Stochastic

Finance
A Numeraire Approach

K10632_FM.indd 1 11/30/10 1:56 PM

 CHAPMAN & HALL/CRC
Financial Mathematics Series

Aims and scope:

The field of financial mathematics forms an ever-expanding slice of the financial sector. This series
aims to capture new developments and summarize what is known over the whole spectrum of this
field. It will include a broad range of textbooks, reference works and handbooks that are meant to
appeal to both academics and practitioners. The inclusion of numerical code and concrete real-
world examples is highly encouraged.

Series Editors
M.A.H. Dempster Dilip B. Madan Rama Cont
Centre for Financial Research Robert H. Smith School Center for Financial
Department of Pure of Business Engineering
Mathematics and Statistics University of Maryland Columbia University
University of Cambridge New York

Published Titles
American-Style Derivatives; Valuation and Computation, Jerome Detemple
Analysis, Geometry, and Modeling in Finance: Advanced Methods in Option Pricing,
Pierre Henry-Labordère
Credit Risk: Models, Derivatives, and Management, Niklas Wagner
Engineering BGM, Alan Brace
Financial Modelling with Jump Processes, Rama Cont and Peter Tankov
Interest Rate Modeling: Theory and Practice, Lixin Wu
Introduction to Credit Risk Modeling, Second Edition, Christian Bluhm, Ludger Overbeck, and
Christoph Wagner
Introduction to Stochastic Calculus Applied to Finance, Second Edition,
Damien Lamberton and Bernard Lapeyre
Monte Carlo Methods and Models in Finance and Insurance, Ralf Korn, Elke Korn,
and Gerald Kroisandt
Numerical Methods for Finance, John A. D. Appleby, David C. Edelman, and John J. H. Miller
Portfolio Optimization and Performance Analysis, Jean-Luc Prigent
Quantitative Fund Management, M. A. H. Dempster, Georg Pflug, and Gautam Mitra
Robust Libor Modelling and Pricing of Derivative Products, John Schoenmakers
Stochastic Finance: A Numeraire Approach, Jan Vecer
Stochastic Financial Models, Douglas Kennedy
Structured Credit Portfolio Analysis, Baskets & CDOs, Christian Bluhm and Ludger Overbeck
Understanding Risk: The Theory and Practice of Financial Risk Management, David Murphy
Unravelling the Credit Crunch, David Murphy

Proposals for the series should be submitted to one of the series editors above or directly to:
CRC Press, Taylor & Francis Group
4th, Floor, Albert House
1-4 Singer Street
London EC2A 4BQ
UK

K10632_FM.indd 2 11/30/10 1:56 PM

 Stochastic
Finance
A Numeraire Approach

Jan Vecer

K10632_FM.indd 3 11/30/10 1:56 PM

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2011 by Taylor and Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed in the United States of America on acid-free paper

10 9 8 7 6 5 4 3 2 1

International Standard Book Number-13: 978-1-4398-1252-5 (Ebook-PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts
have been made to publish reliable data and information, but the author and publisher cannot assume
responsibility for the validity of all materials or the consequences of their use. The authors and publishers
have attempted to trace the copyright holders of all material reproduced in this publication and apologize to
copyright holders if permission to publish in this form has not been obtained. If any copyright material has
not been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmit-
ted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented,
including photocopying, microfilming, and recording, or in any information storage or retrieval system,
without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access [Link].
com ([Link] or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood
Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and
registration for a variety of users. For organizations that have been granted a photocopy license by the CCC,
a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used
only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
[Link]

and the CRC Press Web site at

[Link]
Contents

Introduction ix

1 Elements of Finance 1
1.1 Price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Time Value of Assets, Arbitrage and No-Arbitrage Assets . . 14
1.4 Money Market, Bonds, and Discounting . . . . . . . . . . . . 17
1.5 Dividends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.6 Portfolio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.7 Evolution of a Self-Financing Portfolio . . . . . . . . . . . . 23
1.8 Fundamental Theorems of Asset Pricing . . . . . . . . . . . . 28
1.9 Change of Measure via Radon–Nikodým Derivative . . . . . 44
1.10 Leverage: Forwards and Futures . . . . . . . . . . . . . . . . 48

2 Binomial Models 59
2.1 Binomial Model for No-Arbitrage Assets . . . . . . . . . . . 60
2.1.1 One-Step Model . . . . . . . . . . . . . . . . . . . . . 61
2.1.2 Hedging in the Binomial Model . . . . . . . . . . . . . 65
2.1.3 Multiperiod Binomial Model . . . . . . . . . . . . . . 66
2.1.4 Numerical Example . . . . . . . . . . . . . . . . . . . 67
2.1.5 Probability Measures for Exotic No-Arbitrage Assets . 73
2.2 Binomial Model with an Arbitrage Asset . . . . . . . . . . . 75
2.2.1 American Option Pricing in the Binomial Model . . . 78
2.2.2 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.2.3 Numerical Example . . . . . . . . . . . . . . . . . . . 81

3 Diffusion Models 91
3.1 Geometric Brownian Motion . . . . . . . . . . . . . . . . . . 93
3.2 General European Contracts . . . . . . . . . . . . . . . . . . 99
3.3 Price as an Expectation . . . . . . . . . . . . . . . . . . . . . 109
3.4 Connections with Partial Differential Equations . . . . . . . 111
3.5 Money as a Reference Asset . . . . . . . . . . . . . . . . . . 114
3.6 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.7 Properties of European Call and Put Options . . . . . . . . 122
3.8 Stochastic Volatility Models . . . . . . . . . . . . . . . . . . 127
3.9 Foreign Exchange Market . . . . . . . . . . . . . . . . . . . . 130
3.9.1 Forwards . . . . . . . . . . . . . . . . . . . . . . . . . 131

v
vi Stochastic Finance: A Numeraire Approach

3.9.2 Options . . . . . . . . . . . . . . . . . . . . . . . . . . 133

4 Interest Rate Contracts 137

4.1 Forward LIBOR . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.1.1 Backset LIBOR . . . . . . . . . . . . . . . . . . . . . . 139
4.1.2 Caplet . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.2 Swaps and Swaptions . . . . . . . . . . . . . . . . . . . . . . 141
4.3 Term Structure Models . . . . . . . . . . . . . . . . . . . . . 143

5 Barrier Options 149

5.1 Types of Barrier Options . . . . . . . . . . . . . . . . . . . . 150
5.2 Barrier Option Pricing via Power Options . . . . . . . . . . . 152
5.2.1 Constant Barrier . . . . . . . . . . . . . . . . . . . . . 152
5.2.2 Exponential Barrier . . . . . . . . . . . . . . . . . . . 157
5.3 Price of a Down-and-In Call Option . . . . . . . . . . . . . . 160
5.4 Connections with the Partial Differential Equations . . . . . 165

6 Lookback Options 171

6.1 Connections of Lookbacks with Barrier Options . . . . . . . 171
6.1.1 Case α = 1 . . . . . . . . . . . . . . . . . . . . . . . . 173
6.1.2 Case α < 1 . . . . . . . . . . . . . . . . . . . . . . . . 174
6.1.3 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.2 Partial Differential Equation Approach for Lookbacks . . . . 180
6.3 Maximum Drawdown . . . . . . . . . . . . . . . . . . . . . . 187

7 American Options 191

7.1 American Options on No-Arbitrage Assets . . . . . . . . . . 192
7.2 American Call and Puts on Arbitrage Assets . . . . . . . . . 194
7.3 Perpetual American Put . . . . . . . . . . . . . . . . . . . . 195
7.4 Partial Differential Equation Approach . . . . . . . . . . . . 199

8 Contracts on Three or More Assets: Quantos, Rainbows

and “Friends” 207
8.1 Pricing in the Geometric Brownian Motion Model . . . . . . 209
8.2 Hedging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

9 Asian Options 219

9.1 Pricing in the Geometric Brownian Motion Model . . . . . . 226
9.2 Hedging of Asian Options . . . . . . . . . . . . . . . . . . . . 230
9.3 Reduction of the Pricing Equations . . . . . . . . . . . . . . 233

10 Jump Models 239

10.1 Poisson Process . . . . . . . . . . . . . . . . . . . . . . . . . 240
10.2 Geometric Poisson Process . . . . . . . . . . . . . . . . . . . 243
10.3 Pricing Equations . . . . . . . . . . . . . . . . . . . . . . . . 248
10.4 European Call Option in Geometric Poisson Model . . . . . 251
 Contents vii

10.5 Lévy Models with Multiple Jump Sizes . . . . . . . . . . . . 256

A Elements of Probability Theory 267

A.1 Probability, Random Variables . . . . . . . . . . . . . . . . . 267
A.2 Conditional Expectation . . . . . . . . . . . . . . . . . . . . 271
A.2.1 Some Properties of Conditional Expectation . . . . . . 274
A.3 Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
A.4 Brownian Motion . . . . . . . . . . . . . . . . . . . . . . . . 279
A.5 Stochastic Integration . . . . . . . . . . . . . . . . . . . . . . 283
A.6 Stochastic Calculus . . . . . . . . . . . . . . . . . . . . . . . 285
A.7 Connections with Partial Differential Equations . . . . . . . 287

Solutions to Selected Exercises 293

References 313

Index 323
Introduction

This book is based on lecture notes from stochastic finance courses I have been
teaching at Columbia University for almost a decade. The students of these
courses – graduate students, Wall Street professionals, and aspiring quants –
has had a significant impact on this text and on my teaching since they have
firsthand feedback from the dynamic world of finance. The content of this book
addresses both the needs of practitioners who want to expand their knowl-
edge of stochastic finance, and the needs of students who want to succeed as
professionals in this field. Since it also covers relatively advanced techniques
of the numeraire change, it can be used as a reference by academics working
in the field, and by advanced graduate students.

A typical reader should already have some basic knowledge of stochastic

processes (Markov chains, Brownian motion, stochastic integration). Thus the
prerequisite material on probability and stochastic calculus appears only in
the Appendix, so the reader who wants to review this material should refer
to this section first. In addition, most of the students who previously studied
this material had also been exposed to some elementary concepts of stochastic
finance, so some limited knowledge of the financial markets is assumed in the
text. This book revisits some concepts that may be familiar, such as pricing
in binomial models, but it presents the material in a new perspective of prices
relative to a reference asset.

One of the goals of this book is to present the material in the simplest possi-
ble way. For instance, the well-known Black–Scholes formula can be obtained
in one line by using the basic principles of finance. I often found that it is
quite hard to find the easiest, or the most elegant, solution but certainly a lot
of effort has been spent achieving this. The reader should keep in mind that
this is a demanding field on the level of the mathematical sophistication, so
even the simplest solution may look rather complicated. Nevertheless, most of
the ideas presented here rely on intuition, or on basic principles, rather than
on technical computations.

This book differs from most of the existing literature in the following way:
it treats the price as a number of units of one asset needed for an acquisition
of a unit of another asset, rather than expressing prices in dollar terms exclu-
sively. Since the price is a relationship of two assets, we will use a notation
that will indicate both assets. The price of an asset X in terms of a reference

ix
x Stochastic Finance: A Numeraire Approach

asset Y at time t will be denoted by XY (t). This will allow us to distinguish

between the asset X itself, and the price of the asset XY . This distinction is
important since many financial relationships can be expressed in terms of the
assets. The existing literature tends to mix the concept of an asset with the
concept of the price of an asset.

The reference asset serves as a choice of coordinates for expressing the

prices. The price appears in many different markets, and sometimes it is even
not interpreted as a price process. The simplest example is a dollar price of
an asset, where a dollar is a reference asset. Dollar prices appear in two major
markets: an equity market where the primary assets are stocks, and a foreign
exchange market where the primary assets are currencies. The prices in the
foreign exchange market are also known as exchange rates.

The foreign exchange market shows that the reference asset that is chosen
for pricing can be relative. For instance, information about how many dollars
are required to obtain one euro is the same as how many euros are required to
obtain one dollar. Since in principle there is nothing special about choosing
one or the other currency as a reference asset, it is important to create models
of the price processes that treat both assets equivalently. Thus we treat the
reference asset as relative, and using an analogy from physics, the theory pre-
sented here can be called a theory of relativity in finance. It essentially means
that the observer – an agent in a given economy – should see a similar type
of evolution of prices no matter what reference asset is chosen.

Sometimes a different reference asset than a dollar is used. For instance,

when the reference asset is a money market, or a bond, the resulting price
is known as a discounted price. An even less obvious example of a price is a
forward London Interbank Offered Rate, or LIBOR for short, where the ref-
erence asset is a bond. Markets that trade LIBOR are known as fixed income
markets. Since the prices in the fixed income markets (in this case known as
forward rates) are expressed in terms of bonds, it is strictly suboptimal to
use a dollar as a reference asset in this case. This book presents a unified ap-
proach that explains how to compute the prices of contingent claims in terms
of various reference assets, and the principles presented here apply to different
markets.

Using dollars and currencies in general for hedging or investing is prob-

lematic since holding money in terms of the banknotes creates an arbitrage
opportunity – ability to make a risk free profit – for the issuer of the currency.
Stated equivalently, money has time value; a dollar now is more valuable than
a dollar tomorrow. We can write $t > $t+1 . In order not to lose the value with
the passage of time, currencies have to be invested in assets that do not lose
value with the passage of time, such as bonds, non-dividend paying stocks,
interest bearing money market accounts, or precious metals. Note that the
 Introduction xi

currency and the interest bearing money market account are two different
assets – the first loses value with time, the second does not. When the asset
X keeps that same value with the passage of time, we can write Xt = Xt+1 .
This relationship does not mean that the price of such an asset with respect
to a reference asset Y would stay the same; the price XY (t) can be changing
with time. For instance, an ounce of gold is staying physically the same as an
asset; the gold today is the same as the gold tomorrow, but the dollar price
of the ounce of gold can be changing.

Making a loose connection with physics – money is a choice of a refer-

ence asset (or coordinates) that comes with friction. The time value of money
is analogous to movement with friction. It is always easier to add friction
(money) to the theory of frictionless markets as opposed to removing the
friction (say through adding interest on the money market) in the theory of
markets inherently built with friction. If one holds a unit of the currency, the
unit will keep creating arbitrage opportunities for the issuer of the currency.
Money in terms of banknotes is acceptable if we use it as a spot reference
asset, but it should not be used for hedging or for investment. Therefore we
focus our attention in the following text on reference assets that do not cre-
ate arbitrage opportunities through time, and develop a frictionless theory of
pricing financial contracts.

We call assets that keep the same value with the passage of time as no-
arbitrage assets, as opposed to arbitrage assets that have time value. Note
that an arbitrage asset itself, such as a currency, can be bought or sold, but it
creates arbitrage opportunities as time elapses. Examples of no-arbitrage as-
sets include interest bearing money market accounts, precious metals, stocks
that reinvest dividends, options, or contracts that agree to deliver a unit of a
certain asset in the future. The asset to be delivered may not necessarily be a
no-arbitrage asset, such as in the case of a zero coupon bond – a contract that
delivers a dollar (an arbitrage asset) at some future time. The zero coupon
bond itself does not create arbitrage opportunities in time (until expiration),
and thus can serve as a no-arbitrage reference asset.

The fundamental principle of the modern finance is the non-existence of

any arbitrage opportunity in the markets. Therefore the theory applies only
to no-arbitrage assets that do not lose value with the passage of time. The cen-
tral reason why we can determine the price of a contingent claim is the First
Fundamental Theorem of Asset Pricing which underscores the importance
of the no-arbitrage principle. This theorem states that when the prices are
martingales under the probability measure that corresponds to the reference
asset, the model does not admit arbitrage. The existence of such a martingale
measure allows us to express the prices of contingent claims as conditional
expectations under this measure, giving us a stochastic representation of the
prices. However, the First Fundamental Theorem of Asset Pricing applies only
xii Stochastic Finance: A Numeraire Approach

to prices expressed in terms of no-arbitrage assets as opposed to dollar values,

so only no-arbitrage assets have their own corresponding martingale measure.
Arbitrage assets, such as dollars, do not have their own martingale measure,
and the prices with respect to arbitrage assets have to be computed from the
change of numeraire formula using no-arbitrage assets. The First Fundamental
Theorem of Asset Pricing is introduced early in the text, and all the pricing
formulas follow from this theorem.

In this book we study financial contracts that are written on other under-
lying assets. Such contracts are called derivatives since they depend on other
assets. Sometimes we also call them contingent claims. We study the price
and the hedge of a derivative contract whose payoff depends on more basic
assets. The key idea of pricing and hedging derivative contracts is to identify a
portfolio that either matches or at least closely mimics the contract by active
trading in the underlying assets. It turns out that such a trading strategy in
most cases does not depend on the evolution of the price of the underlying
assets, and thus we can to some extent ignore the real price evolution of the
basic assets.

Single asset contracts depend on only one underlying asset, which we call
X. Such contracts include a contract to deliver a unit of X at some future
time T . This is a special case of a forward. A forward is a contract that deliv-
ers an asset X for K units of an asset Y . Thus a contract to deliver a unit of
X represents a choice of K = 0 in the forward contract. When the underlying
asset to be delivered is a currency, the contract is known as a bond. A zero
coupon bond B T is a contract that delivers one dollar at time T . Contracts
on two assets, say X and Y , include options. An option is a contract that de-
pends on two or more underlying assets that has a nonnegative payoff. This is
essentially the right to acquire a certain combination of the underlying assets
at the time of maturity of the option contract (European-type options), or
any time up to the time of maturity of the contract (American-type options).
Contracts written on three or more assets include quantos and most exotic
options such as lookback and Asian options.

Assets with a positive price that enter a given contract can be used as
reference assets for pricing this financial contract. Such assets are called nu-
meraires. Whenever possible, it is desirable to choose a no-arbitrage asset as
a reference asset since we can apply the results of the First Fundamental The-
orem of Asset Pricing directly. Most existing financial contracts can in fact
be expressed only in terms of no-arbitrage assets with one notable exception
– American stock options are settled in the stock and the dollar, and there
is no way to replace the dollar with a suitable no-arbitrage asset. This makes
American options exceptional in terms of pricing, since the price of the op-
tion has to be expressed with respect to the dollar, which is an arbitrage asset.
 Introduction xiii

Computation of the dollar prices of contingent claims cannot be done di-

rectly by applying the First Fundamental Theorem of Asset Pricing. A widely
used approach is to assume a deterministic evolution of the dollar price of the
money market account, and relate the dollar value to the money market value
by discounting. The First Fundamental Theorem of Asset Pricing applies to
the money market account, and so the dollar prices may be computed from
this relationship. The martingale measure that is associated with the money
market account is also known as the risk neutral measure. This approach has
two limitations. The first limitation is that the dollar price of the money mar-
ket is not typically deterministic due to the stochastic evolution of the interest
rate, in which case this method does not apply at all. The second limitation
is that for more complex financial products, computation of the price of a
contingent claim in terms of a dollar may be unnecessarily complicated when
compared to pricing with respect to other reference assets that are more nat-
ural to use in a given situation.

Our strategy of computing the dollar prices is different and it applies in

general. First, we identify the natural reference no-arbitrage assets which can
be used in the First Fundamental Theorem of Asset Pricing. For instance, we
will show in the later text that a European stock option has two natural refer-
ence no-arbitrage assets: a bond B T that matures at the time of the maturity
of the option, and the stock S itself. We can compute the price of the con-
tingent claim using either the probability measure that comes with the bond
B T (also known as a T-forward measure), or the probability measure that
comes with the stock S. Once we have the price of the contingent claim with
respect to the bond B T (or the stock S), we can trivially convert this price
to its dollar value by a relationship known as the change of numeraire formula.

The advantage of the numeraire approach described above may not be en-
tirely obvious for a relatively simple financial contract. Its price can be found
easily using both methods. However, for more complex products, such as for
barrier options, lookback options, quantos, or Asian options, the numeraire
approach has clear advantages – it leads to simpler pricing equations. We will
also illustrate that the barrier option and the lookback option can be related
to a plain vanilla contract. We will also show how to identify the basic assets
that enter a given contract; for instance, the lookback option depends on a
maximal asset, and the Asian option depends on an average asset.

The understanding of representing prices as a pairwise relationship of two

assets is a fundamental concept, but many books treat it as an advanced topic.
Our approach has several advantages as it leads to a deeper understanding
of derivative contracts. When a given contract depends on several underlying
assets, we can compute the price of the contract using all available reference
assets. It is often the case that a choice of a particular reference asset leads to a
simpler form. We also find some pricing formulas that are model independent.
xiv Stochastic Finance: A Numeraire Approach

Examples that admit a simple solution with the approach mentioned in this
book include a model independent formula for European call options, a simple
method for pricing barrier options, lookback options and Asian options, and
a formula for options on LIBOR.

The book has the following structure. The first chapter of this book intro-
duces basic concepts of finance: price, the concept of no arbitrage, portfolio
and its evolution, types of financial contracts, the First Fundamental Theo-
rem of Asset Pricing, and the change of numeraire formula. The subsequent
chapters apply these general principles for three kinds of models: a binomial
model, a diffusion model, and a jump model. The binomial model tends to be
too simplistic to be used in practice, and we include it only as an illustration
of the concept of the relativity of the reference asset. The novel approach is
that the prices of these contracts have two or more natural reference assets,
and thus there are two or more equivalent descriptions of the pricing problem.
In continuous time, we study both diffusion and jump models of the evolution
of the price processes. We study European options, barrier options, lookback
options, American options, quantos, Asian options, and term structure mod-
els in more detail. The Appendix summarizes basic results from probability
and stochastic calculus that are used in the text, and the reader can refer to
it while reading the main part of the book.

