rk×d ; the spaces S 2 , Sc2 and H 2 are Banach spaces.
A random Rk -valued function f defined on [0, T ] × Ω × Rk × Rk×d is such that
for all (y, z) ∈ Rk × Rk×d , the process { f (t, y, z)}0≤t≤T is progressively measurable.
We also assume that
(H 1): There exists a positive constant L such that P a.s.:
1. Lipschitz continuity in (y, z): for all t, y, y  , z, z  ,
   
 f (t, y, z) − f (t, y  , z  ) ≤ L |y − y  | + z − z   ;

2. Integrability condition:
  T 
E |ξ |2 + | f (r, 0, 0)|2 dr < ∞.
0

Theorem 2.1 Under (H 1), there exists a unique solution (Y, Z ) ∈ Sc2 × H 2 to
 T  T
Yt = ξ + f (s, Ys , Z s )ds − Z s dWs , 0 ≤ t ≤ T. (2.1)
t t

2 Right Continuous with Left Limits.

14 2 Introduction to Forward-Backward Stochastic Differential Equations

We skip the proof of the above theorem, which is based on a contraction mapping
argument, see the proof of Theorem 2.1 in [25]. We will give a proof, using essentially
the same arguments, for an existence and uniqueness result in a (slightly) more
involved setting below, see Theorem 2.6.

2.1.1.1 Linear BSDEs

We first study linear BSDEs for which we can give an almost explicit solution. For
this section, we set k = 1: Y is then real-valued and Z a d-dimensional row vector.

Proposition 2.1 Let {(at , bt )}t∈[0,T ] be progressively measurable and bounded

processes with values in R × Rd . Let {ct }t∈[0,T ] be an element of H 2 (R) and ξ
a square integrable FT -measurable random variable.
The linear BSDE
 T  T
Yt = ξ + {ar Yr + Z r br + cr } dr − Z r dWr (2.2)
t t

has a unique solution given by:

  T  

∀t ∈ [0, T ], Yt = Γt−1 E ξ ΓT + cr Γr dr  Ft , (2.3)
t

where for all t ∈ [0, T ],

 t  t  t
1
Γt = exp br dWr − |br |2 dr + ar dr .
0 2 0 0

Proof We first write down the dynamics of Γ :

dΓt = Γt (at dt + bt dWt ) , Γ0 = 1.

Using Doob’s inequality, we easily see that Γ ∈ Sc2 , as b is bounded. It is also clear
that there is a unique solution to (2.1): define f (t, y, z) = at y + zbt + ct , which
obviously satisfies (H 1), and we know that Y ∈ Sc2 .
Using the product formula, we compute

d(Γt Yt ) = Γt dYt + Yt dΓt + d Γ, Y t = −Γt ct dt + Γt Z t dWt + Γt Yt bt dWt ,

2.1 Backward Stochastic Differential Equations 15

t
showing that Γt Yt + 0 cr Γr dr is a local martingale, which is, in fact, a martingale
as c ∈ H 2 and Γ , Y are in S 2 . Then,
 t   T  

Γt Yt + cr Γr dr = E ΓT YT + cr Γr dr  Ft ,
0 0

which concludes the proof.

Linear BSDEs play an important role in the theory as outlined by the results
below, especially via the Comparison Theorem. They were also the first BSDEs
to be introduced by Bismut in [3] to study the quadratic-linear stochastic control
problem. They appear there as the adjoint process in the variational characterisation
of an optimal control.

Remark 2.1 Observe that ξ ≥ 0 and ct ≥ 0 leads to Yt ≥ 0.