I am grateful to the audiences of my stochastic finance classes given at

Columbia University, the University of Michigan, Kyoto University, and the
Frankfurt School of Finance and Management. I have also received valuable
feedback from the participants in the seminar talks that I gave at Harvard Uni-
versity, Stanford University, Princeton University, the University of Chicago,
Cambridge University, Oxford University, Imperial College, King’s College,
Carnegie Mellon University, Cornell University, Brown University, the Uni-
versity of Waterloo, the University of California at Santa Barbara, the City
University of New York, Humboldt University, LMU Muenchen, Tsukuba Uni-
versity, Osaka University, the University of Wisconsin – Milwaukee, Brigham
Young University, Charles University in Prague, CERGE-EI, and the Prague
School of Economics. The research on the book was sponsored in part by
the Center for Quantitative Finance of the Prague School of Advanced Legal
Studies.

I would also like to thank the following people for comments and suggestions
that helped to improve this manuscript: Mary Abruzzo, Mario Altenburger,
Martin Auer, Jun Kyung Auh, Josh Bissu, Mitch Carpen, Peter Carr, Kan
Chen, Ivor Cribben, Emily Doran, Helena Dona Duran, Clemens Feil, Scott
Glasgow, Nikhil Gutha, Olympia Hadjiliadis, Adrian Hashizume, Gerardo
Hernandez, Amy Herron, Sean Ho, Tomoyuki Ichiba, Karel Janecek, Xiao Jia,
Philip Johnston, Armenuhi Khachatryan, David Kim, Thierry Klaa, Sharat
Kotikalpudi, Ka-Ho Leung, Jianing Li, Sasha Lv, Rupal Malani, Antonio Med-
 Introduction xv

ina, Vishal Mistry, Amal Moussa, Daniel Neelson, Petr Novotny, Kimberli
Piccolo, Radka Pickova, Dan Porter, Libor Pospisil, Cara Roche, Johannes
Ruf, Steven Shreve, Lisa Smith, Li Song, Joyce Yuan Hui Su, Stephen Taylor,
Uwe Wystup, Mingxin Xu, Ira Yeung, Wenhua Zou, Hongzhong Zhang, and
Ningyao Zhang. The editors and the production team from the CRC Press
provided much needed assistance, namely, Sunil Nair, Sarah Morris, Karen
Simon, Amber Donley, and Shashi Kumar. The whole project would not be
possible without the unconditional support of my family.
Chapter 1
Elements of Finance

Some of the basic concepts of finance are widely understood in broad terms;
however this chapter will introduce them from a novel perspective of prices
being treated relative to a reference asset. We first show the difference be-
tween an asset and the price of an asset. The price of an asset is always ex-
pressed in terms of another reference asset. The reference asset is also called
a numeraire. The numeraire asset should never become worthless so that
the price with respect to this asset is well defined. The relationship between
prices of an asset expressed with respect to two different reference assets is
known as a change of numeraire. The concept of price appears in differ-
ent markets under different names, so it may not be obvious that it is just
a particular instance of a more general concept. For instance, an exchange
rate is in fact a price representing a pairwise relationship of two currencies.
An even less obvious example of a price is a forward London Interbank Offer
Rate (LIBOR). By adopting a precise definition of price, we are able to treat
various markets (equities, foreign exchange, fixed income) in one single unified
framework, which simplifies our analysis.

The second section introduces the concept of arbitrage – the possibility of

making a risk free profit. We study models of markets where no agent allows
an arbitrage opportunity. One can create an arbitrage opportunity just by
holding a single asset such as a banknote. This is known as a time value of
money. Thus the concept of no arbitrage splits assets into two groups: no-
arbitrage assets – the assets that do not allow any arbitrage opportunities;
and arbitrage assets – the assets that do allow arbitrage opportunities. In
theory, the market should have only no-arbitrage assets. Financial contracts
are typically no-arbitrage assets; they become arbitrage assets only when their
holder takes some suboptimal action (such as not exercising the American put
option at the optimal exercise time). On the other hand, real markets include
arbitrage assets such as currencies.

Currencies, in terms of banknotes, are losing an interest rate when com-

pared to the corresponding bond or money market account. Since the loss of
the currency value is typically small, money still serves as a primary reference
asset in the economy. However, in order to avoid this loss of value in pricing
contingent claims, one should use discounted prices rather than dollar prices
of the assets. Discounted prices correspond to either a bond or a money mar-

1
2 Stochastic Finance: A Numeraire Approach

ket account as a reference asset. Stocks paying dividends are arbitrage assets
when the dividends are taken out, but an asset representing the equity plus
the dividends is a no-arbitrage asset. We find a simple relationship between
the dividend paying stock and the portfolio of the stock and the dividends.

In the section that follows, we introduce the concept of a portfolio. A port-

folio is a combination of several assets, and it is important to realize that it
has no numerical value. In fact, one should not confuse the concept of a port-
folio (viewed as an asset) with the price of a portfolio (number that represents
a pairwise relationship of two assets). It should be noted that a portfolio may
be staying physically the same, but the price of this portfolio with respect
to some reference asset may be changing. We also introduce the concept of
trading. Self-financing trading is exchanging assets that have the same price
at a given moment. As a consequence, portfolios may be evolving in time by
following a self-financing trading strategy.

When no arbitrage exists in the markets, all prices are martingales with
respect to the probability measure that comes with the specific no-arbitrage
reference asset. Martingales are processes whose best estimator of the fu-
ture value is its present value. Mathematically, a process M that satisfies
Es [M(t)] = M(s), s ≤ t, is a martingale, where Es [.] denotes conditional ex-
pectation. The reader should refer to the Appendix for more details about
martingales and conditional expectation. The result that prices are martin-
gales under the probability measure that is related to the reference asset is
known as the First Fundamental Theorem of Asset Pricing. In particular,
every no-arbitrage asset has its own pricing martingale measure. Other
no-arbitrage assets have different martingale measures. The martingale mea-
sure associated with the money market account is known as a risk-neutral
measure. The martingale measures associated with bonds are known as T-
forward measures. Stocks have martingale measures known as a stock
measure. Arbitrage assets, such as currencies, do not have their own mar-
tingale measures. In particular, there is no dollar martingale measure.

Many authors do not regard currencies as true arbitrage assets because this
arbitrage opportunity is one sided for the issuer of the currency. It is also easy
to confuse money (in terms of banknotes) with the money market account.
Banknotes deposited in a bank start to earn the interest rate and become a
part of the money market account. When borrowing money, the debt is not a
currency, but rather the corresponding money market account. The debt earns
the interest to the lender, and thus it behaves like the money market account.
However, arbitrage pricing theory applies only to no-arbitrage assets, such as
the money market account, bonds, or stocks. It does not apply to money in
terms of banknotes. No-arbitrage assets have their own martingale measure,
while arbitrage assets do not.
 Elements of Finance 3

An important consequence of the First Fundamental Theorem of Asset Pric-

ing is that the prices are martingales with respect to a probability measure
associated with a particular reference asset. Martingales in continuous time
models are under some assumptions just combinations of continuous martin-
gales, and purely discontinuous martingales. Moreover, continuous martin-
gales are stochastic integrals with respect to Brownian motion. This limits
possible evolutions of the price to this class of stochastic processes since other
types of evolutions allow for an existence of arbitrage.

Another related question to the concept of no arbitrage is a possibility of

replicating a given financial contract by trading in the underlying primary as-
sets. The martingale measure from the First Fundamental Theorem of Asset
Pricing may not necessarily be unique; each reference asset may have infinitely
many of such measures. However, each martingale measure under one refer-
ence asset has a corresponding martingale measure under a different reference
asset that agrees on the prices of the financial contracts. The two measures are
linked by a Radon–Nikodým derivative. In particular, when there is a unique
martingale measure under one reference asset, the martingale measures that
correspond to other reference assets are also unique due to the one-to-one
correspondence of the martingale measures.

In the case when the martingale measure is unique, all financial contracts
can be perfectly replicated. This result is known as the Second Fundamental
Theorem of Asset Pricing. The market is complete essentially in situations
when the number of different noise factors does not exceed the number of
assets minus one. Thus models with two assets are complete when there is
only one noise factor, which is, for instance, the case in the binomial model,
in the diffusion model driven by one Brownian motion, or in the jump model
with a single jump size. When the market is complete, the financial contracts
are in principle redundant since they can be replicated by trading in the
underlying primary assets. The replication of the financial contracts is also
known as hedging.

1.1 Price
This section defines price as a pairwise relationship of two assets.

Price is a number representing how many units of an asset Y are

required to obtain a unit of an asset X.
4 Stochastic Finance: A Numeraire Approach

We denote this price at time t by

XY (t).

Here an asset Y serves as a reference asset. The reference asset is known as a

numeraire. Price is always a pairwise relationship of two assets.

For practical purposes the role of a reference asset is typically played by

money, a choice of the reference asset Y being a dollar $. However, the choice
of the reference asset is in principle arbitrary as long as the reference asset is
not worthless. The reader should note that some financial assets may become
worthless at a certain stage (such as options expired out of the money), and
such contracts would be a poor choice of the reference asset. There are also
some desirable properties that the reference asset should satisfy: it should be
sufficiently durable, and there should exist enough identical copies of the as-
set. From this perspective, consumer goods (such as cars, electronic products,
most food products) may be used as a reference asset, but this choice would
not be appropriate since the asset itself has time value; it is deteriorating in
time.

In practice, a small loss of the value of the reference asset is acceptable.

Currencies in particular lose value in time by allowing an arbitrage opportu-
nity with respect to the money market account, and they still play a role of a
primary reference asset in the economy. However, when the loss of the value
becomes large, for instance in a period of hyperinflation, such currency may
no longer be accepted as a reference asset. The property of having sufficient
identical copies of the asset ensures that the individuals in the economy can
easily acquire the reference asset. The reference asset should be sufficiently
liquid. For instance some art works (paintings, sculptures, buildings) have a
significant value, but they cannot be easily bought or sold and thus using
them as a reference asset would not be a good choice.

Typical choices of a reference asset used in practice are currencies (denoted

by $, e, £, ¥, etc.), bonds (denoted by B T ), a money market (denoted by
M ), or stocks and stock indices (denoted by S). A bond B T is an asset that
delivers one dollar at time T . The money market M is an asset that is
created by the following procedure. The initial amount equal to one dollar
is invested at time t = 0 in the bond with the shortest available maturity
(ideally in the next infinitesimal instant), and this position is rolled over to
the bond with the next shortest maturity once the first bond expires. The
resulting asset, the money market M , is a result of an active trading strategy
involving a number of these bonds. In principle, there is a counter party risk
involved in delivering a unit of a currency at some future time. The counter
party may fail to deliver the agreed amount at the specified time. The follow-
ing text assumes situations when there is no such risk present, as in the case
 Elements of Finance 5

when the delivery of the asset is guaranteed by the government.

The reference asset itself does not need to be a traded asset. As we will see
in the chapter on pricing exotic options, some natural reference assets that
are useful for pricing complex financial contracts do not exist in real markets.
For instance, one can use an asset that represents the running maximum of
the price max0≤s≤t XY (s) for pricing lookback options, or one can use an as-
set that represents the average price for pricing Asian options. A price of a
financial contract that is expressed in terms of an asset which is not traded
can be easily converted to a price expressed in terms of a traded asset. Thus
for practical purposes it does not matter if the reference asset exists or not in
real markets.

Let us introduce the following notation. By Xt we mean a unit of an asset

X at time t, not its price in terms of a different asset. In principle, an asset X
that has no time value stays the same at all times (think of an ounce of gold),
so there is really no need to index it with time. However, by adding the time
coordinate we express that a particular asset is used at that time for trading,
pricing, hedging, or for settling some contract. When there is no ambiguity,
we will simply drop the time index, and write only X to stress that the asset
in fact stays the same.

Recall that price is a pairwise relationship of two assets denoted by XY (t)

– the number of units of an asset Y required to obtain one unit of an asset X.
The asset Y is known as a reference asset, or as a numeraire. We can write
that
1 unit of X = XY (t) units of Y ,
or simply
X = XY (t) · Y. (1.1)
Assets X and Y on their own do not have any numerical value (such as an
ounce of gold), and the above equality does not mean that the assets on the
left hand side and on the right hand side of the equation are physically the
same. Note that we cannot divide by Y in the above equation since Y is an
asset.

The relation “=” when used for assets as in Equation (1.1) is an equivalence
relation. We will write Xt = Yt in the sense of assets when XY (t) = 1
in the sense of numbers. Clearly, the relation “=” for assets is reflexive
(Xt = Xt ), symmetric (Xt = Yt implies Yt = Xt ), and transitive (Xt = Yt
and Yt = Zt imply Xt = Zt ). The assets are also ordered according to their
prices. We can write Xt ≥ Yt in terms of assets when XY (t) ≥ 1 in terms
of numbers. It should be noted that two assets X and Y with an equal price
at time t1 (meaning XY (t1 ) = 1) may differ in price at some other time t2
(meaning XY (t2 ) 6= 1). If two assets X and Y have the same price at time t,
6 Stochastic Finance: A Numeraire Approach

they can be exchanged for each other at that time. This procedure is known
as a self-financing trade.

It may not be clear as to why we should adopt notation XY (t) for the price,
instead of using just a single letter for it, say S(t), which is typically used for
the price of a stock in terms of dollars. The following examples illustrate that
the concept of price appears in different markets, such as in equity markets,
in the foreign exchange markets, or in fixed income markets. By using our
notation, we are able to treat these prices in one single framework, rather
than studying them separately.

Example 1.1 Examples of the price

• The dollar price of an asset S, S$ (t), where the role of the asset X is
played by the stock S, and the role of the reference asset Y is played
by the dollar $. Most of the current literature writes simply S(t) for
the dollar price S$ (t) of this asset, but we want to avoid in our text
confusing the asset S itself with the price of the asset S$ (t).
• The price of a stock S in terms of the money market M , SM (t), where
the asset X is a stock S, and the asset Y is a money market M with
M0 = $0 . The price SM (t) is known as a discounted price of an asset
S.
• The price of a stock S in terms of a zero coupon bond B T with maturity
T , SB T (t), where the asset X is a stock S, and the asset Y is a bond B T .
This is also a form of a discounted price which is more appropriate
than SM for pricing derivative contracts that depend on S and $. Note
that we have SB T (T ) = S$ (T ).
• The exchange rate, e$ (t), where X is the foreign currency (e), and Y
is the domestic currency $. The choice of domestic and foreign currency
is relative, and thus $e (t) is also an exchange rate.
• Forward London Interbank Offered Rate, or forward LIBOR for
short,  T 
B − B T +δ δB T +δ (t),
where the role of the asset X is played by a portfolio of two bonds
[B T − B T +δ ], and the reference asset Y is δ · B T +δ .

We will discuss these examples of price in more detail after introducing the
concepts of inverse price, and change of numeraire. Since the assets X and Y
considered in the above are arbitrary, it also makes perfect sense to consider
 Elements of Finance 7

the inverse relationship when X is chosen as a reference asset. For instance,

one may think about X and Y as two currencies. When X = e, and Y = $,
we have both the exchange rate e$ (t) – the number of dollars required to
obtain a unit of a euro, and the exchange rate $e (t) – the number of euros
required to obtain a unit of a dollar. Thus we can also write

1 unit of Y = YX (t) units of X,

or simply
Y = YX (t) · X. (1.2)
The price YX (t) is the inverse price to XY (t). Let us show the relationship
between YX (t) and XY (t). Suppose that an agent starts with a unit of an
asset Y . He can change it for YX (t) units of an asset X. This amount can be
split in two parts: YX (t)−XY (t)−1 and XY (t)−1 units of an asset X. The part
of XY (t)−1 units of an asset X can be exchanged back for a unit Y , which
follows from the relationship

X = XY (t) · Y,

which is equivalent to
Y = XY (t)−1 · X.
We can rewrite the above trading procedure using the following identities

Y = YX (t) · X
= (YX (t) − XY (t)−1 ) · X + XY (t)−1 · X
= (YX (t) − XY (t)−1 ) · X + Y.

Thus the net result of this exchange is YX (t) − XY (t)−1 units of an asset X,
which must be zero in order not to allow a risk-free profit. Therefore the prices
XY (t) and YX (t) are related by the following relationship

1
YX (t) = . (1.3)
XY (t)

This relationship is valid when 0 < XY (t) < ∞, which is the case that neither
the asset X nor the asset Y is worthless. In this case, XY (t) and its inverse
price YX (t) have the same information.

In general, it should not matter which reference asset is chosen, one should
observe similar price evolutions. We will use this as a key principle for pricing
derivative contracts studied in this book. One can look at it as a theory of
relativity in finance: how one views prices depends on one’s choice of the
reference asset.
8 Stochastic Finance: A Numeraire Approach

Given an asset X and two reference assets Y and Z, we can write the price
of X with respect to the reference asset Y using

X = XY (t) · Y. (1.4)

Similarly, we can write

X = XZ (t) · Z (1.5)
when we use Z as a reference asset. Thus we have

X = XY (t) · Y = XZ (t) · Z, (1.6)

which is known as a change of numeraire formula. The above relationship

is written in terms of assets. We can rewrite the above relationship in terms
of the price as
XY (t) = XZ (t) · ZY (t). (1.7)
This relationship is valid for assets X, Y , and Z that are not worthless.

Example 1.2 Foreign Exchange Market

Let us illustrate the concepts of the inverse price and the change of numeraire
on the foreign exchange market. Prices in the real markets satisfy the rela-
tionship
YX (t) = XY (t)−1
at all times (up to the rounding errors). For instance, on January 8th, 2010,
at 8:00PM EST, the exchange rates between e and $ were:

e$ = 1.4415 $e = 0.6937.

We can easily check that

1
$−1
e = = 1.441545...
0.6937
Thus the inverse exchange rate $−1 e matches the first four digits of the ex-
change rate e$ . The exact match is typically not possible since these exchange
rates are quoted in four decimal digits. However, the arbitrage is still not pos-
sible due to the difference of the prices offered and asked. An agent who wants
to acquire a unit of an asset should be ready to pay more than an agent who
wants to sell a unit of the same asset.

More specifically, the market exchange works in the following way: Agents
who want to buy a particular asset place their orders on the market exchange,
and wait until they find corresponding counter parties that are willing to
match their orders. The orders compete according to the price that is quoted;
a higher quote has a higher priority of being executed. The highest quote is
 Elements of Finance 9

known as the best bid. Similarly, agents who want to sell a particular asset
place their orders on the market exchange. A smaller price asked for a unit
of a given asset has a higher priority. The smallest price asked is known as
the best ask. Clearly, the best ask is larger than the best bid. The smallest
difference between two possible quoted prices on the exchange is known as a
tick. In the case of euro/dollar exchange rates, the tick is equal to 0.0001.
The difference between the best bid and the best ask is known as a bid-ask
spread. Bid-ask spreads may be larger than a tick. More liquid assets have
smaller bid-ask spreads, the difference between the buying and the selling
price being smaller.

From the perspective of having both XY (t) and YX (t) as prices, there is no
absolute direction of up and down in the market. Each trade has two sides,
a seller and a buyer. If the market moves in one direction, it is either to the
benefit of the seller and at the expense of the buyer, or vice versa. This is
another way of saying that when one of the prices XY (t) or YX (t) goes up,
the inverse price must go down.

Exchange rates also serve as an example of the change of numeraire formula.

Table 1.1 shows the exchange rate table for four major currencies: dollars,
euros, pounds, and yen as seen on January 8th, 2010 at 8:00PM EST. For
instance the entry ($, e) gives the price $e = 0.6937, etc.

TABLE 1.1: Exchange Rate Table.

$ e £ ¥
$ 1 0.6937 0.6238 92.6300
e 1.4415 1 0.8991 133.5260
£ 1.6032 1.1122 1 148.5040
¥ 0.0108 0.0075 0.0067 1

From the change of numeraire formula, we should also have among other
similar relationships
$ e = $ £ · £e . (1.8)

In fact,
$£ · £e = 0.6238 × 1.1122 = 0.693790...

This matches the original $e rate in four decimal digits if we neglect the
rounding error in the fourth digit. This match is close enough not to allow
for any arbitrage opportunities due to the market imperfections such as the
bid-ask spread, or transaction costs.
10 Stochastic Finance: A Numeraire Approach

REMARK 1.1 The change of numeraire formula (1.7) applies to all

assets, with or without time value. Note that Equation (1.8) is an example of
the change of numeraire formula for assets with time value.

Example 1.3 Forward London Interbank Offered Rate

The Forward London Interbank Offered Rate, or LIBOR for short, is defined
as a simple interest rate that corresponds to borrowing money over the time
interval T and T + δ as seen at time t ≤ T . We denote forward LIBOR by
L(t, T ). When t = T , L(T, T ) is known as a spot LIBOR since it corresponds
to borrowing money at the present time T .

Suppose that one dollar is borrowed at time T , and assume that L(t, T ) is
the simple interest rate for the period between T and T + δ. Then the agent
should return 1 + δL(t, T ) dollars at time T + δ. Thus L(t, T ) can be defined
by the following relationship:

(1 + δL(t, T )) · BtT +δ = BtT . (1.9)

The right hand side of the above relationship indicates that one dollar will be
delivered at time T . The left hand side indicates that (1 + δL(t, T )) dollars
will be returned at time T + δ. Therefore

δL(t, T ) · BtT +δ = BtT − BtT +δ . (1.10)

We can rewrite this relationship in the following form:

 
L(t, T ) = B T − B T +δ δ·B T +δ (t), (1.11)

showing that forward LIBOR L(t, T ) is in fact a price, where the asset X is
a portfolio [B T − B T +δ ] (long the B T bond, and short the B T +δ bond), and
the reference asset Y is δ units of the bond B T +δ .

If we wanted to compute XY (t) for two general assets, we can do so from

the dollar prices of the assets X and Y :
X$ (t)
XY (t) = X$ (t) · $Y (t) = , (1.12)
Y$ (t)
where we substitute Z for $ in the change of numeraire formula. Using Equa-
tion (1.12), we can determine forward LIBOR from dollar prices of bonds by
using
 
L(t, T ) = B T − B T +δ δB T +δ (t)
  B T (t) − B$T +δ (t)
= B T − B T +δ $ (t) · $δB T +δ (t) = $ . (1.13)
δB$T +δ (t)
 Elements of Finance 11

Here we have used the change of numeraire formula, and linearity of the prices:
[aX + bY ]Z (t) = aXZ (t) + bYZ (t). (1.14)

Foreign exchange markets, or fixed income markets that trade on LIBORs,

are in fact much larger than the equity markets in terms of the volume traded,
and thus the main focus of financial markets is on prices that are not expressed
exclusively in dollar terms. It is also not an obvious observation that exchange
rates and forward LIBORs are in fact prices. Calling them the exchange rates
or forward LIBORs is slightly misleading, and the literature tends to study
the asset prices, foreign exchange rates, and forward LIBORs separately. In
our approach, they are just special cases of a more general concept of price.

Price is always a pairwise relationship of two assets, and we will use this
notation throughout this book to indicate the reference asset. This distinction
will help us study derivative contracts later on in the text that are written
on more than one underlying asset. The second (or the third asset when ap-
plicable in the case of exotic options) asset also serves as a viable reference
asset for pricing a given derivative contract. This notation is especially help-
ful when studying quantos and other exotic options, which represent financial
contracts that are written on three underlying assets. The reader should also
note here that every contract is settled in units of particular assets (dollars,
stocks, bonds) rather than in the price itself – the price indicates only how
many units of a particular asset are needed.

1.2 Arbitrage
This section discusses another fundamental concept of finance, namely ar-
bitrage.

Arbitrage is an opportunity to make a risk free profit in the mar-

ket.

It is important to distinguish an arbitrage opportunity from a profitable

trading strategy. Arbitrage means that there is at least one agent that can
make money for sure, while a profitable trading strategy simply works on av-
erage, meaning that some scenarios may lead to a loss.

An arbitrage opportunity means that one can create a guaranteed profit

starting from a portfolio with a zero initial price. It is easy to see that if a
12 Stochastic Finance: A Numeraire Approach

portfolio has a zero price with respect to one asset, it has a zero price with
respect to any reference asset. A typical example of an arbitrage opportunity
is the ability to purchase an asset at a given price and then sell the same
asset immediately or some later time for a higher price. The guarantee of a
higher price is necessary to make it an arbitrage opportunity, assuring that
the portfolio always ends up with more assets than when it started. Such
arbitrage trades can happen when a purchase price in one market is less than
the selling price in a different market.

Example 1.4
Assume that at time t = 0, the price of an asset X with respect to an asset Y
is XY (0) = K. Suppose that at a fixed time T ≥ 0, the price will be exactly
XY (T ) = J with J > K. In such a case one can construct a portfolio, starting
at time t = 0 with P00 = 0, exchange it for the portfolio P01 = X − K · Y that
has a zero price (long one unit of X and short K units of Y ), and end up with
a portfolio PT1 = X − K · Y at time T . This portfolio can be exchanged by
selling a unit of an asset X for J units of an asset Y for a portfolio with the
same price PT2 = (J − K) · Y > 0. This is clearly an arbitrage opportunity.

A slightly less obvious arbitrage opportunity is a free lottery ticket. Al-

though in most cases a typical lottery ticket does not win any prize, one is
certain not to lose any money and still have a possibility of winning something.
That qualifies as an arbitrage opportunity.

Example 1.5
Assume that there is a free lottery ticket L whose price in terms of the dollar $
is zero: L$ (0) = 0. We have seen in the previous example that having dollars
in a portfolio provides an arbitrage opportunity, but let us assume for the
purpose of this example that dollars keep their value with respect to bonds in
order to illustrate a different kind of arbitrage. The lottery ticket either expires
worthless, or it wins N dollars at time T . One can construct the portfolio
starting from zero P00 = 0, acquiring one zero price lottery ticket, thus creating
a portfolio P01 = L0 . This portfolio will convert to PT1 = N · I(ω = Win) · $,
where I(ω = Win) is the indicator function of the win. We have that PT1 ≥ 0
for sure, with the possibility of PT1 > 0. This also constitutes an arbitrage
opportunity.

Another example of an arbitrage opportunity is when the price XY (t) of an

asset X in terms of an asset Y does not correspond to the price YX (t) of an
asset Y in terms of an asset X.
 Elements of Finance 13

Example 1.6 Arbitrage opportunity when XY (t) 6= YX (t)−1 .

If the relationship
1
XY (t) =
YX (t)
does not hold, it is possible to realize a risk-free profit. Assume for instance

1
< YX (t).
XY (t)

In this case, we can start with a unit of an asset Y , and exchange it for YX (t)
units of an asset X. We can split this position in two parts: YX (t) − XY (t)−1
and XY (t)−1 units of an asset X. The second part, XY (t)−1 units of an asset
X, can be exchanged back for a unit of an asset Y . This follows from

X = XY (t) · Y,

which is equivalent to
Y = XY (t)−1 · X.

Therefore one can generate a certain profit of YX (t) − XY (t)−1 > 0 units of
an asset X.

Example 1.7
Assume that XY (t) = 3, and YX (t) = 21 . How can one realize a risk free profit?
First check that YX (t) = 12 6= XY (t)−1 (t) = 31 . Therefore the prices allow for
an arbitrage opportunity. Following the method described in the previous
example, we can start with borrowing one unit of Y . Using YX (t) = 21 , we can
immediately exchange the unit of Y for 12 units of X. We can split 12 units
of X in two parts, consisting of 61 and 13 units of X. The first part 61 units
of X is a net profit from this transaction; the second part can be used for
an acquisition and return of a borrowed unit Y using the price relationship
XY (t) = 3.

Formally, an arbitrage opportunity is defined by:

If one starts with a zero initial portfolio P0 = 0, follows a self-

financing strategy, and ends up with PT ≥ 0 with probability 1, and
has a possible outcome of PT > 0 with positive probability at any
given time T , then an arbitrage opportunity is available in the mar-
ket.

Note that the definition of an arbitrage opportunity does not depend on

the choice of the reference asset Y . If PY = 0 or PY > 0 for the reference
asset Y , then PU = 0 or PU > 0 for any other reference asset U .
14 Stochastic Finance: A Numeraire Approach

1.3 Time Value of Assets, Arbitrage and No-Arbitrage

Assets
As stated in the previous section, an asset can either stay the same over
time or change over time. In the first case, we say that the asset has no time
value. Examples of assets that do not change over time include precious met-
als, a contract to deliver a particular asset in some fixed future time, or a stock
that reinvests dividends. One should not confuse the concept of an asset with
no time value with the concept of the price of an asset with no time value.
For instance an ounce of gold is an asset with no time value, and it does not
change over time, but the price of this asset with respect to a dollar may be
changing over time.

When the asset is changing over time, we say that the asset has a time
value. Assets with time value may deteriorate over the passage of time or
not. Examples of time value assets that deteriorate over time include curren-
cies, stocks that pay out dividends, and most consumer goods. However, some
assets may change over time and not deteriorate, for instance portfolios that
actively exchange assets with no time value.

One certainly does not create an arbitrage opportunity by holding an asset

that has no time value. On the other hand, assets that have time value may
or may not create arbitrage opportunities. It depends if the asset with time
value deteriorates (or appreciates) in time or not. If one creates an arbitrage
opportunity by holding a given asset, we will call this asset an arbitrage
asset. If an arbitrage opportunity is not possible by holding a given asset, we
call this asset a no-arbitrage asset. There is a simple method to determine
whether a given asset X is an arbitrage or a no-arbitrage asset. Let V be a
contract to deliver a unit of the asset X at some future time T . We can write

VT = XT .

When Vt = Xt at all times t ≤ T , the asset X is a no-arbitrage asset. When

Vt 6= Xt for some t ≤ T , the asset X is an arbitrage asset.

The identity Vt = Xt means that V , the contract to deliver a unit of an as-

set X, is identical to the asset X itself. The only way to deliver a no-arbitrage
asset is to hold it at all times up to time T . For instance the contract to deliver
a stock costs the stock itself, a contract to deliver an ounce of gold costs the
ounce of gold (neglecting a possible cost of carry which is close to zero for
financial assets). Some hedge funds try to realize arbitrage opportunities even
in these primary assets, so it may be hard to tell which asset is a no-arbitrage
asset without observing the corresponding contract to deliver. A contract to
deliver usually does not exist for a no-arbitrage asset since it coincides with
 Elements of Finance 15

the asset itself, and thus it is completely redundant. However, the nonexis-
tence of the contract to deliver can happen for two reasons: the underlying
asset is a no-arbitrage asset, or there is no market for the contract to deliver.
This makes it harder to determine whether the asset is a no-arbitrage asset.

Rational investors do not allow any arbitrage opportunities, and thus their
portfolios hold only no-arbitrage assets, or arbitrage assets that provide one
sided advantage for the investor. If the market has only rational investors,
there would be no arbitrage assets at all. For a given asset X, the contract
V to deliver an asset X is always a no-arbitrage asset, even when
the asset X to be delivered is an arbitrage asset. This is easily seen from
the following argument. Let U be a contract to deliver the asset V at time
T , or in other words, UT = VT . From the identity VT = XT , we also have
UT = XT . Thus U is also a contract to deliver X at time T , and there-
fore U is identical to V . This proves that V , a contract to deliver an asset X
at time T , is a no-arbitrage asset. In particular, bonds are no-arbitrage assets.

On the other hand, assets with Vt 6= Xt for some t < T are arbitrage assets.
We have either Vt < Xt , or Vt > Xt . When Vt < Xt , it is possible to deliver
the asset X at time T at a cheaper price than just holding the asset X itself.
The exact procedure to lock the arbitrage opportunity for an arbitrage asset
is described in Example 1.8 which follows. When Vt < Xt , one should buy
a contract to deliver V and sell a corresponding number of units of an asset X.

Arbitrage assets do exist in real markets, mostly representing assets with

deteriorating time value (food, consumer goods, banknotes). However, these
assets are not typically included in financial portfolios as holding them would
create arbitrage opportunities that are not favorable for the holders of such
assets. But the arbitrage assets still may appear in the payoffs of financial
contracts, such as a contract to deliver a unit of the asset in a fixed future
time. We have already seen that a contract to deliver any asset is always a
no-arbitrage asset. Such derivative contracts facilitate trading of assets with
deteriorating time value. While the underlying asset creates arbitrage oppor-
tunities, the contract to deliver does not, and as such it may be included in a
financial portfolio that does not deteriorate over time.

Examples of arbitrage assets that appear in such payoffs include certain

food products (orange juice, coffee, pork bellies), currencies, or stocks that
pay dividends. A stock together with the corresponding dividend payments
is a no-arbitrage asset. However, a stock when taken separately without the
dividends is an arbitrage asset. Taking away the dividends is an obvious ar-
bitrage opportunity. Another example of an arbitrage asset is an asset that
corresponds to a maximum price of an asset X with respect to a reference
asset Y defined as [max0≤s≤t XY (s)] · Yt . This asset appears in the payoff of
a lookback option, and although it does not exist in the real markets, it can
16 Stochastic Finance: A Numeraire Approach

still be used as a reference asset for pricing lookback options.

Arbitrage assets do change over some periods of time; in particular we have

$t > $t+1 , (1.15)

which means that a dollar today is worth more than a dollar tomorrow. In-
equality (1.15) is known as the time value of money.

Example 1.8 Arbitrage opportunity created by an arbitrage asset

Let V be a contract that delivers a unit of an asset X at time T , or in other
words,
VT = XT .

This equality is written in the sense of two assets, the contract to deliver V
has the same price as an asset X at time T . In terms of prices, we can write

VX (T ) = 1,

which means that the price of the contract to deliver V with respect to the
reference asset X is one at time T . When V0 < X0 , we can realize a risk free
profit by buying a unit of an asset V , and sell VX (0) < 1 units of an asset X,
thus creating a zero price portfolio

P0 = 1 · V − VX (0) · X.

Clearly, PX (0) = 1 · VX (0) − VX (0) = 0. This portfolio is kept until time T ,

when it becomes

PT = 1 · V − VX (0) · X
= (1 − VX (0)) · X > 0.

Thus one can get a portfolio with a guaranteed positive price starting from a
portfolio with a zero price.

The most typical examples of arbitrage assets are currencies. Let X be a

dollar $. A contract to deliver a dollar at time T is known as a bond, and
it is denoted by B T . The dollar price of the bond is typically less than one
(B$T (0) < 1), making a dollar an arbitrage asset. In order to lock the risk free
profit, one would have to buy a bond B T , and sell B$T (0) units of a dollar.
This means one would have to borrow money to get a short position in dollars,
which leads us to the following important remark.
 Elements of Finance 17

REMARK 1.2 Borrowing money

When one borrows money in terms of a dollar $, the resulting

asset that is owed is not money but rather a money market account
M , an interest bearing account. The asset that is borrowed is different
from the asset that is owed. In contrast, if one borrows a stock S (in terms of
short-selling on the stock exchange), the debt is still the same stock S. The
exchange may charge a fee for that, but the asset that is borrowed is the same
as the asset that is owed.

Even governments have to pay interest when borrowing money. The only
exception when interest is not paid is when governments issue banknotes.
Governments typically have a limited intention to print more banknotes in
order to finance their debts, and thus exploration of this arbitrage opportunity
is not significant.

1.4 Money Market, Bonds, and Discounting

The fact that currencies have time value means that prices in terms of a
dollar may not be consistent in time. This is known as time value of money: A
dollar today is worth more than a dollar tomorrow. Thus when one expresses
prices of an asset S in terms of a dollar, these prices will have an upward drift
component that corresponds to the loss of value of the reference asset.

In order to remove the effect of the depreciation of the reference asset, one
can express the price of the asset S in terms of no-arbitrage proxy assets to a
dollar, such as a money market M , or a bond B T . Prices SM (t) and SB T (t)
are known as discounted prices of the asset S.

Recall that the money market M is an asset created by the following pro-
cedure. The initial amount equal to one dollar is invested at time t = 0 in the
bond with the shortest available maturity (ideally in the next infinitesimal
instant), and this position is rolled over to the bond with the next shortest
maturity once the first bond expires. The resulting no-arbitrage asset, the
money market M , is a result of an active trading strategy involving a num-
ber of these no-arbitrage bonds. The dollar price of the money market is given
by
R 
t
M$ (t) = exp 0 r(u)du , (1.16)

where r(t) is a parameter known as the interest rate. In practice, the money
market asset is replicated as a portfolio of different bonds by banks or invest-
18 Stochastic Finance: A Numeraire Approach

ment funds.

Equation (1.16) can be written in a differential form as

dM$ (t) = r(t)M$ (t)dt. (1.17)

The interest rate r(t) can be viewed as a rate of deterioration of an arbitrage

asset $ with respect to a no-arbitrage asset M , the money market account.
Since the parameter r(t) is related only to the shortest available bond, in
this case B t , a bond that matures immediately at time t, a simple analog of
Equation (1.17) for a bond B T is not available. Only if we take a simplifying
assumption that the interest rate r(t) is deterministic, can we also write
 R 
T
B$T (t) = exp − t r(u)du . (1.18)

The reason is that there is only one way to deliver one dollar at time T by in-

RT
vesting in the money market account M . If one starts with exp − t r(u)du
units of a dollar at time t and invests it in the money market account M , it
will be worth
 R  R 
T T
exp − t r(u)du · exp t r(u)du = 1

unit of a dollar at time T . Therefore the price of the bond B T at time t must be
given by Equation (1.18); otherwise we would have an arbitrage opportunity.
In this case, the price of the bond B T and the price of the money market M
are related by the formula
 R 
T
BtT = exp − 0 r(u)du · Mt . (1.19)

Thus the money market M is just a constant multiple of the bond B T .

In the case of a deterministic interest rate r(t), we can also write

dB$T (t) = r(t)B$T (t)dt, (1.20)

which is similar to Equation (1.17). Moreover, when the interest rate is con-
stant, the above relationships lead to

Mt = ert · $t , (1.21)

BtT = e−r(T −t) · $t , (1.22)

and
BtT = e−rT · Mt . (1.23)
T
The relationship between the money market M and the bond B is no
longer trivial when the interest rate r(t) is stochastic. In this case, the price
 Elements of Finance 19

of the money market starts at a deterministic value M$ (0) = 1, but at later

time t, M$ (t) will be stochastic in general. On the other hand, the price of
the bond B$T (t) is random in general for times t < T before the expiration of
the bond, but it becomes one at time T (B$T (T ) = 1), which is a deterministic
value. We study the evolution of bond prices in detail in the chapter on term
structure models.

As seen earlier, we can regard both SM (t) and SB T (t) as discounted prices
of an asset S. When we express the price of S with respect to the money
market M using the change of numeraire formula for assets X = S, Y = M ,
and Z = $, we get
 R 
T
SM (T ) = S$ (T ) · $M (T ) = exp − 0 r(u)du · S$ (T ) ≤ S$ (T ), (1.24)

with
SM (0) = S$ (0) · $M (0) = S$ (0). (1.25)
Similarly, when we express the price of S with respect to the bond B T using
the change of numeraire formula for assets X = S, Y = M , and Z = $, we
get
SB T (T ) = S$ (T ) · $B T (T ) = S$ (T ), (1.26)
with
S$ (0)
SB T (0) = S$ (0) · $B T (0) = ≥ S$ (0). (1.27)
B$T (0)
The two types of discounting are also related by
SB T (t) = SM (t) · MB T (t). (1.28)
In particular, when the interest rate r is constant, the relation between SB T
and SM is simply
SB T (t) = erT · SM (t). (1.29)
The important difference between SM and SB T is that the price of SM agrees
with the price S$ at time t = 0, while the price of SB T agrees with the price
S$ at time T . The reference point for discounting with the money market M
is at time t = 0, while the reference point for discounting with the bond B T is
at time T . Since typical European-type derivative contracts explained in the
next chapter pay off f (S$ (T )) for some function f , discounting with respect
to the bond B T makes more sense as SB T (T ) = S$ (T ).

Bonds usually deliver units of a currency at multiple times until their matu-
rity. However, without loss of generality we consider only bonds with a single
delivery time T . A bond B T that pays one dollar at time T is also known as a
zero coupon bond. A bond with multiple delivery times is just a combination
of several zero coupon bonds. A zero coupon bond is also a possible choice of
a no-arbitrage reference asset.
20 Stochastic Finance: A Numeraire Approach

1.5 Dividends
It is often the case that a stock S pays dividends, making it an arbitrage
asset. However, the portfolio Se of the stock and the dividends is a no-arbitrage
asset. Let us find the relationship between the dividend-paying stock S and
the asset representing the stock plus dividends S. e

Consider first the situation when the dividends are paid in discrete times
t1 , t2 , . . . , tn . At the time of the first dividend payment t1 , the stock S splits
into two parts; one representing the equity part after the dividend, and one
representing the dividend. At the time t1 − immediately before the dividend
payment, we have
St1 − = Set1 − .
Assuming that the dividend payment is a fraction a(t1 ) ∈ (0, 1) of the stock
S taken at time t1 −, we get
St1 = (1 − a(t1 )) · St1 − = (1 − a(t1 )) · Set1 − = (1 − a(t1 )) · Set1 .
While the value of the equity S jumps down at the time of the dividend
payment, the value of the equity plus dividends Se does not, and thus we have
Set1 − = Set1 .
At the time of the second dividend payment t2 , the dividend is the fraction
a(t2 ) ∈ (0, 1) of the equity part St2 − = (1 − a(t1 )) · Set2 − . Thus the stock S
satisfies
St2 = (1 − a(t1 )) · (1 − a(t2 )) · Set2 .
Continuing this procedure, we conclude that the stock S and the asset repre-
senting the stock plus the dividends Se are related by
" n #
Y
Stn = (1 − a(ti )) · Setn (1.30)
i=1

after n dividend payments.