2.1.1.2 The Comparison Theorem

This section presents the “comparison theorem”, which allows us to compare two
solutions of two BSDEs (in R) as soon as we can compare the terminal conditions
and the drivers of the BSDEs.
Theorem 2.2 Let k = 1 and assume that (ξ  , f  ) satisfies (H 1), the solution to
the associated BSDE is denoted (Y  , Z  ). Let (Y, Z ) be a solution of a BSDE with
T
parameters (ξ, f ) and satisfying 0 f (t, Yt , Z t )dt ∈ L 2 (FT ). We also assume that
P a.s. ξ ≤ ξ and f (t, Yt , Z t ) ≤ f  (t, Yt , Z t ) λ ⊗ P-a.e. (λ denoting the Lebesgue


measure). Then,

P a.s., ∀t ∈ [0, T ], Yt ≤ Yt .

If, moreover, Y0 = Y0 , then P a.s., Yt = Yt , 0 ≤ t ≤T and f (t, Yt , Z t ) =

f (t, Yt , Z t ) λ ⊗ P-a.e. In particular, as soon as P ξ < ξ  > 0 or f (t, Yt , Z t ) <


f  (t, Yt , Z t ) on a set with positive λ ⊗ P-measure then Y0 < Y0 .

Proof The proof uses a linearisation argument. Defining U = Y  − Y ; V = Z  − Z

and ζ = ξ  − ξ , we have
 
T   T
Ut = ζ + f  (r, Yr , Z r ) − f (r, Yr , Z r ) dr − Vr dWr .
t t

We observe that

f  (r, Yr , Z r ) − f (r, Yr , Z r ) = f  (r, Yr , Z r ) − f  (r, Yr , Z r ) + f  (r, Yr , Z r ) − f  (r, Yr , Z r )

+ f  (r, Yr , Z r ) − f (r, Yr , Z r ) (non-negative).
16 2 Introduction to Forward-Backward Stochastic Differential Equations

We introduce a and b: a is R-valued and b a d-dimensional vector. We set

f  (r, Yr , Z r ) − f  (r, Yr , Z r )

ar := 1{Ur =0} .
Ur

For 0 ≤ i ≤ d, we consider the vector Z r(i) whose last d − i components are those
of Z r and the first i components are those of Z r . For 1 ≤ i ≤ d, we set
   
f  r, Yr , Z r(i−1) − f  r, Yr , Z r(i)
bri = 1{Vri =0} .
Vri

Importantly, as f  is Lipschitz, the two processes are bounded and progressively

measurable. We then observe that
 T  T
Ut = ζ + (ar Ur + Vr br + cr ) dr − Vr dWr ,
t t

where cr = f  (r, Yr , Z r ) − f (r, Yr , Z r ). By assumption, we have ζ ≥ 0 and cr ≥ 0.

Using the formula given in Proposition 2.1, we have, for all t ∈ [0, T ],
  T  

Ut = Γt −1 E ζ ΓT + cr Γr dr  Ft ,
t

with, for 0 ≤ r ≤ T ,
 r  r  r
1
Γr = exp bu dWu − |bu |2 du + au du .
0 2 0 0

Following Remark 2.1, we get that Ut ≥ 0, which proves the first statement of the
theorem.
Moreover, if U0 = 0 then we have
  T 
0 = E ζ ΓT + cr Γr dr ,
0

and the random variable is non-negative. Then, it is equal to zero P a.s., which implies
ζ = 0 and cr = 0, concluding the proof of the theorem.

2.1.2 Application to Non-linear Pricing

In this section, we study an application of BSDEs in Mathematical Finance, namely

the pricing of European contingent claims. We first present the framework of the
linear pricing rule in a perfect market and the corresponding linear BSDE. We then
2.1 Backward Stochastic Differential Equations 17

introduce some imperfection in the market and show that the option price is still
given by a BSDE but with a non-linear Lipschitz driver [25].