We can also consider the situation when the stock pays the dividend at
the continuous rate. A standard approach is to assume that the relationship
between the stock S and the asset representing the stock and the dividends
Se is given by
 Z t 
St = exp − a(s)ds · Set , (1.31)
0

or stated equivalently,
dSeS (t) = a(t)SeS (t)dt. (1.32)
The process a(t) represents the dividend yield.
 Elements of Finance 21

1.6 Portfolio
This section addresses the following questions: What is a portfolio? What
is the price of a portfolio? What is a self-financing trading strategy?

A portfolio is a sum of one’s assets

N
X
Pt = ∆i (t) · X i , (1.33)
i=0

where ∆i (t) represents how many units of an asset X i are held at

time t.

When ∆i (t) > 0, we say that the portfolio has a long position in the asset
X . When ∆i (t) < 0, we say that the portfolio has a short position in the
i

asset X i . When ∆i (t) = 0, we say that the portfolio has a neutral position
in the asset X i .

Note that a portfolio is not a number. A car, a house, paintings, and jewelery
are assets that do not take numerical values. Thus a portfolio is a distinct
concept from the price of a portfolio, the number of units of the reference
asset that is required to acquire the entire portfolio. As mentioned earlier,
price is relative to the chosen reference asset. If we fix Y = X 0 to be the
reference asset, the price of a portfolio with respect to the reference asset
(numeraire) Y is given by

N
X
PY (t) = ∆i (t) · XYi (t). (1.34)
i=0

In other words, PY (t) is the number of units of the asset Y that one would
obtain, should one exchange all assets in one’s portfolio for an asset Y at
time t.

The individual portfolio position ∆i (t) has to be known at time t; it cannot

be set in retrospect after observing prices in the future. It is similar to bet-
ting in a casino – one first places the stake before observing the outcome of
a given game. Mathematically, each ∆i (t) has to be a predictable process,
which means that the portfolio position is set before the market observes the
price move. Predictable processes are generated by the processes that have
left continuous paths.
22 Stochastic Finance: A Numeraire Approach

A portfolio, Pt , together with prices XYi (t) determine the price of a portfolio
PY (t). On the other hand, different portfolios may have the same price at
a given time t. We assume that one can exchange one’s portfolio for any
other portfolio that has an equal price at time t. We also assume that all
assets in the portfolio are no-arbitrage assets. This procedure of exchanging
no-arbitrage assets with equal price is known as a self-financing trading
strategy. Trading portfolios with equal prices means that no asset is either
added or withdrawn from the portfolio without being properly exchanged
with a combination of assets of an equal price. Holding only no-arbitrage
assets ensures that the resulting portfolio is also a no-arbitrage asset. If the
prices of two portfolios are the same with respect to one asset Y , the prices
are also the same with respect to any other asset Z. This is easily seen from
the change of numeraire formula

PZ (t) = PY (t) · YZ (t).

Since exchanging portfolios with equal price can be done in principle at any
given time t, one can have continuously rebalanced portfolios as a result.

Let us give an example of self-financing trading.

Example 1.9 Self-financing trading

The portfolio
N
X
Pt1 = ∆i (t) · X i
i=0

can be exchanged for the portfolio

"N #
X
2 i i
Pt = ∆ (t) · XY (t) · Y
i=0

since the two have the same price. This is easily seen from
N
X
PY1 (t) = ∆i (t) · XYi (t),
i=0

and
N
X
PY2 (t) = ∆i (t) · XYi (t).
i=0

Therefore we have
PY1 (t) = PY2 (t).
However, the two portfolios are physically different. The first portfolio Pt1 has
∆it units of an asset X i , for i = 1, . . . , N , while the second portfolio Pt2 has
 Elements of Finance 23
PN
i=0 ∆i (t) · XYi (t) units of an asset Y , and zero positions in the remaining
assets. But since they have the same price, they can be exchanged for each
other at time t.

REMARK 1.3 Note that self-financing trading may come with some
limitations. For instance in the economy consisting of just two assets X and
Y , portfolios of the form
P = ∆X (t) · X + (PY (t) − ∆X (t)XY (t)) · Y
have the same price PY (t) with respect to the reference asset Y , where ∆X (t)
is an arbitrary number. But in reality, one usually cannot take arbitrarily
large or arbitrarily small (negative) positions in the underlying assets. These
positions are usually bounded. For instance, sometimes it may not be possible
to take a short position in a particular asset. The bounds on the portfolio
position may depend on a given situation, and they may even be different
for different agents (think about credit lines). Therefore it is not clear how
to define acceptable portfolio positions in order to reflect the reality of the
market. There can be also a physical limit on the number of assets that can
be held: some assets are nondivisible, and thus one can have only an integer
number of them in a given portfolio.

Another limit is that the price of the portfolio may be required to stay above
a certain minimal threshold; otherwise a bankruptcy occurs. An adapted port-
N
folio process ∆i (t)i=0 that guarantees PY (t) ≥ L for some lower bound L for
all t is called admissible.

The last concern we mention is continuous trading. The traders in the real
markets are allowed to change their portfolio positions rather frequently, but
only finitely many times in a given time interval. However, mathematical
models in continuous time assume that the portfolio positions can be changed
continuously. Such an approach gives realistic results, but one should be care-
ful not to construct portfolios that require an infinite number of trades that
are not the result of a limit of discrete trading.

We will not be specific in this text about these limitations since this is not
a prime focus of the book, but the reader should be aware of them.

1.7 Evolution of a Self-Financing Portfolio

Let us discuss how the portfolio can evolve in time, using a self-financing
trading strategy. We also assume that all assets are no-arbitrage as-
24 Stochastic Finance: A Numeraire Approach

sets; otherwise the portfolio itself is an arbitrage asset. Consider first

the discrete time case. Let the portfolio at time k be given by
N
X
Pk = ∆i (k) · X i .
i=0

At time k + 1, the portfolio will have the same positions ∆ik in each asset X i :
N
X
Pk+1 = ∆i (k) · X i ,
i=0

but since X i stays the same over time for each i = 0, 1, . . . , N , the portfolios
Pk and Pk+1 are the same, only taken at two different time periods.

While the portfolio remains unchanged, its price with respect to a reference
asset may be changing. When we write the difference of the prices of the
portfolio taken at two consecutive times k and k + 1, we get

N
X  
PY (k + 1) − PY (k) = ∆i (k) · XYi (k + 1) − XYi (k) . (1.35)
i=0

Note that we can omit the changes in the reference asset Y = X 0 since

YY (k + 1) − YY (k) = 1 − 1 = 0.

For example, one ounce of gold in the portfolio will still be one ounce of gold in
the portfolio in the next time interval, and its price will stay unchanged if the
reference asset is chosen to be gold. Similarly, a particular asset will remain
the same in the portfolio, but its price with respect to gold may fluctuate in
time.

Equation (1.35) says that the change of the price of the portfolio is explained
only by the changes of the prices of individual assets in the portfolio. On
the other hand, possible changes in the asset positions ∆ik from time k to
k + 1 do not enter this equation. At time k + 1, the holder of the portfolio is
free to exchange his present portfolio for a portfolio that has the same price.
old
If we denote the old portfolio that was inherited from time k by Pk+1 =
PN i i
Pk = i=0 ∆ (k) · X , and the newly exchanged portfolio at time k + 1 by
new
PN
Pk+1 = i=0 ∆i (k + 1) · X i , we have

PYold (k + 1) = PYnew (k + 1).

The holder of the portfolio can change his position in the underlying assets
X i from ∆i (k) to ∆i (k + 1) given that the two portfolios under consideration
 Elements of Finance 25

have the same price. It means that

N
X N
X
∆i (k) · XYi (k + 1) = ∆i (k + 1) · XYi (k + 1),
i=0 i=0

or in other words,

N
X  
∆i (k + 1) − ∆i (k) · XYi (k + 1) = 0. (1.36)
i=0

This is the condition a discretely rebalanced portfolio must satisfy in order to

be self-financing. The above identity can be also expressed as

N h
X  
(∆i (k + 1) − ∆i (k)) · XYi (k + 1) − XYi (k)
i=0
i
+ (∆i (k + 1) − ∆i (k)) · XYi (k) = 0. (1.37)

When we consider continuous time models, the above identities will take
the following forms. For the evolution of the price of the portfolio, we have

N
X
dPY (t) = ∆i (t) · dXYi (t), (1.38)
i=0

a continuous analog of Equation (1.35). Similarly, the identity corresponding

to Equation (1.37) is

N
X  
(d∆i (t)) · dXYi (t) + (d∆i (t)) · XYi (t) = 0. (1.39)
i=0

Indeed, if we applied Ito’s formula for the evolution of the price of the portfolio,
we would get
N
!
X
i
dPY (t) = d ∆ (t) · XYi (t)
i=0
N
X  i 
= ∆ (t) · dXYi (t) + (d∆i (t)) · dXYi (t) + (d∆i (t)) · XYi (t) ,
i=0

But since the last two terms of the above identity sum to zero from (1.39),
we have Equation (1.38).
26 Stochastic Finance: A Numeraire Approach

Example 1.10  
Consider ha portfolio P that
i holds ∆X (t) = 1 − Tt units of an asset X, and
Rt
∆Y (t) = T1 0 XY (s)ds units of an asset Y at time t, where t ∈ [0, T ]. In
other words, h R i
  t
Pt = 1 − Tt · X + T1 0 XY (s)ds · Y. (1.40)
We can show that this is a self-financing portfolio. The condition of self-
financing trading (1.39) reads as

(d∆Y (t)) · dYY (t) + (d∆Y (t)) · YY (t)

+ (d∆X (t)) · dXY (t) + (d∆X (t)) · XY (t) = 0,

where we substituted X 0 = Y , and X 1 = X. Since YY (t) = 1, the above

relationship simplifies to

(d∆Y (t)) + (d∆X (t)) · dXY (t) + (d∆X (t)) · XY (t) = 0. (1.41)

Note that

d∆Y (t) = 1
T XY (t)dt,
X
d∆ (t) = − T1 dt,

and thus

d∆Y (t) + d∆X (t) · dXY (t) + d∆X (t) · XY (t) =

1
= T XY (t)dt + (− T1 dt) · dXY (t) + (− T1 dt)XY (t) = 0.

Therefore we have the self-financing evolution of the prices of the portfolio

from (1.38). When we choose Y to be the reference asset, we have
 
dPY (t) = ∆X (t)dXY (t) = 1 − Tt dXY (t),

when we choose X to be the reference asset, we have

h R i
t
dPX (t) = ∆Y (t)dYX (t) = T1 0 XY (s)ds dYX (t).

hNoteR
that the iportfolio Pt starts with P0 = X0 and ends with PT =
1 T
T 0 XY (s)ds · YT . Therefore the above described self-financing strategy
h R i h R i
T T
delivers T1 0 XY (s)ds units of Y at time T . The number T1 0 XY (s)ds
represents the average price of the asset X in terms of the reference asset Y .

The trading strategy described in Example 1.10 does not depend on the
evolution of the underlying price XY (t). Also, d∆X (t) and d∆Y (t) have only
 Elements of Finance 27

a dt term, so ∆X (t) and ∆Y (t) are smooth. Because of that, the (d∆X (t)) ·
dXY (t) cross term is zero. However, the positions ∆X (t) and ∆Y (t) in the
underlying assets can be even diffusions, such as in the following example. In
that case, the (d∆X (t)) · dXY (t) cross term may not disappear. The reader
should be familiar with stochastic calculus, or return to this example after
reading the Chapter 3 Diffusion Models in order to fully appreciate it.

Example 1.11
Assume that an asset price follows geometric Brownian motion

dXY (t) = σXY (t)dW Y (t),

where X and Y are two no-arbitrage assets. Consider a portfolio Pt which is

given by
Pt = [N (d+ )] · X + [−KN (d− )] · Y,

where
Z x y2
N (x) = √1
2π
· e− 2 dy,
−∞

and  
XY (t)
√
d± = √1 · log ± 21 σ T − t.
σ T −t K

The portfolio P holds ∆X (t) = N (d+ ) units of an asset X, and ∆Y (t) =

−KN (d− ) units of an asset Y . It turns out (Exercise 1.5) that this portfolio
is indeed self-financing. The self-financing condition is given by

d∆Y (t) + d∆X (t) · dXY (t) + d∆X (t) · XY (t) =

= −KdN (d− ) + dN (d+ ) · dXY (t) + dN (d+ ) · XY (t).

It is not trivial to show that

−KdN (d− ) + dN (d+ ) · dXY (t) + dN (d+ ) · XY (t) = 0,

but it is true. Thus we have

dPY (t) = N (d+ ) dXY (t),

and
dPX (t) = −KN (d− ) dYX (t).

The portfolio Pt in this example is in fact a hedging portfolio for a European

option with a payoff (XT − K · YT )+ in a geometric Brownian motion model.
28 Stochastic Finance: A Numeraire Approach

1.8 Fundamental Theorems of Asset Pricing

The general assumption in finance is that the market does not contain arbi-
trage. If an arbitrage opportunity appears, the market usually corrects itself
in a short time period. On the other hand, profitable trading strategies may
exist for long periods. Some profitable trading strategies may even come with
a risk of a catastrophic loss.

Obviously, the entire theory depends upon the fact that the assets in the
portfolio are no-arbitrage assets to start with; otherwise the portfolio is not
arbitrage free. The central result of finance theory is the First Fundamental
Theorem of Asset Pricing:

THEOREM 1.1 First Fundamental Theorem of Asset Pricing

If there exists a probability measure PY such that the price processes XY (t)
are PY -martingales, where X is an arbitrary no-arbitrage asset, and Y is an
arbitrary no-arbitrage asset with a positive price, then there is no arbitrage in
the market.

PROOF Let Y be a fixed reference asset. If there is an arbitrage opportu-

nity, one can start with a zero price portfolio PY (0) = 0 and obtain a portfolio
PY (T ) in the form PY (T ) = ξ(ω), where ξ(ω) is a non-negative random vari-
able with PY (ξ(ω) > 0) > 0. In this case, PY (T ) cannot be a martingale since
EY [PY (T )] > 0 = PY (0).

REMARK 1.4 Market interpretation of PY

The probability measure PY associated with a no-arbitrage reference asset

Y has the following market interpretation. Let A be an event in FT , which
can be viewed as a set of market scenarios ω that satisfy a condition ω ∈ A.
As an example of such an event, consider A = {ω ∈ Ω : XY (T, ω) ≥ K}. This
is a set of scenarios where the market price of X with respect to the reference
asset Y exceeds a fixed constant K at time T . Each set A from the informa-
tion set FT has some objective probability P(A). The probability measure P
is known as the real probability measure. However, the real probability
measure does not play any role in the First Fundamental Theorem of Asset
Pricing, and thus its role in pricing financial contracts is limited.

What is relevant to pricing financial contracts is the probability measure

PY . Imagine that there is a security V that pays off one unit of the asset Y at
 Elements of Finance 29

time T when the scenario ω is in A; otherwise it pays nothing. In mathematical

notation,
VT = IA (ω) · YT ,
where I denotes an indicator function. We can also rewrite the above equation
in terms of the prices as
VY (T ) = IA (ω).
The contract V is known as an Arrow–Debreu security. If we want to find
the price of this contract at time t = 0, we can use the fact that VY (t) is a
martingale under the probability measure PY . Therefore

VY (0) = EY [VY (T )] = EY [IA (ω)] = PY (A).

In terms of the assets, we have

V0 = PY (A) · Y0 .

In other words, PY (A) is the initial market price of the contract V in terms
of the asset Y . Clearly, delivering a unit of Y at time T for a set of scenarios
in A should cost at most a unit of Y at time t = 0. So PY (A) indicates what
fraction of Y is required to start with in order to deliver the Arrow–Debreu
security at time T . The probability PY does not indicate directly how likely
is the event A to occur, but rather how costly it is with respect to the asset Y .

When the number of possible scenarios in Ω is finite, we can consider events

with a single scenario only, meaning A = {ω}. The price corresponding to the
Arrow–Debreu security for this event, PY (ω), is known as an Arrow–Debreu
state price. This concept generalizes to a countable number of states. When
the number of states is not countable, representing a continuous random vari-
able, the Arrow–Debreu state price can be interpreted as a density:
Z
Y
P (A) = dPY (ω).
ω∈A

In this situation, dPY (ω) is known as an Arrow–Debreu state price den-

sity.

Note that the probability measure PX that is associated with a different no-
arbitrage reference asset X is in general different from PY . The corresponding
Arrow–Debreu security U would pay off one unit of an asset X when a scenario
ω is in A; otherwise it would pay nothing. In other words,

UT = IA (ω) · XT .

This contract differs from V only in the underlying asset. The initial price of
U is given by
U0 = PX (A) · X0 .
30 Stochastic Finance: A Numeraire Approach

In general, the fraction PY (A) of the asset Y needed for the security V and
the fraction PX (A) of the asset X needed for the security U will differ.

Consider for instance a geometric Brownian motion model for an asset price


XY (t) = XY (0) · exp σW Y (t) − 21 σ 2 t .
This model will be studied in detail in Chapter 3 Diffusion Models. Brownian
motion and stochastic calculus are also discussed in the Appendix. Let A be
the set of scenarios where the terminal price XY (T ) of the asset ends up below
the initial price of the asset XY (0), or in other words,
A = {ω ∈ Ω : XY (T, ω) ≤ XY (0)}.
Let V be the corresponding Arrow–Debreu security that delivers a unit of
the asset Y at time T when the asset price XY (T ) ends up below XY (0),
and let U be the corresponding Arrow–Debreu security that delivers a unit
of the asset X at time T . It turns out that PY (A) > 21 , but PX (A) < 12 .
While the mathematical details behind this result are discussed in Chapter
3 on Diffusion Models, intuitively this makes sense. Take as an example X
to be a stock market, and Y to be a money market. The set of scenarios in
A represents outcomes when the market makes a downturn with respect to
the reference asset Y . Should one deliver a unit of Y on the downturn, this
happens to cost more than half a unit of Y to start with. But that is not
surprising; when the market takes a downturn, the reference asset Y , such as
the money market in this case, becomes more expensive to deliver. The reason
is that Y has appreciated with respect to X, and thus it takes more than one
half units of Y to cover the payoff of the corresponding Arrow–Debreu security.
On the other hand, it costs less than half a unit of X to deliver a unit of X
on the market downturn. This is also not surprising, since on the downturn,
the asset X becomes less valuable, and cheaper to deliver.

REMARK 1.5 The inverse statement in the First Fundamental Theorem

of Asset Pricing that a no-arbitrage condition implies existence of a martin-
gale measure PY is also true, at least in typical mathematical models. This
means no arbitrage implies that prices are martingales with respect to the
corresponding probability measure. A proper mathematical statement of this
theorem requires a careful definition of an admissible trading strategy. The
interested reader should refer to academic literature on this topic. For practi-
cal purposes, it is enough that we start with a martingale evolution
of the price. Furthermore, martingales in continuous time are just combi-
nations of diffusions and jumps, so no other processes (such as a fractional
Brownian motion for Hurst index 6= 21 ) can be considered for a no-arbitrage
description of the prices.
 Elements of Finance 31

We now consider how to determine the probability measure PY . One should

start with describing the set of possible outcomes Ω that represent the individ-
ual scenarios of a price evolution. One can consider discrete time and discrete
space models, which are known as tree models (binomial or trinomial tree).
Continuous time models can have either continuous paths, which lead to dif-
fusion models, or they can have jumps, which lead to purely discontinuous
models. Note that requiring prices to be martingales limits possible types of
the price evolution.

A general martingale in continuous time can be written as a sum of a

martingale with continuous paths and a purely discontinuous martingale:

M(t) = Mc (t) + Md (t). (1.42)

A martingale Md (t) is called purely discontinuous if its product with any

continuous martingale remains a martingale. For instance, a compensated
Poisson process N (t) − λt is a purely discontinuous martingale. Note that
a purely discontinuous martingale may have continuous paths. Continuous
martingales adapted to a filtration FtW generated by a Brownian motion W
are in fact diffusions; they can be represented as stochastic integrals with
respect to Brownian motion. Thus

Z t
c c
M (t) = M (0) + φ(s)dW (s), (1.43)
0

where φ(t) is adapted to FtW . This result is known as the Martingale Repre-
sentation Theorem (Theorem A.3).

The following example lists some possible martingale evolutions of the price.

Example 1.12 Martingale evolution of the price

Trinomial Model The price XY (0) is assumed to take three possible values
in the next time instant: event A – go up to u · XY (0) (u > 1), event B
– stay the same, or event C – go down to d · XY (0) (d < 1).
32 Stochastic Finance: A Numeraire Approach

XY (1, A) = u · XY (0)

XY (0) XY (1, B) = XY (0)

XY (1, C) = d · XY (0)

When the probabilities of the events A, B and C are given by

1−d u−1
PY,ξ (A) = · ξ, PY,ξ (B) = 1 − ξ, PY,ξ (C) = · ξ. (1.44)
u−d u−d
where ξ ∈ [0, 1], the price process XY (n) is a martingale. Note that
each ξ defines a different probability measure, so in this case there exist
infinitely many martingale measures PY,ξ . One can check that
EY,ξ XY (1) = XY (1, A) · PY,ξ (A) + XY (1, B) · PY,ξ (B)
+XY (1, C) · PY,ξ (C)
1−d
= u · XY (0) · · ξ + XY (0) · (1 − ξ)
u−d
u−1
+d · XY (0) · ·ξ
u−d
= XY (0).
It means that the prices of Arrow–Debreu securities may not be uniquely
defined, meaning that there exists a range of the prices when there is no
arbitrage present. Consider for instance an Arrow–Debreu security that
pays off one unit of Y when the scenario A happens. The initial price of
this security is
1−d
PY,ξ (A) = · ξ,
u−d
1−d
which can be any number in the interval [0, u−d ], depending on the value
of the parameter ξ. The market can quote any price in that interval, and
there would be no arbitrage opportunity. The question is which martin-
gale measure should one use when there is more than one in order to
determine the prices of financial securities? The answer is that it is the
market that chooses the martingale measure. For instance, if the market
quotes the price of the above mentioned Arrow–Debreu security, it al-
ready determines the value of the parameter ξ, thus effectively choosing
only one martingale measure.
 Elements of Finance 33

Binomial Model A binomial model is a special case of a trinomial model

with ξ = 1. The price either goes up to u · XY (0) (u > 1), or goes down
to d · XY (0) (0 < d < 1).

XY (1, H) = u · XY (0)

XY (0)

XY (1, T ) = d · XY (0)

We have a martingale evolution of the price when

1−d u−1
PY (H) = , PY (T ) = . (1.45)
u−d u−d
See Chapter 2 Binomial Models for more details. Note that the martin-
gale measure here is unique.

Geometric Brownian Motion Geometric Brownian motion is a process

that satisfies the following stochastic differential equation:

dXY (t) = σXY (t)dW Y (t). (1.46)

The parameter σ is known as the volatility. The above stochastic differ-

ential equation admits a closed form solution


XY (t) = XY (0) · exp σW Y (t) − 12 σ 2 t , (1.47)

which is a martingale. The market noise process, namely Brownian mo-

tion W Y (t), comes with the reference asset Y and determines the mar-
tingale measure PY . Chapter 3 Diffusion Models studies this model in
detail. More information about Brownian motion and stochastic calculus
is given in the Appendix.

Geometric Poisson Process Geometric Poisson process satisfies the fol-

lowing stochastic differential equation:

dXY (t) = (eγ − 1) · XY (t−)d(N (t) − λY t). (1.48)

34 Stochastic Finance: A Numeraire Approach

The price with the above dynamics is given by

XY (t) = XY (0) exp γ · N (t) − (eγ − 1)λY t . (1.49)

This is also a martingale process. In contrast to a geometric Brownian

motion model, the market noise process N (t) that represents Poisson
jumps does not come with a particular asset. However, different assets
come with different martingale measures, which is captured by the in-
tensity of jumps λY that comes with a particular reference asset Y .
Pricing in the presence of Poisson jumps is studied in Chapter 10 Jump
Models.

Note that when there is more than one asset with a positive price available,
any of them can be used as a reference asset. Consider a situation when both
X and Y are no-arbitrage assets with a positive price, and let V be an arbi-
trary no-arbitrage asset. Then we have that VY (t) is a PY martingale, but also
VX (t) is a PX martingale. The relationship between martingale measures PY
and PX is explained in detail in the following text. It turns out that an im-
portant assumption is that both prices XY (t) and YX (t) stay positive, which
is a reasonable assumption for primary reference assets that are represented
by currencies, stocks, or precious metals. It is possible that even such basic
assets may become worthless, in which case the worthless asset cannot be used
as a numeraire. For instance when X = 0, we still have a well-defined price
XY (t) = 0, but YX (t) is not well defined. Note that derivative contracts can
have in principle any price, they may even take negative values, but in this
case they cannot be used as reference assets.

An example of a situation when we have two assets with a positive price is

a foreign exchange market, where X stands for a domestic bond and Y stands
for a foreign bond. Bonds serve as no-arbitrage proxies to the respective cur-
rencies. However, an asset is domestic relative to a location, and thus Y is a
domestic asset and X is a foreign asset for somebody else. Therefore it makes
sense to consider the price of the foreign asset in terms of the domestic asset
XY (t), and vice versa, the price of the domestic asset in terms of the foreign
asset YX (t).

When the underlying asset is a bond B T with maturity T , the correspond-

ing PT measure is known as a T-forward measure. The term risk-neutral
measure is used when the underlying asset is the money market account M .
We will denote the risk-neutral measure by PM . The risk-neutral measure and
T-forward measure coincide when the interest rate evolution is deterministic.
The reader should note that the natural choice for the pricing measure for
contracts that are settled in money is the T-forward measure which works
also in situations of random interest rates. The risk-neutral measure can be
 Elements of Finance 35

used for pricing such contracts only when the interest rate is deterministic.
There is no martingale measure P$ that would correspond to a dollar as a ref-
erence asset since the dollar is an arbitrage asset. Other no-arbitrage reference
assets have their own martingale measure. When the underlying reference as-
set is a stock S, the corresponding PS measure is known as a stock measure.

The price of an arbitrary no-arbitrage asset V can be computed from the

First Fundamental Theorem of Asset Pricing, which gives us a stochastic
representation of the prices. The theorem states that the prices are martingales
under a proper probability measure, and thus their expected value does not
change with time. We have the following relationship:

VY (t) = EYt [VY (T )] , (1.50)

where V and Y are two no-arbitrage assets. The symbol Et [.] denotes con-
ditional expectation. Rewriting the above relationship in terms of assets, we
get
V = EYt [VY (T )] · Y. (1.51)

This literally means that V is worth EYt [VY (T )] units of Y at time t. Note
that EYt [VY (T )] is an Ft measurable random variable that represents the price
VY (t). Computing this conditional expectation is a key aspect of pricing fi-
nancial contracts. The computation can be done in the following ways: finding
a closed form solution for a particular contract; using Monte Carlo simulation
to estimate the expected value; or by using differential methods to compute
the price as explained later in the text.

REMARK 1.6 Computing dollar prices

The First Fundamental Theorem of Asset Pricing does not apply when a
dollar is used as a reference asset since it is an arbitrage asset. The dollar
prices have to be computed from the change of numeraire formula. Consider
a contingent claim V with a payoff at a fixed maturity T . The claim will pay
V$ (T ) units of a dollar $ at time T . We can use any no-arbitrage asset Y to
compute the price of V using the formula

VY (t) = EYt [VY (T )].

From the change of numeraire formula, we can compute the dollar price of
the contract by
V$ (t) = VY (t) · Y$ (t).
A natural no-arbitrage asset to use is the bond B T that matures at time T .
In this case we can write

VB T (t) = ETt [VB T (T )].

36 Stochastic Finance: A Numeraire Approach

Converting to dollar prices by the change of numeraire formula and using the
fact that B$T (T ) = 1, we can also write

V$ (t) = VB T (t) · B$T (t) = ETt [V$ (T ) · $B T (T )] · B$T (t) = ETt [V$ (T )] · B$T (t).
Thus we have
V$ (t) = B$T (t) · ETt [V$ (T )]. (1.52)
Equation (1.52) is of central importance in the current literature on derivative
pricing. The advantage is that one can immediately obtain the dollar value
of a given contingent claim by using the corresponding T-forward measure.
Note that the interest rate r(t) does not enter the formula. It appears only
indirectly in the price of the bond B$T (t) if we assumed some dependence of
this price on the interest rate. However, such a step is not needed as we can
get the value of B$T (t) directly from the price quoted on the market.

Another possible choice of a no-arbitrage proxy asset to a dollar is the

money market M . We can write
VM (t) = EM
t [VM (T )].

Converting to dollar prices, we get

V$ (t) = VM (t) · M$ (t) = EM
t [V$ (T ) · $M (T )] · M$ (t).

We have already
 seen in Equation (1.16) that M$ (t) is given by M$ (t) =
Rt
exp 0 r(s)ds , and thus the above formula simplifies to
h  R  i
T
V$ (t) = EM
t exp − t r(s)ds · V$ (T ) . (1.53)

Equation (1.53) says that “the price of a contingent claim V is the expected
value of its discounted payoff under the risk-neutral measure.” Some authors
use this equation as a starting point of pricing financial contracts, but this
method can be safely used only in the case of a deterministic interest rate r.
When the interest rate process r(t) is stochastic,
 R which
 is a typical case in
T
real markets, the random variables exp − t r(s)ds and V$ (T ) that show
up in the expectation in (1.53) could be correlated, and the problem of pric-
ing a contingentclaim V would have to address the joint distribution of
RT
exp − t r(s)ds and V$ (T ). This may not be a trivial task, especially when
V itself is an interest rate product.
 R 
T
When the interest rate is deterministic, the discount factor exp − t r(s)ds
is also deterministic and thus independent of the payoff V$ (T ). Thus it can be
factored out from the expectation, and we have
 R 
T
V$ (t) = exp − t r(s)ds · EtM [V$ (T )].
 Elements of Finance 37

However, in the case of a deterministic interest rate we also have B$T (t) =
 R 
T
exp − t r(s)ds , and we can rewrite Equation (1.52) as
 R 
T
V$ (t) = exp − t r(s)ds · EtT [V$ (T )].

This shows that

EtM [V$ (T )] = EtT [V$ (T )]
for an arbitrary claim V , and thus the T-forward measure PT and the risk-
neutral measure PM are the same, but only when the interest rate is deter-
ministic.

When the contingent claim V depends on a stock S, we can also choose S

as the reference asset. Converting the price to dollar values, we obtain
V = ESt [VS (T )] · S = ESt [VS (T )] · S$ (t) · $. (1.54)

Due to the symmetry of the First Fundamental Theorem of Asset Pricing,

a similar formula is valid for the choice of a different reference asset X:
VX (t) = EX
t [VX (T )] , (1.55)
or in other words,
V = EX
t [VX (T )] · X. (1.56)
This means that V is worth EX
t [VX (T )] units of X at time t.

Let us illustrate the concept of X being a reference asset on a trinomial

model. The cases of a binomial model, a geometric Brownian motion model,
and a geometric Poisson model are discussed in detail in the corresponding
chapters.

Example 1.13 Trinomial model with X as a reference asset

Recall that the trinomial model assumes the following evolution of the price
process. The price can take three different values in one time step. When Y
was chosen as a reference asset, the price can go up to XY (1, A) = u · XY (0)
for u > 1 (event A), it can stay the same XY (1, B) = XY (0) (event B), or it
can go down to XY (1, C) = d · XY (0) for 0 < d < 1 (event C). Let us take X
as a reference asset, and let us study the inverse price process YX . On event
A, the price YX (1) is equal to YX (1, A) = u1 · YX (0). This follows from the
relationship between the price XY and its inverse price YX : YX (t) = XY (t)−1 .
When the price XY goes up (such as in the case of event A), the inverse price
YX goes down, and vice versa. On event B, the price YX stays the same:
YX (1, B) = YX (0). On event C, the price YX goes up to YX (1, C) = 1d · YX (0).
38 Stochastic Finance: A Numeraire Approach

1
YX (1, A) = u · YX (0)

YX (0) YX (1, B) = YX (0)

1
YX (1, C) = d · YX (0)

When the probabilities of the events A, B and C are given by

1−d u−1
PX,ξ (A) = u · · ξ, PX,ξ (B) = 1 − ξ, PX,ξ (C) = d · · ξ. (1.57)
u−d u−d
where ξ ∈ [0, 1], the price process YX (n) is a martingale. As in the case of Y
being a reference asset, we get infinitely many martingale measures PX,ξ , one
for each choice of the parameter ξ. One can check that
EX,ξ YX (1) = YX (1, A) · PX,ξ (A) + YX (1, B) · PX,ξ (B) + YX (1, C) · PX,ξ (C)
1−d
= u1 · YX (0) · u · · ξ + YX (0) · (1 − ξ)
u−d
u−1
+ d1 · YX (0) · d · ·ξ
u−d
= YX (0).
The probability measure PX corresponds to Arrow–Debreu securities that use
the asset X as the underlying asset. For instance, a security U that pays off
one unit of an asset X when the event A happens has the initial price
1−d
PX,ξ (A) = u · · ξ.
u−d
The price of U is also not uniquely defined, it can be any number in the
1−d
interval [0, u · u−d ].

We have two possible representations of the price of a contract V : it is either

EYt [VY (T )] units of an asset Y , or EX
t [VX (T )] units of an asset X. This leads
to the following variant of the change of numeraire formula

V = EYt [VY (T )] · Y = EX
t [VX (T )] · X. (1.58)

The reference asset appears in three places in the pricing formula: X – the
reference asset; EX
t – the conditional expectation that is associated with the
 Elements of Finance 39

reference asset; and X – the discount factor in the payoff function. Thus if
one wants to price a contract under a different numeraire Y , one just needs
to replace the formula with Y at these three locations.

Note that the probability measure PY in the change of numeraire formula

(1.58) may not be unique, and the price VY (t) = EYt [VY (T )] of a general con-
tingent claim Y may depend on a particular choice of PY . We have seen this
situation in the trinomial model. Similarly, the probability measure PX may
not be unique, and the price VX (t) = EXt [VX (T )] may depend on a particular
choice of PX . However, to one particular probability measure PY corresponds
one particular probability measure PX that agrees on the prices in the sense
of the change of numeraire formula (1.58). It turns out that the two measures
PY and PX are linked by a Radon–Nikodým derivative as we will show in the
next section.

Example 1.14 One-to-one correspondence of the probability mea-

sures PY and PX in the trinomial model

Let us show a one-to-one correspondence of the probability measures PY and

PX in the trinomial model. Let V be an arbitrary contingent claim. Since the
price VY is a martingale with respect to PY , we can write

VY (0) = EY [VY (1)].

This expectation depends on a particular choice of the probability measure

PY,ξ . When we fix a parameter ξ, we get

VY (0) = EY,ξ [VY (1)].

Note that a different choice of ξ may lead to a different value of VY (0). Ex-
panding the expectation, we can also write

VY (0) = VY (1, A) · PY,ξ (A) + VY (1, B) · PY,ξ (B) + VY (1, C) · PY,ξ (C).

Using the change of numeraire formula VY = VX · XY , the above equality can

be rewritten as

VX (0) · XY (0)
= VX (1, A) · XY (1, A) · PY,ξ (A) + VX (1, B) · XY (1, B) · PY,ξ (B)
+ VX (1, C) · XY (1, C) · PY,ξ (C).
40 Stochastic Finance: A Numeraire Approach

After dividing by XY (0), we can also write

VX (0)
XY (1, A) Y,ξ XY (1, B) Y,ξ
= VX (1, A) · · P (A) + VX (1, B) · · P (B)
XY (0) XY (0)
XY (1, C) Y,ξ
+ VX (1, C) · · P (C). (1.59)
XY (0)

But VX is a martingale under some probability measure PX , and thus we have

VX (0) = EX [VX (1)],

or

VX (0) = VX (1, A) · PX (A) + VX (1, B) · PX (B) + VX (1, C) · PX (C) (1.60)

after expanding the expectation. The prices in (1.59) and (1.60) should agree,
so we must have
XY (1, ω) Y,ξ
PX (ω) = · P (ω). (1.61)
XY (0)
Thus for a particular choice of the martingale measure PY,ξ there is a sin-
gle corresponding measure PX given by (1.61) that gives the same prices of
contingent claims V . Since

XY (1, A) XY (1, B) XY (1, C)

= u, = 1, = d,
XY (0) XY (0) XY (0)

the measure PX is given by

PX (A) = u·PY,ξ (A), PX (B) = PY,ξ (B), PX (C) = d·PY,ξ (C). (1.62)

It turns out that the probability measure PX corresponds to the probability

measure PX,ξ that is given in (1.57). Therefore the price of a contingent claim
V would be the same if computed both under PY,ξ or under PX,ξ for a fixed
parameter ξ. The relationship between PY and PX in a general model is given
by the so-called Radon–Nikodým derivative, and it is studied in the next
section.

REMARK 1.7 All martingale measures PY agree on the price of a

contract to deliver
A given model of a price evolution may come with infinitely many martingale
measures PY , and the price of a general contingent claim may depend on
the choice of the probability measure PY . We have seen this situation in the
trinomial model. However, all martingale measures PY agree on a price of a
contract to deliver a no-arbitrage asset Y . Let us denote this contract by V ,
 Elements of Finance 41

with VT = YT at the delivery time T . Since VY is a martingale, the initial

price VY (0) is given by

VY (0) = EY [VY (T )] = EY [YY (T )] = EY [1] = 1,

and thus V0 = Y0 . This result is independent on the choice of the probability

measure PY . From the change of numeraire formula (1.58), we would get the
same price of the contract V by using any probability measure PX . Similarly,
all martingale measures PX and PY agree on the price of a contract to deliver
a no-arbitrage asset X.

In perfectly symmetric situations when the roles of X and Y can be ex-

changed, it makes sense to study models where XY (t) and its inverse price
YX (t) admit similar evolutions. That would make the observer unable to iden-
tify the reference asset just by looking at the price process. More specifically,
we can consider the situation when the distribution of the price X Y (T )
XY (0) un-
der the probability measure PY has the same distribution as YYXX(T )
(0) under the
probability measure PX . When this is the case, we can also write
   
XY (T ) YX (T )
LY = LX , (1.63)
XY (0) YX (0)

meaning that the laws of the two distributions agree. We will call this princi-
ple the exchangeability of the reference assets. We show in the following
text that it is possible to model the prices of assets in a way that the role of
X and Y can be freely exchanged, for instance in the binomial model or in
the diffusion model.

Another important question is: when is it possible to replicate a contingent

claim V whose payoff depends on underlying assets X i by trading in these
assets? Or in other words, is there a portfolio P of the form
X
P = ∆i (t) · X i

such that Pt = Vt ? We call such a situation a complete market. A market

is incomplete if it is not complete.

THEOREM 1.2 Second Fundamental Theorem of Asset Pricing

A market is complete if and only if the martingale measure PY is unique.

Rule of Thumb: The market is typically complete in situations when the

number of different noise factors does not exceed the number of assets minus
one asset that serves as a numeraire.
42 Stochastic Finance: A Numeraire Approach

Example 1.15 Complete models

Consider a situation when there are just two assets X and Y . The binomial
model has one noise factor which can be thought of as a coin toss, and the
market is complete. Similarly, the market is complete in a geometric Brownian
motion model, where the only source of uncertainty is the Brownian motion.
In the case when the asset price has stochastic volatility, there are two noise
factors (the original Brownian motion, and stochastic volatility), and the mar-
ket is incomplete. Jump models are complete only if the jump size takes one
single value, such as in a geometric Poisson process which represents the only
noise factor. Jump models with multiple jump sizes are incomplete.

Example 1.16 Trinomial model

We have already seen that a trinomial model, with XY (1, A) = u · XY (0),
XY (1, B) = XY (0), and XY (1, C) = d · XY (0), where 0 < d < 1 < u, does
not have a unique probability measure PY .

XY (1, A) = u · XY (0)

XY (0) XY (1, B) = XY (0)

XY (1, C) = d · XY (0)

The price process is a martingale when the probability PY is given by

1−d u−1
PY,ξ (A) = · ξ, PY,ξ (B) = 1 − ξ, PY,ξ (C) = · ξ,
u−d u−d
where ξ ∈ [0, 1]. Here one can think of PY (A) and PY (B) as two sources of
uncertainty (or noise factors). The probability of event C, PY (C) is already
determined since PY (C) = 1 − PY (A)− PY (B). Let us show that this is indeed
an incomplete market. Let V be a contingent claim that pays off VY (1) units
of Y at time T = 1. A hedging portfolio for this claim takes the form
P0 = ∆X (0) · X + ∆Y (0) · Y. (1.64)
If P replicates a contract V , we should have P1 = V1 for all outcomes A, B,
and C. Note that the portfolio P remains unchanged from time t = 0 to time
t = 1, and thus we also have
P1 = ∆X (0) · X + ∆Y (0) · Y.
 Elements of Finance 43

The identity P1 = V1 can also be written in terms of the prices as PY (1) =

VY (1). Thus we have three equations, one for each outcome:

VY (1, A) = ∆X (0) · XY (1, A) + ∆Y (0),

VY (1, B) = ∆X (0) · XY (1, B) + ∆Y (0),
VY (1, C) = ∆X (0) · XY (1, C) + ∆Y (0).

However, we have only two unknowns, ∆X (0), and ∆Y (0) and there is no way
to match all three different values of VY (1) in general. Since P1 = V1 cannot
be satisfied in general, this model is incomplete.

One way to overcome the incompleteness of the model is to consider more

underlying assets that may exist in the real markets, thus completing the
model. Let us assume for instance that the market trades an Arrow–Debreu
security Z that pays one unit of an asset Y when the outcome A happens.
The quote of the price ZY (0) already determines the probability measure PY,ξ
uniquely from the relationship
1−d
ZY (0) = PY,ξ (A) = · ξ,
u−d
and thus
u−d
ξ = ZY (0) · .
1−d
The market becomes complete if we consider a portfolio in the form

P0 = ∆X (0) · X + ∆Y (0) · Y + ∆Z (0) · Z.

At time t = 1, the portfolio P will remain unchanged. In order to match

P1 = V1 for a general claim V , we must have

VY (1, A) = ∆X (0) · XY (1, A) + ∆Y (0) + ∆Z (0),

VY (1, B) = ∆X (0) · XY (1, B) + ∆Y (0),
VY (1, C) = ∆X (0) · XY (1, C) + ∆Y (0),

where we used the fact that ZY (1, ω) = IA (ω). We can always find a solution
for ∆X (0), ∆Y (0), and ∆Z (0) that would match the payoff of the contingent
claim V .