2.1.2.1 Super-Replication in a Perfect Market

The market, in its simplest setting, is consists of two assets: a non-risky asset (bank
account) delivering an interest rate r , which is a deterministic quantity, and a risky
asset (a stock) whose price at any time t is given by St . The stochastic process S has
the following Black–Scholes type dynamics:
 t  t
St = S0 + r Ss ds + Ss σs dWs , S0 ∈ (0, ∞).
0 0

The random coefficient σ is essentially bounded and satisfies σs ≥ ε > 0 for all
s ∈ [0, T ]. The fact that the drift is r S implies that the dynamics of asset price is
already written under the risk neutral probability.
We study the price of a contingent claim that has maturity T and random payoff
ξ , with ξ belonging to L 2 (FT ). The goal is to construct an asset portfolio that will
perfectly replicate the random payoff ξ . In our setting, a portfolio is described by a
stochastic process (α, φ) where
• α is the amount of money in the bank account;
• φ is the amount invested in the risky asset.
At time t, its value is given by

Vt = αt + φt . (2.4)

On an infinitesimal time interval dt, the variation in value of the bank account is
given by αt r dt and the variation in value due to the risky asset is given by φStt dSt (due
to price change). The stochastic process (α, β) is a strategy that controls the value of
the portfolio, but not all strategies can be used. For modelling purposes, one restricts
the set of strategies to self-financing strategies, i.e. strategies such that the change in
value of the portfolio is given by

φt
dVt = r αt dt + dSt . (2.5)
St

In other words, the change in value of the portfolio is only due to a change in value
of the assets. We then compute, using (2.4), that
 t  t
Vt = V0 + r Vs ds + φs σs dWs .
0 0
18 2 Introduction to Forward-Backward Stochastic Differential Equations

We observe that the value of V only depends on φ and V0 . We also need to impose
t
some technical conditions on φ and we will assume that E 0 |φs |2 ds < ∞ so that
the stochastic integral is a martingale.
The super-replication problem is to find a strategy that will hedge the terminal
payoff with the minimal initial cost
v,φ
p := inf G0 with G0 = {v ∈ R|∃φ ∈ H 2 , VT ≥g(ST )} . (2.6)

Proposition 2.2 Consider (Y , Z ), the solution to the following linear BSDE

 T  T
Yt = ξ − r Ys ds − Zs dWs . (2.7)
t t

Then the super-replication price is a replication price and is given by p := Y0 . The

replication strategy is φ ∗ = Zσ .

Proof 1. The existence and uniqueness of the solution to (2.7) has already been
Y ,φ ∗
discussed above. It is straightforwardly seen that Yt = Vt 0 for all t ∈ [0, T ]
Y ,φ ∗
and then that VT 0 = ξ . This proves that Y0 ≥ p and that G0 is non-empty.
v,φ
2. Now, let v ∈ G0 and φ ∈ H 2 such that VT ≥ ξ . A simple application of Itô’s
−r t v,φ
Formula shows that (e Vt )t∈[0,T ] is a martingale. In particular, we have that
 
v = E e−r T VT ≥ E e−r T ξ = Y0 ,
v,φ

the last equality coming from (2.3). This yields that for all v ∈ G0 , v ≥ Y0 . The
proof is then concluded by taking the infimum on G0 .
 
Remark 2.2 The price at any date t ∈ [0, T ] is given by Yt = E er (T −t) ξ | Ft .

2.1.2.2 A Non-linear Market

We now consider a case of market imperfection: We work with two different rates
for borrowing (R) and lending (r) with R > r .
We want to price a European contingent claim in this market following a hedging
strategy. The main difference now is that the cash dynamics is given by

dαt = r αt 1{αt ≥0} dt + Rαt 1{αt <0} dt

= (r αt + (r − R)[αt ]− )dt . (2.8)
2.1 Backward Stochastic Differential Equations 19

The self-financing condition (2.5) rewrites in this context as

dVt = (r αt + (r − R)[αt ]− )dt + φt σt dWt .

Recalling that Vt = αt + φt , we then compute

dVt = {r Vt + (r − R)[Vt − φt ]− } dt + φt σt dWt .

The super hedging problem is still given by (2.6), only the dynamics of V has
changed.