An alternative way to complete the market with other securities is to change

the condition on the hedging portfolio P . Instead of requiring P1 = V1 which
corresponds to a perfect hedge, one may require P1 ≥ V1 which corresponds
to a superhedge. A superhedging portfolio guarantees that the contractual
payoff represented by a claim V is always met, but in some scenarios the re-
sulting portfolio P may have a higher price than V . Unfortunately, it often
happens the the superhedging portfolio P has a substantially higher price
44 Stochastic Finance: A Numeraire Approach

than the actual claim V . Even the superhedging portfolio that has the small-
est initial price PY (0) may give unrealistically high prices. For this reason,
superhedging is almost never used in practice.

A perfect hedge in the assets X and Y is only possible in two notable

situations: either when VY (1) = 1, or when VY (1) = XY (1). The first case
represents a situation when V1 = Y1 , so the payoff is the asset Y itself. In
this case, V becomes a contract to deliver an asset Y , and the corresponding
hedge is ∆X (0) = 0 and ∆Y (0) = 1. All martingale measures PY,ξ do agree
that the initial price of this contract is simply V0 = Y0 . The second case
represents a situation when V1 = X1 , so the payoff is the asset X itself. The
contract V becomes a contract to deliver the asset X, with the initial price
V0 = X0 that is independent of the choice of the martingale measure PY,ξ and
the corresponding hedge is ∆X (0) = 1, ∆Y (0) = 0.

1.9 Change of Measure via Radon–Nikodým Derivative

This section describes the relationship between measures implied by using
a different numeraire. Suppose that X is a no-arbitrage reference asset, Y is
another no-arbitrage reference asset, and V is a contract to be priced. From
the change of numeraire formula, we have

V = EY [VY (T )] · Y = EX [VX (T )] · X. (1.65)

Recall that we may have in principle infinitely many different martingale mea-
sures PY and PX , but the change of numeraire formula links one probability
measure PY with another probability measure PX that agrees with PY on the
same prices for an arbitrary claim V .

The two measures PY and PX can be also related through a scaling factor
Z(T ) in the following sense:

EY [VX (T ) · Z(T )] = EX [VX (T )]. (1.66)

Rewriting this equation in integral form

Z Z
VX (T, ω)Z(T, ω)dPY (ω) = VX (T, ω)dPX (ω)
Ω Ω

which is valid for any integrable random variable VX (T, ω), we get the follow-
ing representation of Z(T ):
dPX
Z(T ) = . (1.67)
dPY
 Elements of Finance 45

In other words,
Z
PX (A) = Z(T, ω)dPY (ω), A ∈ F. (1.68)
A

Intuitively this represents how much one must increase or decrease the weight
placed upon the probability of ω under the PY measure so that one gets the
same answer as if one used the PX measure to start with. The scaling factor Z
is known as the Radon–Nikodým derivative. When the space of outcomes
Ω is discrete, Equation (1.68) can be expressed as

PX (ω) = Z(T, ω) · PY (ω), ω ∈ Ω. (1.69)

We can also consider a reciprocal change of measure

1 dPY
= , (1.70)
Z(T ) dPX
meaning that  
VY (T )
EY [VY (T )] = EX . (1.71)
Z(T )
The Radon–Nikodým derivative has the following financial interpretation.
We can write

EX [VX (T )] · X0 = EY [VX (T ) · Z(T )] · X0 = EY [VY (T )] · Y0 ,

where the first equality results from changing measures, and the second equal-
ity comes from the change of numeraire formula. Since this relationship is valid
for an arbitrary payoff V , we must have

[VX (T ) · Z(T )] · X0 = [VY (T )] · Y0 ,

or
dPX XY (T )
Z(T ) = Y
= . (1.72)
dP XY (0)
We used that
VY (T )
= XY (T ),
VX (T )
which follows from the change of numeraire formula. Note that the Radon–
Nikodým derivative for the reciprocal change of measure is given by

1 dPY YX (T )
= = , (1.73)
Z(T ) dPX YX (0)

which preserves the symmetry between assets X and Y .

46 Stochastic Finance: A Numeraire Approach

REMARK 1.8 Condition for equivalence of the martingale mea-

sures PX and PY
1 Y X
When both Z(T ) and Z(T ) stay positive, the two measures P and P agree
Y
on zero probability events in FT . When P (A) = 0 for A ∈ FT we also have
PX (A) = 0 and vice versa, PX (A) = 0 implies PY (A) = 0. This follows from
the relationships Z
PX (A) = Z(T, ω)dPY (ω),
A
and Z
1
PY (A) = dPX (ω).
A Z(T, ω)
When two probability measures agree on zero probability events in FT , they
are equivalent. Thus the probability measures PY and PX are equivalent
when both prices XY (T ) and YX (T ) stay positive.

REMARK 1.9 The risk-neutral measure PM agrees with the T-

forward measure PT when the interest rate is deterministic
We have already seen that when the interest rate is deterministic, the risk-
neutral measure PM that comes with the money market account M and the T-
forward measure PT that comes with the bond B T that matures at time T give
the same prices of contingent claims. This means that the two measures are
the same. We can also check this result using the Radon–Nikodým derivative
R 
T
dPM MB T (T ) M$ (T ) · $B T (T ) exp 0 r(t)dt · 1
Z(T ) = = = = R  = 1,
dPT MB T (0) M$ (0) · $B T (0) 1 · exp
T
r(t)dt
0
(1.74)
which implies that
Z Z
PM (A) = Z(T, ω)dPT (ω) = 1dPT (ω) = PT (A), A ∈ F. (1.75)
A A

Therefore the two martingale measures PT and PM are the same when the
interest rate is deterministic. When the interest rate is stochastic, the Radon–
Nikodým derivative becomes

dPM M T (T ) M (T ) · $B T (T )
Z(T ) = = B = $
dPT MB T (0) M$ (0) · $B T (0)
R 
T
exp 0 r(t)dt · 1 R 
T
= = exp 0 r(t)dt · B$T (0),
1 · $B T (0)
which is no longer one, and the relationship between the risk-neutral measure
PM and the T-forward measure PT is no longer trivial. We study the rela-
tionship of PT and PM in more detail in Chapter 4 Interest Rate Contracts.
 Elements of Finance 47

REMARK 1.10 Radon–Nikodým derivative for conditional expec-

tations
The Radon–Nikodým derivative as described in the above text corresponds to
changing the measure at time t = 0. However, we can generalize this concept
to any time t ≤ T . From the change of numeraire formula, we have

EX Y Y
t [VX (T )] · Xt = Et [VY (T )] · Yt = Et [VX (T ) · XY (T )] · Yt .

This can be rewritten as

   
XY (T ) Z(T )
EX
t [VX (T )] = EYt VX (T ) · Y
= Et VX (T ) · .
XY (t) Z(t)
Therefore we have
1
EX
t [VX (T )] = · EYt [VX (T ) · Z(T )] (1.76)
Z(t)

in terms of the original Radon–Nikodým derivative Z. This relationship is

known as the Bayes formula.

REMARK 1.11 European call option

A European call option is a contract that pays off (XT − K · YT )+ at
maturity time T , where K is a constant defined by the contract and is known
as the strike. Let us denote the European call option contract as V . We
can assume that both assets X and Y are no-arbitrage assets. If not, we can
consider corresponding no-arbitrage assets that deliver a unit of an asset X,
or a unit of an asset Y respectively, at time T . We can rewrite the option
payoff as

VT = (XT − K · YT )+ = I (XY (T ) ≥ K) · X − K · I (XY (T ) ≥ K) · Y.

The above expression suggests that a European option is simply a combination

of two Arrow–Debreu securities, one that pays off a unit of an asset X when
XY (T ) ≥ K, and one that pays off K units of an asset Y on the same event
when XY (T ) ≥ K. We have already seen in Remark 1.4 that the initial value
of the Arrow–Debreu security that pays off a unit of an asset X when event A
happens is PX (A) units of an asset X. Similarly, the initial value of the Arrow–
Debreu security that pays off a unit of an asset Y when event A happens is
PY (A) units of an asset Y . If we consider A to be event XY (T ) ≥ K, the
value of the European call option at time t is simply

Vt = PX Y
t (XY (T ) ≥ K) · X − K · Pt (XY (T ) ≥ K) · Y. (1.77)

The above relationship is known as the Black–Scholes formula. Note that

deriving the Black–Scholes formula in this form does not require any compu-
tation. The question how to determine the probabilities PX
t (XY (T ) ≥ K) and
48 Stochastic Finance: A Numeraire Approach

PYt (XY (T ) ≥ K) more explicitly for more specific martingale models of the
price evolution is the subject of following chapters.

Note that the choice of the probability measure PY in situations when

there is more than one such measure already determines the corresponding
probability measure PX , and vice versa. The two probability measures must
agree on the prices of all contingent claims, and thus they are related by the
Radon–Nikodým derivative. This follows from
 
V = EYt (X − K · Y )+ Y (T ) · Y
= EYt [XY (T ) · I(XY (T ) ≥ K)] · Y − EYt [K · YY (T ) · I(XY (T ) ≥ K)] · Y
= EX Y
t [XX (T ) · I(XY (t) ≥ K)] · X − K · Pt (XY (T ) ≥ K) · Y
= PX Y
t (XY (T ) ≥ K) · X − K · Pt (XY (T ) ≥ K) · Y,

where we have used the change of numeraire formula

EYt [XY (T ) · I(XY (T ) ≥ K)] · Y = EX

t [XX (T ) · I(XY (T ) ≥ K)] · X.

This shows that the probability measures PX and PY are indeed linked by the
Radon–Nikodým derivative.

1.10 Leverage: Forwards and Futures

Leverage is one of the most important concepts of finance. It allows in-
vestors to magnify their positions in the underlying assets. Let us consider a
situation when an investor believes that the price XY of a specific asset X
will appreciate in the near future. A straightforward way how to realize the
potential profit is to buy the asset X now, and sell it at some subsequent time
T . The result of this trading is summarized in Table 1.2. At time t = 0, the
investor has one unit of an asset X that costs him XY (0) units of an asset Y .
At time t = T , the asset X is sold for XY (T ) units of an asset Y . Therefore at
time t = T , the position in the asset X is zero, and the position in the asset Y
is XY (T )−XY (0). The net profit or loss of this trading is thus XY (T )−XY (0)
units of an asset Y . When XY (T ) − XY (0) is positive, this trade results in a
net profit, when XY (T ) − XY (0) is negative, this trade results in a net loss.

There is an alternative way to realize this profit or loss by trading in con-

tracts to deliver. Instead of buying the asset X at time t = 0, one can buy a
contract U that delivers the asset X at time T , and pay for it in terms of a
contract V that delivers the asset Y at time T . Consider first the case that X
and Y are both no-arbitrage assets. We have seen that a contract to deliver
a no-arbitrage asset agrees with the asset itself at all times, so Ut = Xt , and
 Elements of Finance 49
TABLE 1.2: Trading in the asset X.
Time t = 0 Time t = T
Asset X 1 0
Asset Y −XY (0) XY (T ) − XY (0)

Vt = Yt . Thus it may not be obvious why this approach gives any advantage
over the case when the investor trades in the primary assets X and Y . Table
1.3 shows the positions in the assets U , V , X and Y . Note that at time t = 0,
the investor has zero positions in the assets X and Y . The major advantage
in this trade is that the investor does not need to have a short position in
the reference asset Y . The choice of Y is typically a money market account.
In contrast to the previous case, the investors do not need to decrease their
position in the money market by paying XY (0) units of an asset Y for a unit
of an asset X.

TABLE 1.3: Trading in the contracts to deliver U and

V.
Time t = 0 Delivery, t = T Sale of X, t = T
Asset U 1 0 0
Asset V −XY (0) 0 0
Asset X 0 1 0
Asset Y 0 −XY (0) XY (T ) − XY (0)

The contract U delivers a unit of an asset X at time T . Similarly, XY (0)

units of the contract V delivers the corresponding number of units of an asset
Y to the counter party of this trade at time T . Furthermore, the holder of the
asset X may immediately sell it for XY (T ) units of an asset Y , resulting in
the net profit or loss of XY (T ) − XY (0) units of an asset Y . This is the same
as in the case when the asset X was bought at time t = 0, and sold at time
T.

Developing this idea even further, one can introduce a contract that pays
off one unit of an asset X for K units of an asset Y at time T :

FT = XT − K · YT . (1.78)

The contract F is known as a forward. When X and Y are no-arbitrage

assets, the price of the forward contract is given by

FY (t) = EYt [XY (T ) − K · YY (T )]

= EYt [XY (T ) − K · YY (T )] = XY (t) − K. (1.79)
50 Stochastic Finance: A Numeraire Approach

Thus we have
Ft = Xt − K · Yt
at all times t ≤ T . More generally, the forward can be written as
Ft = Ut − K · Vt ,
where U is a contract that delivers a unit of an asset X, and V is a contract
that delivers a unit of an asset Y . This relationship is valid in both cases when
assets X and Y are arbitrage or no-arbitrage assets.

The forward price For(t, T ) is the value of K that makes the forward
contract F have zero price at time t. It is obvious that
For(t, T ) = XY (t) (1.80)
when X and Y are no-arbitrage assets. Table 1.4 shows that one receives
XY (T ) − XY (0) units of an asset Y at time T by buying a forward contract
F . The forward contract itself has a zero price at time t = 0, and entering
this contract does not require any change of positions in the assets X and Y .
Since the price of the forward contract F is zero, one can potentially enter an
unlimited number of forward contracts at a given time. Although the forward
contract should formally deliver a unit of the asset X, it is still typically set-
tled entirely in the asset Y . Thus the number of the forward contracts may
exceed the total supply of the asset X. This is indeed the case for many typ-
ical assets. For instance there are many more contracts to deliver gold or oil
than is physically available. However, these contracts are typically settled in
money; the asset itself is delivered only in rare cases.

TABLE 1.4: Trading in the forward

contract F .
Time t = 0 Time t = T
Asset F 1 0
Asset X 0 0
Asset Y 0 XY (T ) − XY (0)

Obviously, entering a huge number of forward contracts comes with a sig-

nificant risk of a bankruptcy. The contractual payoff XY (T ) − XY (0) at time
T can be both positive or negative, and having a substantial number of such
contracts may lead to a significant gain, or to a significant loss. In order to
prevent the situation that one of the contractual parties fails to meet its obli-
gations, one can split the payoff XY (T )−XY (0) into a series of daily payments
that reflect the change of the price of the forward contract.
 Elements of Finance 51

Splitting the payoff into a series of payments is done in the following way.
Let 0 = t0 < t1 < · · · < tn = T be the times of the payments. One can think
about them as days if the payments come on a daily basis. At time t0 = 0,
one enters a forward contract F t0 = X − XY (t0 ) · Y that has a zero price. At
time t1 , the price of F t0 will change to

FYt0 (t1 ) = EYt1 [XY (T ) − XY (t0 )] = XY (t1 ) − XY (t0 ).

In order to make F t0 a zero price contract at time t1 , one should subtract

XY (t1 ) − XY (t0 ) units of the asset Y from it. This technically creates a new
forward contract F t1 that has a zero price at time t1 . The relationship between
F t1 and F t0 is the following

F t1 = F t0 − [XY (t1 ) − XY (t0 )] · Y

= X − XY (t0 ) · Y − XY (t1 ) · Y + XY (t0 ) · Y = X − XY (t1 ) · Y.

One can continue this procedure for other times tk . Table 1.5 shows the result
of this procedure between times tk−1 and tk .

TABLE 1.5: Splitting the payments.

Time t = tk−1 Time t = tk
Asset F tk−1 1 0
Asset F tk 0 1
Asset X 0 0
Asset Y 0 XY (tk ) − XY (tk−1 )

In contrast to the forward contract, this procedure does not wait until its
expiration T , but rather settles the changes of the price of the forward contract
daily. The forward contract F tk−1 from the previous time tk−1 is replaced by
a new forward contract F tk at time tk so that F tk has a zero price. The
difference between the prices of F tk−1 and F tk is settled in the asset Y . At
the end of this procedure, one would collect
n
X
[XY (tk ) − XY (tk−1 )] = XY (T ) − XY (0)
k=1

units of an asset Y . Splitting the payments is a principle of a contract known

as futures. A futures contract is defined as a series of payments

n
X
[Fut(tk , T ) − Fut(tk−1 , T )] · Ytk (1.81)
k=1
52 Stochastic Finance: A Numeraire Approach

that are settled in the asset Y at the corresponding times tk . The futures
price Fut(tm , T ) is a number that makes the series of the remaining payments
n
X
[Fut(tk , T ) − Fut(tk−1 , T )] · Ytk
k=m+1

have a zero price at time tm . At time t = T , Fut(T, T ) agrees with the price
XY (T ), the number of units of an asset Y required to obtain a unit of an asset
X.

Let us determine Fut(tm , T ) when X and Y are two no-arbitrage assets. At

time tn−1 , the futures contract has only one payment left, namely

[Fut(T, T ) − Fut(tn−1 , T )] · YT = [XY (T ) − Fut(tn−1 , T )] · YT .

If the price of this contract be zero at time tn−1 , we must have

0 = EYtn−1 [XY (T ) − Fut(tn−1 , T )] = XY (tn−1 ) − Fut(tn−1 , T )

from the martingale property of XY (t). We conclude that

Fut(tn−1 , T ) = XY (tn−1 ).

Repeating this argument, we obtain

Fut(tm , T ) = XY (tm )

at all times tk . Thus in the case when both assets X and Y are no-arbitrage
assets, the forward and the futures price agree:

Fut(t, T ) = For(t, T ) = XY (t),

and futures is the same as the forward contract. However, by splitting the
payments, one minimizes the default risk of the counter party.

One can avoid the counter party risk completely by trading such contracts
on an exchange. Members of the exchange are required to deposit enough
funds to cover for all their potential losses that may happen within one day.
This deposit is known as a margin account. When the funds in the margin
account become critically low, the member receives a margin call, a request
to add more funds. If the member fails to do so, his positions are closed. Clos-
ing the existing positions does not cost anything as the prices of the futures
contracts are set to zero continuously.

The most typical futures contracts are settled in currencies, rather than in
a no-arbitrage asset. It slightly changes the situation since we also need to
take into account the time value of money. Let us assume that the asset X
 Elements of Finance 53

is a stock S, and the asset Y is a dollar $. The futures contract is defined in

this case as a series of payments of the following form
n
X
[Fut(tk , T ) − Fut(tk−1 , T )] · $tk . (1.82)
k=1

Fut(tm , T ) is the value that makes the price of the remaining payments
n
X
[Fut(tk , T ) − Fut(tk−1 , T )] · $tk .
k=m+1

to be zero at time tm . Equation (1.82) is written in terms of an arbitrage asset

$. However, the investor would immediately convert the dollar position into a
position in the money market M . Assume that the price of the money market
M$ (0) starts at one, so we have M$ (0) = 1. From the relationship

M$ (tk ) · $tk = Mtk ,

we can write
1
$tk = · Mtk .
M$ (tk )
The dollar $ at time tk can be exchanged for M$1(tk ) number of units of the
money market M . Thus the payoff of the futures contract can be reexpressed
as
Xn
1
[Fut(tk , T ) − Fut(tk−1 , T )] · · Mtk . (1.83)
M$ (tk )
k=1

Note that this makes the money market M a natural reference asset for com-
puting the price of the futures contract. Let us determine Fut(tm , T ). At the
terminal time tn = T , Fut(T, T ) agrees with the dollar price of the stock
S$ (T ). At time tn−1 the futures contract has only a single payment

1 1
[Fut(T, T )− Fut(tn−1 , T )]· ·MT = [S$ (T )− Fut(tn−1 , T )]· ·MT .
M$ (T ) M$ (T )

Should the price of this payment be zero at time tn−1 , we must have
 
1
0= EM
tn−1 [S$ (T ) − Fut(tn−1 , T )] ·
M$ (T )
1 h i
= · EM [S
tn−1 $ (T )] − Fut(t n−1 , T ) .
M$ (T )

We have used the fact that the price of the money market account M$ (T )
is already known at the prior time tn−1 . The reason is that the interest rate
that corresponds to the time interval [tn−1 , tn ] is set at time tn−1 , so the
54 Stochastic Finance: A Numeraire Approach

investor knows the price M$ (tn ) of the money market account one period
ahead. Therefore
Fut(tn−1 , T ) = EM
tn−1 [S$ (T )].

Repeating this procedure for the previous times t, we get

Fut(t, T ) = EM
t [S$ (T )]. (1.84)

Let us compare the futures price Fut(t, T ) with For(t, T ), the price of the
corresponding forward contract. The forward contract F when written on a
stock S and a dollar $ pays off

FT = ST − K · $T .

This payoff can be rewritten in terms of a bond B T that delivers a dollar $

at time T as
FT = ST − K · BTT .

The forward price is a number For(t, T ) that corresponds to a choice of K

such that the price of the forward contract F is zero at time T . Thus For(t, T )
satisfies the equation

0 = ETt [SB T (T ) − For(t, T )].

The natural choice of the reference asset is a bond B T . Solving for For(t, T ),
we get

For(t, T ) = ETt [S$ (T )]. (1.85)

We used a simple relationship S$ (T ) = SB T (T ) · B$T (T ) = SB T (T ).

Both the futures price Fut(t, T ) and the forward price For(t, T ) are expecta-
tions of the terminal price of the stock S$ (T ), but under different probability
measures. The futures price is computed under the risk-neutral measure PM ,
while the forward price is computed under the T-forward measure PT . We
have already seen that when the interest rate r(t) is deterministic, the two
measures agree: PM = PT . In this case, the futures price and the forward price
agree.

When the interest rate r(t) is stochastic, the two measures PM and PT are
in general different, and the futures price may be different from the forward
 Elements of Finance 55

price. Let us compute the difference between them:

Fut(0, T ) − For(0, T ) = (1.86)
M T
= E S$ (T ) − E S$ (T )
 
T MB T (T ) ET MB T (T )
= E S$ (T ) · − · ET S$ (T )
MB T (0) MB T (0)
h i
= B$T (0) ET [S$ (T ) · M$ (T )] − ET [S$ (T )] · ET [M$ (T )]
= B$T (0) · covT (S$ (T ), M$ (T ))
 R 
T
= B$T (0) · covT S$ (T ), exp 0 r(t)dt .

Thus the difference between Fut(0, T ) and For(0, T ) is proportional to the

covariance between the stock price S$ (T ) and the price of the money market
account M$ (T ). The covariance is computed in the T-forward measure PT
that corresponds to the bond B T as choice of the reference asset. The price
of the money market M$ (T ) is directly related to the interest rate r(t): the
higher is the interest rate, the higher is the price of the money market.

When the covariance between S$ (T ) and M$ (T ) is positive, the futures price

is higher than the forward price. This can be explained by the following argu-
ment. In the scenarios when the stock price S$ (T ) ends up above the initial
stock price S$ (0), the corresponding price of the money market M$ (T ) will
also tend to increase more than in the scenarios when the stock price S$ (T )
ends up lower than S$ (0). This follows from the positive correlation of S$ (T )
and M$ (T ). When the price of the stock goes up, the holder of the futures
contract will be receiving a positive cash flow, and this cash flow will tend to
earn a higher interest rate r(t) on those scenarios. On the other hand, when
the stock goes down, the holder of the futures contract will be receiving a
negative cash flow, and this cash flow will tend to earn a lower interest rate
r(t) on those scenarios. The fact that the resulting cash flow from the futures
contracts earns a favorable interest means that the futures price should be
higher than the corresponding forward price. In contrast to the futures con-
tract, the forward contract is settled in one single payment at its maturity
time, and thus it cannot benefit from varying interest rate.

The reader should check (Exercise 1.7) that the difference between the
futures price and the forward price can also be expressed as
Fut(0, T ) − For(0, T ) = −B$T (0) · covM (S$ (T ), M$1(T ) )
  R 
T
= −B$T (0) · covM S$ (T ), exp − 0 r(t)dt

if we use the risk-neutral measure PM that corresponds to the money market

M as a reference asset. The idea is to follow the computation in (1.86), but
apply the change of measure from PT to PM in the third line of the equation.
56 Stochastic Finance: A Numeraire Approach

References and Further Reading

The concept of Arrow–Debreu securities traces back to Arrow and Debreu

(1954). The idea of pricing financial securities by a no-arbitrage argument was
already present in the papers of Black and Scholes (1973) and Merton (1973).
However, the theory of pricing under the martingale measure was fully de-
veloped only in later papers of Harrison and Kreps (1979) and Harrison and
Pliska (1981). Different formulations of the absence of arbitrage and the exis-
tence of a martingale measure appear in Delbaen and Schachermayer (1996).
An extensive survey of the mathematics of arbitrage appears in the mono-
graph Delbaen and Schachermayer (2006).

The fact that one may use different reference assets for pricing appeared
early in the relevant literature. Margrabe (1978) was the first to use a stock
measure for pricing an exchange option, a contract written on two stocks.
Jamshidian (1989) used bonds as a numeraire in pricing problems of the
fixed income markets, introducing the T-forward measure. A more system-
atic theory of the change of numeraire was developed in Geman et al. (1995).
The change of numeraire is now a mainstream technique used in the fi-
nance theory, as illustrated in the papers of Gourieroux et al. (1998), Long
(1990), Papell and Theodoridis (2001), Brekke (1997), Flemming et al. (1977),
Schroder (1999), Platen (2006), Johansson (1998), Karatzas and Kardaras
(2007), Platen (2004), and Filipovic (2008). The distinction between the for-
ward and futures contracts was pointed out by Margrabe (1976) and Black
(1976).

There are many additional books on quantitative finance that may be useful
for the reader who is interested in a more thorough study of the field. An
overview of many financial products is given in Hull (2008), which also serves
as an introduction to option pricing for practitioners. The book by Baxter and
Rennie (1996) serves as a very intuitive introduction to contingent pricing
in continuous time. An incomplete list of quantitative finance monographs
includes Shreve (2004b), Merton (1992), Bjork (2004), Musiela and Rutkowski
(2008), Duffie (2001), Dana and Jeanblanc (2007), Jeanblanc et al. (2009),
Karatzas and Shreve (2001), Shiryaev (1999), Joshi (2008), Wilmott (2006),
Cerny (2009), or Neftci (2008).
 Elements of Finance 57

Exercises
1.1 Assume that the change of numeraire formula does not hold at time t,
and we have XY (t) < XZ (t) · ZY (t). Show how to make a risk-free profit by
trading in assets X, Y , and Z.

1.2 Consider a dividend paying stock S in a continuous dividend payment

model given by Equation (1.32). Assume that the price the asset representing
the stock plus the dividends Se with respect to the money market M follows
geometric Brownian motion

dSeM (t) = σ SeM (t)dW M (t).

Determine
dSM (t).
Hint: Use SM (t) = SSe(t) · SeM (t) and apply the product rule (Remark A.4).

1.3 Let X and Y be two no-arbitrage assets. Determine whether the follow-
ing portfolios are self-financing or not:
(a) Pt = [max
h R 0≤s≤t XY (s)]i · Yt (portfolio representing the running maximum)
1 t
(b) Pt = t 0 XY (s)ds · Yt (portfolio representing the running average).

1.4 Show that the price of a discretely rebalanced portfolio Pk =

P N i i Y
i=1 ∆ (k) · X in terms of the reference asset Y is a P martingale, where
i
X are no-arbitrage assets.

1.5 Assume that the asset price follows geometric Brownian motion

dXY (t) = σXY (t)dW Y (t),

where X and Y are two no-arbitrage assets. Show that a portfolio Pt which
is given by
Pt = N (d+ ) · X − KN (d− ) · Y,
where Z x y2
N (x) = √1
2π
· e− 2 dy,
−∞
and   √
XY (t)
d± = √1 · log ± 21 σ T − t,
σ T −t K

is self-financing. Show that the portfolio Pt is self-financing when

XY (t)dN (d+ ) + dXY (t)dN (d+ ) − KdN (d− ) = 0,

and prove this relationship using the Ito’s formula.

58 Stochastic Finance: A Numeraire Approach

1.6 Assume a geometric Brownian motion model for XY (t). Determine which
of the following portfolios are self-financing:
(a) Pt = Nh (d− ) · Y. i h i
(b) Pt = φ(d− ) · σ√1T −t · YX (t) · X + N (d− ) − φ(d− ) · σ√1T −t · Y.
h (d+ ) · X.
(c) Pt = N i h i
(d) Pt = N (d+ ) + φ(d+ ) · √1 · X + −φ(d+ ) · √1 · XY (t) · Y.
σ T −t σ T −t

1.7 Show that the difference between the futures price and the forward price
satisfy

Fut(0, T ) − For(0, T ) = −B$T (0) · covM (S$ (T ), M$1(T ) )

  R 
T
= −B$T (0) · covM S$ (T ), exp − 0 r(t)dt .
Chapter 2
Binomial Models

Binomial models for the price evolution assume that the price of an asset X
in terms of a reference asset Y in the next time instant will take only two pos-
sible values – an uptick or a downtick. These models are typically too simple
to capture market reality. The main reason to include them in this book is to
illustrate the fundamental concepts of derivative pricing in a simple model.
We will later extend our analysis to more complex models, in particular to
diffusions, and to models with jumps. In this chapter we show how to ap-
ply the First Fundamental Theorem of Asset Pricing on contracts written on
two assets X and Y . Both assets X and Y can be used as reference assets
for pricing European option contracts. We show how the pricing martingale
measures that come with the assets X and Y are related, using both the ba-
sic martingale principles, and their relationship through the Radon–Nikodým
derivative. The approach of using both reference assets X and Y is novel; most
of the current literature uses only one reference asset, typically represented
by a money market account, to price derivative contracts. In particular, we
give two alternative characterizations of the price of a contingent claim using
both reference assets.

This chapter has two main parts. The first part deals with pricing models
with no-arbitrage assets X and Y . This applies to pricing all European-type
options. Even when a European option is written on one or two arbitrage
assets, such as in the case of stock options, or on foreign exchange options,
it is always possible to substitute the arbitrage asset with the corresponding
contract to deliver, which is itself a no-arbitrage asset. For example, a stock
option is written on a stock and on a dollar, and the dollar can be substituted
with a zero coupon bond. Thus we have a martingale evolution of the price to
start with, and we can show how to price and hedge European-type contracts
with respect to both assets X and Y . We also illustrate via the contract rep-
resenting the average asset that any no-arbitrage asset has its own martingale
measure.

The second part of this chapter studies the case when one of the assets is an
arbitrage asset and the contract is an American option. An American option is
similar to the European option, but the payoff can be collected at any time up
to the maturity of the contract. In this case the arbitrage asset that enters a
given American contract cannot be substituted with a no-arbitrage asset. This

59
60 Stochastic Finance: A Numeraire Approach

happens, for instance, for American stock options whose underlying assets are
a stock S and the dollar $. Thus we also have to consider a no-arbitrage proxy
asset for the dollar $ that is used for hedging such an option, a corresponding
zero coupon bond B T . Therefore American option pricing uses three assets:
S, B T and $.

2.1 Binomial Model for No-Arbitrage Assets

This section applies to a European-type contract V when the payoff is either
VY (T ) units of an asset Y, or equivalently when the payoff is VX (T ) units of an
asset X. The prices VY and VX with respect to two different reference assets
are linked by the change of numeraire formula VY (t) = VX (t)·XY (t). We have
already seen that we can assume that both X and Y are no-arbitrage assets.
If not, we can replace the arbitrage asset with a corresponding no-arbitrage
asset that delivers X or Y at time T . Thus we can directly apply the First
Fundamental Theorem of Asset Pricing, using both assets X and Y . Then we
have
V = EYt [VY (T )] · Y (2.1)
if we use Y as a reference asset Y , and

V = EX
t [VX (T )] · X (2.2)

if we use X as a reference asset for all 0 ≤ t ≤ T . This section determines the

probability measures PY and PX and the hedging portfolio for a general Eu-
ropean option. Both measures PY and PX are relevant for pricing derivative
contracts. For instance the Black–Scholes formula (1.77) is expressed in terms
of the two martingale measures that correspond to assets X and Y .

When the contract is a European stock option, the two no-arbitrage assets
are a stock S and a bond B T . Most of the current literature uses a dollar
$ instead of the bond B T , but this only increases the dimensionality of the
problem to three assets, a step that is not necessary for pricing European
options. An option must be hedged in no-arbitrage assets only, which are the
stock S and the bond B T in this case, so the dollar $ becomes an extra asset.
The dollar prices of derivative contracts can be obtained simply by the change
of numeraire formula from their prices expressed in terms of the stock S or the
bond B T as opposed to computing them from the pricing model that involves
all three assets S, B T , and $.

Consideration of all three assets S, B T and $ is needed only for American

options, and such a model will be described in the following section. The
reason is that the intrinsic value of the American option is expressed in terms
 Binomial Models 61

of the dollar $, and it cannot be replaced by one corresponding no-arbitrage

proxy asset.

2.1.1 One-Step Model

Consider the following one step model for the evolution of an asset price
where X and Y are two no-arbitrage assets. At time n = 0, the price is
XY (0). At time n = 1, the price is either at XY (1, H) = u · XY (0), or at
XY (1, T ) = d · XY (0), where u > d.

XY (1, H) = u · XY (0)

XY (0)

XY (1, T ) = d · XY (0)

An obvious no-arbitrage condition is given by

0 < d < 1 < u.

Given that there is no arbitrage, there is a probability measure PY associated

with this model such that the price process XY (t) is a PY -martingale. This
is a direct consequence of the First Fundamental Theorem of Asset Pricing.
The probability measure PY is determined by pY (u, d) = PY (H) and depends
on the parameters u and d. Using the martingale property, we must have

XY (0) = EY [XY (1)] = pY (u, d) · XY (1, H) + (1 − pY (u, d)) · XY (1, T )

= pY (u, d) · u · XY (0) + (1 − pY (u, d)) · d · XY (0).

Therefore

1−d u−1
PY (H) = pY (u, d) = , PY (T ) = q Y (u, d) = , (2.3)
u−d u−d

where q Y = 1 − pY = PY (T ). When 0 < d < 1 < u, then clearly 0 < pY < 1,

so the corresponding PY measure is well defined.
62 Stochastic Finance: A Numeraire Approach

If we want to compute the price of the contingent claim V using Y as a

reference asset, we can use the following formula

VY (0) = EY [VY (1)] = VY (1, H) · PY (H) + VY (1, T ) · PY (T ) (2.4)

1−d u−1
= VY (1, H) · + VY (1, T ) · . (2.5)
u−d u−d

Recall that the probability PY (A) of an event A does not represent the real
odds of this event P(A), but rather how costly the event is in terms of the
asset Y . Consider for instance an Arrow–Debreu security V that pays off one
unit of Y when ω = H:

V1 = I(ω = H) · Y1 . (2.6)

The price of this security is

VY (0) = EY [VY (1)] = EY [I(ω = H) · YY (1)] = PY (H), (2.7)

where we use the fact that the price process VY (t) is a PY martingale. In order
to deliver an asset Y on the event ω = H, one should charge PY (H) units of
Y to start. Exercise 2.1 asks for one to construct a hedging portfolio for this
contract.

Because of the symmetry in the statement of the First Fundamental Theo-

rem of Asset Pricing, we can also use X as a reference asset. In particular, the
no arbitrage condition is equivalent to the existence of a probability measure
PX such that the price process YX (t) is a martingale.

Let us study the inverse price YX in the binomial model. Assuming the
same dynamics as above,

1 1
YX (1, H) = = u · YX (0).
XY (1, H)

This is a down tick for YX , but an up tick for XY . Similarly,

1 1
YX (1, T ) = = d · YX (0),
XY (1, T )

which is an up tick for YX , but a down tick for XY .

 Binomial Models 63

1
YX (1, H) = u · YX (0)

YX (0)

1
YX (1, T ) = d · YX (0)

The reference asset X will imply a different probability measure PX that

is determined by pX (u, d) = PX (H). This probability also depends on the
parameters u and d. We can determine pX (u, d) from the martingale property

YX (0) = EX [YX (1)] = pX (u, d) · YX (1, H) + (1 − pX (u, d)) · YX (1, T )

= pX (u, d) · 1
u · YX (0) + (1 − pX (u, d)) · 1
d · YX (0).

Solving for pX and q X = 1 − pX , we get

1−d u−1
PX (H) = pX (u, d) = u · , PX (T ) = q X (u, d) = d · . (2.8)
u−d u−d

Note that the up tick factor u in XY (1, H) = u · XY (0) has a corresponding

down tick factor u1 in YX (1, H) = u1 · YX (0), and the down tick factor d
in XY (1, T ) = dXY (0) has a corresponding up tick factor d1 in YX (1, T ) =
1
d · YX (0). Therefore the following symmetry relationship holds:

1 − d1 1
−1
pX (u, d) = pY ( u1 , d1 ) = 1 1, q X (u, d) = q Y ( u1 , d1 ) = u
1 . (2.9)
u − d u − d1

If we want to price a contingent claim V in terms of a reference asset X,

we can do so from the following relationship:

VX (0) = EX [VX (1)] = VX (1, H) · PX (H) + VX (1, T ) · PX (T ) (2.10)

1−d u−1
= VX (1, H) · u · + VX (1, T ) · d · . (2.11)
u−d u−d
As in the case of the martingale measure PY , the martingale measure PX
represents how costly individual events are in terms of the reference asset X.
Consider an Arrow–Debreu security U that pays off one unit of an asset X
on the event ω = H:
U1 = I(ω = H) · X1 . (2.12)
64 Stochastic Finance: A Numeraire Approach

The initial value of this contract is given by

UX (0) = EX [UX (1)] = EX [I(ω = H) · XX (1)] = PX (H). (2.13)

In order to deliver a unit of an asset X on the event ω = H, one should charge

PX (H) units of X to start with. Constructing the hedging portfolio for this
claim is the subject of Exercise 2.1.

REMARK 2.1 Radon–Nikodým derivative

Consider a one-step binomial model with tick factors u and d, and correspond-
ing assets X and Y . We have seen that the measure PY that corresponds to
the reference asset Y is given by

1−d u−1
PY (H) = pY (u, d) = , PY (T ) = q Y (u, d) = .
u−d u−d

We can also determine the PX measure that corresponds to the reference asset
X by using the Radon–Nikodým derivative. From (1.69) we have

PX (ω) = Z(1, ω) · PY (ω) (2.14)

for ω = H, T . But according to the financial representation of Z from (1.72),

we also have

XY (1, ω) Y
PX (ω) = Z(1, ω) · PY (ω) = · P (ω), (2.15)
XY (0)

leading to
XY (1, H) Y 1−d
PX (H) = · P (H) = u · , (2.16)
XY (0) u−d
and
XY (1, T ) Y u−1
PX (T ) = · P (T ) = d · . (2.17)
XY (0) u−d

This confirms the formulas for PX (ω) we have previously obtained by using
the basic martingale property.

Note that the probability measure PY under which the price process XY (t)
is a martingale is inherently tied to the numeraire asset Y . When pricing a
contract under the probability measure PY , one computes how many units
of asset Y are needed in order to settle this contract. Other reference assets
imply different probabilities since the number of those assets needed to settle
the same contract is, in general, different.
 Binomial Models 65

2.1.2 Hedging in the Binomial Model

Let us assume that there is a contingent claim with a given payoff V at
time T = 1 defined on both outcomes V (1, H) and V (1, T ). It is possible to
find a portfolio
P0 = ∆X (0) · X + ∆Y (0) · Y,
such that
PY (0) = VY (0),
and
PY (1, H) = VY (1, H) PY (1, T ) = VY (1, T ). (2.18)
X
Let us find ∆ (0), the number of asset X to be held at time n = 0. Since

P1 = ∆X (0) · X + ∆Y (0) · Y,

we also have
PY (1, H) = ∆X (0) · XY (1, H) + ∆Y (0), (2.19)
and
PY (1, T ) = ∆X (0) · XY (1, T ) + ∆Y (0). (2.20)
Subtracting Equation (2.20) from Equation (2.19), and using the fact that the
price of the portfolio P at time n = 1 should match the price of the payoff V
(Equation (2.18)), we get

VY (1, H) − VY (1, T ) = ∆X (0) · (XY (1, H) − XY (1, T )) ,

or in other words,

VY (1, H) − VY (1, T )
∆X (0) = . (2.21)
XY (1, H) − XY (1, T )

Similarly, the position ∆Y (0) in the asset Y can be determined from the
following equations:

PX (1, H) = ∆X (0) + ∆Y (0) · YX (1, H), (2.22)

PX (1, T ) = ∆X (0) + ∆Y (0) · YX (1, T ). (2.23)

Subtracting Equation (2.23) from Equation (2.22), and using the fact that the
price of the portfolio P at time n = 1 should match the price of the payoff V ,
we get

VX (1, H) − VX (1, T ) = ∆Y (0) · (YX (1, H) − YX (1, T )) ,

or
VX (1, H) − VX (1, T )
∆Y (0) = . (2.24)
YX (1, H) − YX (1, T )
66 Stochastic Finance: A Numeraire Approach

Therefore the replicating portfolio is given by the following formula:

   
VY (1, H) − VY (1, T ) VX (1, H) − VX (1, T )
P0 = ·X + · Y. (2.25)
XY (1, H) − XY (1, T ) YX (1, H) − YX (1, T )

REMARK 2.2 The hedging position ∆Y (0) in the asset Y can also be
computed from the relationship

VY (1, H) = PY (1, H) = ∆X (0) · XY (1, H) + ∆Y (0),

or in other words,
VY (1, H) − VY (1, T )
∆Y (0) = VY (1, H) − · XY (1, H).
XY (1, H) − XY (1, T )
After some simplifications, we get
[XY (1, H) − XY (1, T )] · VY (1, H)
∆Y (0) =
XY (1, H) − XY (1, T )
[VY (1, H) − VY (1, T )] · XY (1, H)
−
XY (1, H) − XY (1, T )
−VY (1, H) · XY (1, T ) + VY (1, T ) · XY (1, H)
=
XY (1, H) − XY (1, T )
VX (1, H) − VX (1, T )
= ,
YX (1, H) − YX (1, T )

which confirms the previously obtained formula for ∆Y (0). If one wanted to
compute ∆Y (0) using the prices with respect to the reference asset Y only,
one can use the formula

VY (1, T ) · XY (1, H) − VY (1, H) · XY (1, T )

∆Y (0) = . (2.26)
XY (1, H) − XY (1, T )

2.1.3 Multiperiod Binomial Model

The multiperiod binomial model is a generalization of the one-step binomial
model. At time n, the price of XY (n) moves either to XY (n+1, H) = u·XY (n),
or to XY (n + 1, T ) = d · XY (n). In general, the price XY at time n is given
by the formula
XY (n) = XY (0) · u#H · d#T ,
where #H is the number of heads, and #T is number of tails in N trials.
Clearly, #H + #T = N .
 Binomial Models 67

Let V be a European option with a payoff VN at time N . One can determine

the price of this contract with respect to the reference asset Y using the
martingale property of VY (n) under the PY measure:

VY (n) = EYn [VY (N )].