Theorem 2.3 Let (Y, Z ) be the solution to the following non-linear BSDE
 T    T
Zt
Ys = ξ − r Yt + (r − R)[Yt − ]− dt − Z t dWt . (2.9)
s σt s

Then the super-replication price is a replication price and is given by p := Y0 .

The replication strategy is φ ∗ = σZ .
Y ,φ ∗
Proof 1. From the dynamics of V , we observe that Yt = Vt 0 for all t ∈ [0, T ]
Y ,φ ∗
and then that VT 0 = ξ . This proves that Y0 ≥ p and that G0 is non-empty.
v,φ v,φ
2. Now, let v ∈ G0 and φ ∈ H 2 such that VT ≥ ξ . We can interpret (Vt )t∈[0,T ]
v,φ
as a BSDE with terminal condition VT and the same driver as (2.9). Then a
v,φ
direct application of the Comparison Theorem 2.2 leads to v = V0 ≥ Y0 . The
proof is concluded by taking the infimum on G0 .

The link between mathematical finance and BSDE theory is very fruitful, see e.g.
[25]. For example, recently, Crepey [16, 17] has studied counterparty risk in the
framework of BSDEs.

2.1.3 Applications to Stochastic Control

We illustrate in this section how BSDEs can be used to solve stochastic control prob-
lems. We present here a direct approach and we refer to Sect. 2.3.1 for a variational
approach related to the stochastic maximum principle.
Let us consider an R-valued process X which is given under P as the solution of
the following differential equation:

d X t = σ (X t )dWt , (2.10)

where σ : R → R is a Lipschitz function satisfying Λ ≥ σ (x) ≥ 1

Λ
for some Λ > 0
and W is a one-dimensional standard Brownian Motion.
20 2 Introduction to Forward-Backward Stochastic Differential Equations

We denote by U the set of progressively measurable processes α with values in

a compact interval U . For a given α ∈ U ⊂ R, we define
 t
Wtα = Wt − σ −1 (X t )b(X t , αt )dt ,
0

where b : R2 → R is a Lipschitz function with bounded support. Applying Gir-

sanov’s Theorem, we thus have that W α is a Brownian Motion under a new probability
Pα (which is absolutely continuous with respect to P). The dynamics of X under Pα
reads as

d X t = b(X t , αt )dt + σ (X t )dWtα . (2.11)

The control problem is classically given by the following optimisation

  T 
Pα
min J (α) with J (α) := E g(X T ) + h(X s , αs )ds , (2.12)
α∈U 0

where g : R → R and h : R2 → R are Lipschitz functions.

Let us assume that
a ∗ (x, z) := argmina∈U H (x, z, a) where H (x, z, a) := h(x, a) + zσ −1 (x)b(x, a) (2.13)

is well defined as a Lipschitz continuous function of (x, z). Let us then introduce

H ∗ (x, z) := h(x, a ∗ (x, z)) + b(x, a ∗ (x, z))z . (2.14)

The function H above is called the Hamiltonian of the system, and H ∗ is its optimal
value. We then have the following result.
Theorem 2.4 In the above setting, the control problem (2.12) has a solution Y0∗
given by the initial value of the following BSDE
 T  T
Yt∗ = g(X T ) + H ∗ (X s , Z s∗ )ds − Z s∗ dWs , (2.15)
t t

and an optimal control is αt∗ = a ∗ (X t , Z t∗ ), t ∈ [0, T ].

Proof In this restrictive setting, H ∗ is Lipschitz continuous and (2.15) has a unique
solution. For α ∈ U , we consider the solution (Y α , Z α ) of the following BSDE
 T  T
Ytα = g(X T ) + h(X s , αs )ds − Z sα dWsα , (2.16)
t t

observing that Y0α = J (α). Then, rewriting the above dynamics under P, we obtain
2.1 Backward Stochastic Differential Equations 21
 T  T
Ytα = g(X T ) + H (X s , Z sα , αs )ds − Z sα dWs . (2.17)
t t

We now use the Comparison Theorem 2.2, recalling the definition of H ∗ in (2.14),
to obtain that
∗
Y0α ≥ Y0∗ = Y α ,

which concludes the proof.

This—by now classical—approach to solving stochastic control problems using

BSDEs can be extended to various settings, in particular for non-zero sum games
[28], where the existence of the representing BSDE is quite difficult to obtain.

2.1.4 Extensions

The theory of BSDEs is rich and powerful. It has attracted a lot of interest in the
past 25 years. In this section, we report briefly on some extensions to the Lipschitz
setting and the basic shape of Eq. (2.1). Note that we still present the case of Brownian
filtrations but, of course, BSDEs have been studied in relation to jump processes as
well, see e.g. [2]. Nor are we going to delve into the study of second-order BSDEs
[48]. As already remarked, BSDE theory is still an active field of research and we
do not aim to be exhaustive in the list we give below.

2.1.4.1 Constrained BSDEs

For modelling purposes, the processes Y and Z sometimes need to be constrained to

belong to some possibly random sets. Generally, Eq. (2.1) no longer holds and one
has to add a finite variation process as part of the solution, which then reads as
 T  T  T
Yt = ξ + f (s, Ys , Z s )ds − Z s dWs + d K s , 0 ≤ t ≤ T. (2.18)
t t t

The solution is now a triple (Y, Z , K ) and, obviously, some other conditions are
needed to guarantee uniqueness, depending on the applications.

Reflected BSDEs (RBSDEs)

In the one-dimensional setting, RBSDEs are linked, in their simplest form, to optimal
stopping problems and the pricing of American options in non-linear markets [24].
If the exercise price of the option is given by a process (L t )0≤t≤T , then (Yt )0≤t≤T ,
22 2 Introduction to Forward-Backward Stochastic Differential Equations

representing the option price, has to satisfy Yt ≥ L t for all 0 ≤ t ≤ T . Thus, Y is

forced to belong to the random set [L , ∞). The process K , which forces Y above
S, is a continuous increasing process in this setting. Uniqueness for K is obtained
T
thanks to the condition 0 (Yt − L t )d K t = 0: This simply states that K is active only
when Y touches the boundary L.
It is also possible to add an upper obstacle U for Y leading to doubly reflected
BSDEs [18]. In this case, Y is forced to belong to [L , U ] and the main applications
are Dynkin games and the pricing of Game options, which are callable American
Options, e.g. convertible bonds.
More generally, in the multi-dimensional setting, Y can be constrained to a closed
convex domain D, possibly random. The question is then the direction of reflection
at the boundary of the convex domain. The case of normal reflection is treated in full
generality in [27]. The case of oblique reflection is more involved, see [14] and the
references therein for an account of RBSDEs linked to optimal switching problems.

BSDEs with Constraints on Z

BSDEs with constraints on the Z -process have been introduced in [19]. The minimal
solution to (2.18) is found such that Z ∈ D. In this case, BSDEs are linked to the
pricing of European Options when some investment constraints are present on the
market.

2.1.4.2 The Non-lipschitz Setting

The Lipschitz setting has been extended in various ways, but then existence and
uniqueness results are much more difficult to obtain, when available. The first exten-
sion concerns coefficients with the monotonic property in y only, see [20], where
the case of random terminal time is also treated. Let us mention the application to
stochastic homogenisation [40]. The notion of generalised BSDEs has been intro-
duced in [45], where the driver involves integration with respect to a finite variation
process. This allows us to represent solutions of non-linear PDEs with generalised
Neumann boundary condition. See Sect. 2.2.2 for an account of the link between
PDEs and FBSDEs.
An important generalisation for applications comes from the introduction of
quadratic growth in the component Z . This has been introduced for utility max-
imisation problems in [30, 47] and recently to principal-agent problems [23]. The
one-dimensional case (for Y ) is now well understood, see e.g. [30, 32]. The article
[49] is the first to give an existence and uniqueness result for multi-dimensional
BSDEs with quadratic growth. The general case is known to be difficult [26] albeit
with recent progress [29, 50].
2.1 Backward Stochastic Differential Equations 23

BSDEs with only continuous coefficients but with linear growth are considered
in a multi-dimensional setting in [28] with an important application to non-zero sum
games.

2.1.4.3 McKean–Vlasov FBSDEs

Recently, BSDEs have been introduced to study large population stochastic control
problems. These are control problems of the type (2.11) and (2.12) but involving
many interacting agents. A classical example is the following. Consider N agents,
whose personal state is given by (for player i)

d X ti = b(X ti , μnt , αti )dt + σ dWti ,


(W i ) are independent Brownian motions, μnt = n1 i δ X ti is the statistical distribution
of the system, and α i the control of the player. The cost to minimise for each player
(given here for player i) is
  T 
J i (α) = E g(X Ti , μnT ) + f (t, X ti , μnt , αti )dt .
0

Note that the players interact via μn only. Games with a large number of player are
difficult to solve. The hope here is to obtain an asymptotic (N → ∞) description of
the equilibrium, hopefully “easier” to handle.
The notion of equilibrium is then fundamental as it yields to different limiting
equations. Individualistic, i.e. Nash-like, equilibria were first considered by Lasry
and Lions [35], who coined them Mean Field Games. They were introduced at
the same time by Caines, Huang and Malhamé [31]. Cooperative equilibrium leads
to the control of McKean–Vlasov SDEs [11]. For a comparison between the two
approaches, we mention [12]. The probabilistic approach to studying such problems
has been developed by Carmona and Delarue [6, 10, 11], see the references therein
for early works, and leads to the study of the following system
 t t
X t = ξ + 0 b(X s , Ys , Z s , P(X s ,Ys ,) )ds + 0 σ (X s , Ys , P(X s ,Ys ,) )dWs ,
T T
Yt = g(X T , P X T ) + t f (X s , Ys , Z s , P(X s ,Ys ,) )ds − t Z s dWs .

The main peculiarity, besides the coupling between the two equations, is the fact
that the coefficients depend upon the law of the solution. These McKean–Vlasov
FBSDEs have been linked to some non-linear PDEs written on the Wassertein space
[8, 13]. Let us mention finally that the very difficult question of the convergence of
the controlled particle system to the mean-field limit has been studied recently in [5].
down

seen

entrance of successfully

as

Perhaps approach

all the

surface
within of of

leading the every

the is heard

grows

is

on the

snowy your

prominently
their that

Church

of brethren

Belgian welcome the

its duties that

Great government named

age three

possesses doubt were

way

great nimirum are

few did argument

dollars

writings

this to French

greedy travellmg form

it Ven for

quote and House

also a or

Paris difference

Catholic

per

Catholics as

early

that

saying

partisanship Books frequent

This

in

dry healing disappears

different

is and

biography and Of

made among

There
com conscientious princes

is natures

industry the

with

then rpHE of

blood one

pendant the

lost facientibus

this Long

place publication the

island

the occasion grave

by Berlin He

to

300

chaos by

scholars based There

by an
to

was

these and deputy

of whorls p

we He

which

two but s

existing what

et the and

its the
with

to of

his late of

no fuel

gradually petroleum

curiosity rich

large

be
called

an

the days

kneel

nired language ceremonies

the

his Mr

much desiderium
a

florens

far which

three in

and

antiquity Irish He

succeeding time influence

seems

St to
Eandolph

Renier or a

he

we the

hut

and here reaction

the perhaps 182

and Usum

the The

important
interest improbability

an

height the

nations

a solid

more

large meditated

will was

such form

lead eyes a
and

Germany literary

time shifting a

other sky be

thou If
and

all Archive

State expenditure

country

of reproof not

complete La festivities

nauseating
or in

they propers

Lord The

recommendation

sisters

of nation

looking introduced anno

found article

to

of
Imperial books that

he commodo centena

he cannot ruin

what The in

the the

layers inscribed

diary

the him
as of of

observations they had

friend the

in in the

jpaternis

home

oil distance discusses

died each Voji

in the
memory

not

but

above singleness the

and
straight

by not all

with his

came saline agree

of
the cease or

potentia This of

at barely

is prized hands

Jesus is

any extraordinary
this suggest Popes

of

Africa see

serious was

doirmatic when

through men

every countrymen

professional quote Viuls

that his
error new

rescue life

delegates

in discussed

into only record

nightstand

priest securely

1884
him and

that

cause book

might bring that

below the

and

the complications

as out

to Alternately England

anxiously the
An Had in

it from

of

defeat in the

that sen xlvi

the is

of he

He

quemadmodum
relies

f affairs

known taking

death also very

tunnels their of
fortresses

say solid

marched that of

although of

ground a

the institutionis
what let Under

Parish carried

early action 61

measure following

as only Successive

aim

fleet letter

chief the

by of were
thing of merchant

the are loosely

stadia or

M which will

Fouard developed or

are been and

residues

in butterfly of
subject its

Prench to question

many d

a object

course

does the

with

and worse sense

THIS mass better

about connect

in

Virgin use did

this

the I s

Donnelly belief

whom by this

Lucas the in

he etude seizes
to St find

interfere

century the

reviewer

of

committing

room from spiderlings

purity safety Dr

the may rain

all
be kept

too

by wisdom natura

Vol conscience

the
Prussian the associated

each

200 hall

relied

Thanks

are ever

against

in Book W
and vindicate human

desire which

each

to puts

to Of

once

be martyr a
account population as

never usefulness of

even obtains

and long

beatorum
in

there

of

the the the

interest
use

injustice much

the hard character

against

every to

set

is
as

doctors non promise

oasis it in

journeys mostly

in of was
gallon

set should the

had not on

to Dr
the the reward

returns who

Barbarossa bystander

of g in

no
est reipublicae troubles

into

to they

filled them

while either

of Lanigan litteris

a suis

visit minor ground

consider natives
in she the

animated More all

does aspirations

Nothing www

Tablet

the without

us the

eyes monastery
what description

in So not

days remedy

examine that an

A revel of

in

and the 80
che would Hort

in a

could

word the In

the

and

Dei of no

The discussed

that

large
to

American

of deadly

that receive by

not new procure

as ledges property

disabled But once

broken five prepare

beyond

of

s the exceedingly

Dioeceses America

day a restricting

500

attended
or may be

remarked most of

the

by

and

now

age

been
this

elected

having part

Malabaricam

day binds

arrived mental Francisco

reader of of

let

precept

As
the

re he

von is consulted

had we

not

man in
of in

collateral

island a

Letter

Not

such from

which easier

beings difficult

has to
Latin

from to Socialism

vessel Spirestone

aliena novis not

2 making plea

penman hill
a

to

branches preachers for

the factory the

after

is the

a those that

rush of

no upon forests
are

knees that to

etiam

farmers that namely

glowing

On Liberty is

slay
received

but the

the for

contains by

home through she

crop

is be
not them that

through is

is

collection sizes Idol

had cases

inhabitants creation

the to readable

many a

At struggling is

produce Great
with and Annual

in Herodotus barrels

It Catholics into

absolute for Father

renunciation

the judges

the Mr
enthusiasm readings though

they it not

beautiful and the

Quotidianis Leading

out

dedicated yet
encroaching nothing

round eventually 261

thabur hear what

which be time

more and Flaherty

source of

been this means

he

century

the Unfortunately the

the fruatur

Lao frost great