Using the tower property, it is possible to compute this expectation recursively

using the relationship

VY (n) = EYn [VY (n + 1)] = pY · VY (n + 1, H) + q Y · VY (n + 1, T ), (2.27)

for n = N − 1, N − 2, . . . , 0. Similarly, the price of the contract with respect

to the reference asset X is given by

VX (n) = EX
n [VX (N )]

using the martingale property of VX (n) under the PX measure. The corre-
sponding recursive computation is given by

VX (n) = EX X X
n [VX (n + 1)] = p · VX (n + 1, H) + q · VX (n + 1, T ), (2.28)

for n = N − 1, N − 2, . . . , 0.

The hedging portfolio is analogous to that of the one-step binomial model

by the following formulas:

VY (n + 1, H) − VY (n + 1, T )
∆X (n) = , (2.29)
XY (n + 1, H) − XY (n + 1, T )

VX (n + 1, H) − VX (n + 1, T )
∆Y (n) = , (2.30)
YX (n + 1, H) − YX (n + 1, T )
and thus we can write
 
VY (n + 1, H) − VY (n + 1, T )
Pn = ·X
XY (n + 1, H) − XY (n + 1, T )
 
VX (n + 1, H) − VX (n + 1, T )
+ · Y. (2.31)
YX (n + 1, H) − YX (n + 1, T )

2.1.4 Numerical Example

Consider pricing a European call option with a payoff (XN − K · YN )+ in a
two step (N = 2) binomial model with parameters u = 2, d = 12 , initial price
XY (0) = 4, and strike K = 14
5 . When we take Y as a reference asset, the price
XY (n) has the following evolution:
68 Stochastic Finance: A Numeraire Approach

XY (2, HH) = 16
1
3

XY (1, H) = 8
1
3

2
3
XY (0) = 4 XY (2, HT, T H) = 4
1
3

2
3
XY (1, T ) = 2

2
3
XY (2, T T ) = 1

The probabilities pY = PY (H) and q Y = P(T ) are given by

1
1−d 1− 1 u−1 2−1 2
pY = = 2
1 = , and qY = = = .
u−d 2− 2
3 u−d 2 − 21 3

The price process XY (t) is a PY martingale, and thus we have

XY (n, ω) = EYn [XY (n + 1)] (ω) = pY · XY (n + 1, ωH) + q Y · XY (n + 1, ωT )
for n = 0, 1.

Let us denote the European option by V , and first determine the price of
V in terms of the asset Y . At the maturity time N = 2, the price of V with
respect to the reference asset Y is given by

+
VY (2) = XY (2) − 14
5 .
14 +
The corresponding payoff function is f Y (x) = (x − 5 ) . More specifically,
66
VY (2, HH) = 5 , VY (2, HT ) = VY (2, T H) = 56 , VY (2, T T ) = 0.
From the First Fundamental Theorem of Asset Pricing, the price VY (n) is
a martingale, and thus
VY (n, ω) = EYn [VY (n + 1)] (ω) = pY · VY (n + 1, ωH) + q Y · VY (n + 1, ωT )
for n = 0, 1. We first compute VY (1) from the known values of VY (2) to obtain:
VY (1, H) = EY1 [VY (2)] (H)
= pY · VY (2, HH) + q Y · VY (2, HT ) = 1
3 · 66
5 + 2
3 · 6
5 = 26
5 ,
 Binomial Models 69

VY (1, T ) = EY1 [VY (2)] (T )

= pY · VY (2, T H) + q Y · VY (2, T T ) = 1
3 · 6
5 + 2
3 · 0 = 25 .

At time n = 0, we have

VY (0) = EY0 [VY (1)]

= pY · VY (1, H) + q Y · VY (1, T ) = 1
3 · 26
5 + 2
3 · 2
5 = 2.

66
VY (2, HH) = 5
1
3

26
VY (1, H) = 5
1
3

2
3
6
VY (0) = 2 VY (2, HT, T H) = 5
1
3

2
3
2
VY (1, T ) = 5

2
3
VY (2, T T ) = 0

When we take X as a reference asset, the price evolution YX (n) is the

inverse of XY (n):
70 Stochastic Finance: A Numeraire Approach

1
YX (2, HH) = 16
2
3

1
YX (1, H) = 8
2
3

1
3
1 1
YX (0) = 4 YX (2, HT, T H) = 4
2
3

1
3
1
YX (1, T ) = 2

1
3
YX (2, T T ) = 1

The probability measure PX that makes the prices with respect to the refer-
ence asset X martingales is given by
1
1−d 1− 2 u−1 1 2−1 1
pX = u · =2· 2
1 = , and qX = d · = · 1 = ,
u−d 2− 2
3 u−d 2 2− 2 3

where pX = PX (H) and q X = PX (T ).

The price of the contract with respect to the reference asset X at time
N = 2 is given by

+
VX (2) = 1 − 14
5 YX (2) .

+
The payoff function is f X (x) = 1 − 14
5 ·x . This gives us the values
33 3
VX (2, HH) = 40 , VX (2, HT ) = VX (2, T H) = 10 , VX (2, T T ) = 0.

Using the First Fundamental Theorem of Asset Pricing, we can compute the
prices of the contract at an earlier time using the martingale property

VX (n, ω) = EX X X
n [VX (n + 1)] (ω) = p · VX (n + 1, ωH) + q · VX (n + 1, ωT )

for n = 0, 1. When n = 1, we get

VX (1, H) = EX
1 [VX (2)] (H)
= pX · VX (2, HH) + q X · VX (2, HT ) = 2
3 · 33
40 + 1
3 · 3
10 = 13
20 ,
 Binomial Models 71

VX (1, T ) = EX
1 [VX (2)] (T )
= pX · VX (2, T H) + q X · VX (2, T T ) = 2
3 · 3
10 + 1
3 · 0 = 15 .

At time n = 0, we have

VX (0) = EX
0 [VX (1)]
= pX · VX (1, H) + q X · VX (1, T ) = 2
3 · 13
20 + 1
3 · 1
5 = 21 .

33
VX (2, HH) = 40
2
3

13
VX (1, H) = 20
2
3

1
3
1 3
VX (0) = 2 VX (2, HT, T H) = 10
2
3

1
3
1
VX (1, T ) = 5

1
3
VX (2, T T ) = 0

Note that the prices computed with respect to the reference asset Y and X
are indeed consistent through the change of numeraire formula

VX (n) = VY (n) · YX (n). (2.32)

Similarly, we have
VY (n) = VX (n) · XY (n). (2.33)
For instance
1
VX (0) = VY (0) · YX (0) = 2 · 4 = 21 ,
which agrees with the price computed from the martingale property of the
price VX (n) under the PX measure.

Note that we can also obtain the prices of this European call option from
the Black–Scholes formula (1.77):
   
V = PX Y
n (XY (2) ≥ K) · X + −K · Pn (XY (2) ≥ K) · Y.
72 Stochastic Finance: A Numeraire Approach

At time n = 0,

PX (XY (2) ≥ K) = PX (HH) + PX (HT ) + PX (T H) = 2

3 · 23 + 32 · 13 + 31 · 23 = 98 ,

PY (XY (2) ≥ K) = PY (HH) + PY (HT ) + PY (T H) = 1

3 · 13 + 13 · 23 + 23 · 1
3 = 95 ,
resulting in
8 14 5 8 14
V0 = 9 · X0 − 5 · 9 · Y0 = 9 · X0 − 9 · Y0 .
Thus we have
8 14 8 14
VY (0) = 9 · XY (0) − 9 = 9 ·4− 9 = 2,

and
8 14 8 14 1
VX (0) = 9 − 9 · YX (0) = 9 − 9 · 4 = 21 .
At time n = 1, we have

PX X X
1 (XY (2) ≥ K)(H) = P (H) + P (T ) = 1,

PY1 (XY (2) ≥ K)(H) = PY (H) + PY (T ) = 1,

which leads to
14
V1 (H) = X1 − 5 · Y1 .
Therefore
14 14 26
VY (1, H) = XY (1, H) − 5 =8− 5 = 5 ,

and
14 14 1 13
VX (1, H) = 1 − 5 · YX (1, H) = 1 − 5 · 8 = 20 .

Similarly,
PX X 2
1 (XY (2) ≥ K)(T ) = P (H) = 3 ,

PY1 (XY (2) ≥ K)(T ) = PY (H) = 31 ,

which leads to
2 14 1 2 14
V1 (T ) = 3 · X1 − 5 · 3 · Y1 = 3 · X1 − 15 · Y1 .

Therefore
2 14 2 14
VY (1, T ) = 3 · XY (1, T ) − 15 = 3 ·2− 15 = 25 ,
and
2 14 2 14 1
VX (1, T ) = 3 − 5 · YX (1, T ) = 3 − 5 · 2 = 51 .
These results are indeed consistent with the prices we obtained from binomial
pricing.
 Binomial Models 73

The hedging portfolio is given by

   
VY (1, H) − VY (1, T ) VX (1, H) − VX (1, T )
P0 = ·X + ·Y
XY (1, H) − XY (1, T ) YX (1, H) − YX (1, T )
 26 2   13 1 
5 − 5 −5
= · X + 20 1 1 ·Y
8−2 8 − 2
4 6
= 5 ·X− 5 · Y,

   
VY (2, HH) − VY (2, HT ) VX (2, HH) − VX (2, HT )
P1 (H) = ·X + ·Y
XY (2, HH) − XY (2, HT ) YX (2, HH) − YX (2, HT )
 66 6   33 3 
5 − 5 40 − 10
= ·X + 1 1 ·Y
16 − 4 16 − 4
14
= 1·X − 5 · Y,

and
   
VY (2, T H) − VY (2, T T ) VX (2, T H) − VX (2, T T )
P1 (T ) = ·X + ·Y
XY (2, T H) − XY (2, T T ) YX (2, T H) − YX (2, T T )
6   3 
5 −0 10 − 0
= ·X + 1 ·Y
4−1 4 −1
2 2
= 5 ·X − 5 · Y.

Note that the hedging positions in this case do not satisfy

∆X (n) = PX
n (XY (2) ≥ K),

or
∆Y (n) = −K · PYn (XY (2) ≥ K)
in general, as one may expect from the Black–Scholes formula
   
V = PX Y
n (XY (2) ≥ K) · X + −K · Pn (XY (2) ≥ K) · Y.

However, these formulas for hedging are valid in the geometric Brownian mo-
tion model, as we show in Chapter 3 Diffusion Models.

2.1.5 Probability Measures for Exotic No-Arbitrage Assets

Every no-arbitrage asset comes with its own martingale measure. So far
we have identified the corresponding martingale measures associated with the
assets Y and X. It is interesting to note that even self-financing portfolios
created by trading in the assets X and Y generate their own martingale mea-
sures, as long as the value of the portfolio stays positive at all times. The
74 Stochastic Finance: A Numeraire Approach

resulting portfolio is itself a no-arbitrage asset.

Let us illustrate the martingale property on an asset A defined by a payoff

1
A1 = 2 [XY (0) + XY (1)] · Y1 . (2.34)

The asset A is known as an average asset and it represents the average price
of the asset X in terms of the asset Y . A contract whose payoff depends on
the asset A defined above is known as an Asian option. It is obvious that
A is a result of a self-financing strategy. One should start with a unit of an
asset X, sell immediately half of the unit of X for 21 · XY (0) units of Y at
time n = 0, and sell the remaining half of the unit of X for 12 · XY (1) units of
Y at time n = 1. In terms of the formulas, we have

A0 = X0 ,

and
∆X (0) = 12 .

The martingale measure associated with the asset A should assign proba-
bility to both possible outcomes ω, namely to ω = H, and to ω = T . The
most straightforward way to determine PA (H) and PA (T ) is from its relation-
ships with the measure PY via the Radon–Nikodým derivative described in
the formulas (1.69) and (1.72). Using these two relationships, we can write
AY (1, ω) Y
PA (ω) = Z(1, ω) · PY (ω) = · P (ω)
AY (0)
1
[XY (0) + XY (1)] Y
= 2 · P (ω)
XY (0)
 
= 12 · 1 + XXYY(1,ω)
(0) · PY (ω). (2.35)

Thus we have
1−d
PA (H) = 1
2 · (1 + u) · , (2.36)
u−d
and
u−1
PA (T ) = 1
2 · (1 + d) · . (2.37)
u−d
The pricing measure PA is useful for pricing Asian options; see the following
example, or Chapter 9 Asian Options.

Example 2.1
Consider a contract V that pays off

V1 = I(ω = H) · A1 .
 Binomial Models 75

Its price at time n = 0 is simply

1−d
VA (0) = PA (H) = 1
2 · (1 + u) · .
u−d

The price with respect to the reference asset Y can be obtained from the
change of numeraire formula

1 1−d
VY (0) = VA (0) · AY (0) = 2 · (1 + u) · · XY (0).
u−d

2.2 Binomial Model with an Arbitrage Asset

Some contracts are written on one or several arbitrage assets and there is no
way to replace them with a suitable no-arbitrage asset in the same way as for
European options. This is, for instance, the case for American stock options
with two underlying assets, a stock S, and the dollar $. The American option
pays off either f Y (XY (τ )) units of an asset Y , or f X (YX (τ )) units of an asset
X at the exercise time τ ∈ [0, T ]. The exercise time τ is chosen by the holder
of the option.

When the underlying assets are the stock S, and the dollar $, the payoff is
either f $ (S$ (τ )) units of dollars $, or f S ($S (τ )) units of the stock S at the
exercise time τ ∈ [0, T ]. Since the payoff has to be delivered at an arbitrary
time chosen by the holder of the option, it is not possible to substitute the
dollar with the corresponding bond. When there is more than one possible
delivery time, there is no corresponding no-arbitrage asset.

However, hedging of a contract whose underlying is one or more arbitrage

assets must be done exclusively in no-arbitrage assets. For an American stock
option, the pricing can be done with respect to both reference assets $ and S,
but the hedging is done in S and in a suitable no-arbitrage proxy of $ such
as a bond B T . Thus we have three assets to consider: a stock S, a bond B T ,
and the dollar $.

Let us start with a one-step model at time n. The binomial model then
assumes that the stock price in terms of dollars either increases by a factor
u, or decreases by a factor d. Thus we have S$ (n + 1, H) = u · S$ (n), and
S$ (n + 1, T ) = d · S$ (n, T ).
76 Stochastic Finance: A Numeraire Approach

S$ (n + 1, H) = u · S$ (n)

S$ (n)

S$ (n + 1, T ) = d · S$ (n)

The bond has deterministic evolution given by B$T (n + 1) = (1 + r) · B$T (n).

B$T (n) B$T (n + 1) = (1 + r) · B$T (n)

Neither of these two evolutions are justified by the First Fundamental Theo-
rem of Asset Pricing which considers only no-arbitrage assets. In particular,
these prices are not martingales under the corresponding reference measures.
The two no-arbitrage assets here are S and B T . The First Fundamental The-
orem of Asset Pricing says that the price evolution SB T (n) is a PT martingale
(under the T-forward measure that corresponds to a B T bond), and that the
price evolution BST (n) is a PS martingale (under the stock measure that cor-
responds to S). The measures PT and PS are the two measures that can be
used for the pricing of an American option; there is no pricing measure that
corresponds to the dollar.

From the price evolutions of S$ and B$T , we can already determine the
pricing measures PT and PS . Note that

SB T (n + 1, H) = S$ (n + 1, H) · $B T (n + 1)
1 u
= u · S$ (n) · 1+r · $B T (n) = 1+r · SB T (n),

and

SB T (n + 1, T ) = S$ (n + 1, T ) · $B T (n + 1)
1 d
= d · S$ (n) · 1+r · $B T (n) = 1+r · SB T (n).
 Binomial Models 77

u
SB T (n + 1, H) = 1+r · SB T (n)

SB T (n)

d
SB T (n + 1, T ) = 1+r · SB T (n)

This is the same model as described in the previous section for two no-
arbitrage assets with the exception that the scaling factors u and d are slightly
u d
modified. The u factor is now 1+r , and the d factor is now 1+r . The asset Y
T
is now the bond B , and the asset X is now the stock S. Rewriting Equation
u d
(2.3) with the new scaling factors 1+r and 1+r , we obtain that the probabil-
ities that correspond to the T-forward measure are given by

1+r−d u − (1 + r)
pT (u, d) = , q T (u, d) = . (2.38)
u−d u−d

Similarly, by rewriting Equation (2.8) we obtain the stock measure

u 1+r−d d u − (1 + r)
pS (u, d) = · , q S (u, d) = · . (2.39)
1+r u−d 1+r u−d

Even when the contract depends on an arbitrage asset, we still have to

compute its price with respect to a no-arbitrage asset and convert it back to
the given arbitrage asset using the change of numeraire formula. When the
contract is written on a stock and on dollars, we have two possible reference
assets: a bond B T , and a stock S. Recall that when V is a European option
with a payoff VN at time N , we can determine the price of this contract with
respect to the reference asset B N (maturity time T = N ) using the martingale
property of VB N (n) under the PT measure:

VB N (n) = ETn [VB N (N )].

Using the tower property, it is possible to compute this expectation recursively

from the relationship

VB N (n) = ETn [VB N (n + 1)] = pT · VB N (n + 1, H) + q T · VB N (n + 1, T ), (2.40)

78 Stochastic Finance: A Numeraire Approach

for n = N − 1, N − 2, . . . , 0. Since we assume

1
BnN = · $n ,
(1 + r)N −n

Equation (2.40) can be rewritten in terms of the dollar as a reference asset as

 
V$ (n) = 1
1+r · ETn [V$ (n + 1)] = 1
1+r · pT · V$ (n + 1, H) + q T · V$ (n + 1, T ) .
(2.41)
This is a mainstream way of computing a price of a derivative contract in a
binomial model, even for European-type options. Note that V$ (n) is not a PT
martingale, and thus the formula requires discounting the expectation by a
1
factor 1+r .

Similarly, the price of the contract with respect to the stock S is given by

VS (n) = ESn [VS (N )]

using the martingale property of VS (n) under the PS measure. Since VS is a

martingale, the corresponding recursive computation is given by

VS (n) = ESn [VS (n + 1)] = pS · VS (n + 1, H) + q S · VS (n + 1, T ), (2.42)

for n = N − 1, N − 2, . . . , 0.

As we pointed out in the previous section, bringing an arbitrage asset to

pricing European-type options only increases the dimensionality of the prob-
lem, a step that is not necessary. However, it is typical to use the approach
presented in this section that needs three assets: a stock S, a bond B N , and
the dollar $. The bond plays a role of a no-arbitrage proxy of the dollar, and
is required for hedging. The hedging portfolio has no positions in arbitrage as-
sets. A cleaner solution for European-type options is to consider only a stock
S, and a bond B N . The dollar value of a contract is easily determined from
the dollar value of the stock

V$ (n) = VS (n) · S$ (n)

or from the dollar value of the bond

V$ (n) = VB N (n) · B$N (n).

2.2.1 American Option Pricing in the Binomial Model

A general treatment of pricing American options is given in Chapter 7. In
this section, we study pricing and hedging of an American stock option V
in the binomial model. The holder of the contract has two possible actions
 Binomial Models 79

at each time n: either exercise the option and collect the payoff, or keep the
contract for the future. Both actions have value: immediate exercise has a
value known as an intrinsic value of the option; keeping the option has
a value known as a continuation value of the option. Thus the holder
should evaluate the intrinsic and continuation value of the option, and choose
the one with a higher price so as not to produce arbitrage opportunities for
the seller of the option.

Recall that an American stock option pays off either f $ (S$ (τ )) units of
the dollar $ or f S ($S (τ )) units of a stock S at the exercise time τ ∈ [0, T ].
The intrinsic value can be expressed in terms of both assets that enter this
contract. In particular, the intrinsic value of the option at time n is f $ (S$ (n))
when the dollar $ is chosen as a reference asset, and it is f S ($S (n)) when a
stock S is chosen as a reference asset. Similarly, the continuation value is given
1
by 1+r · ETn [V$ (n + 1)] when the dollar $ is chosen as a reference asset, and
S
it is En [VS (n + 1)] when a stock S is chosen as a reference asset. Comparing
the intrinsic and the continuation values in both cases, we find that
 
V$ (n) = max f $ (S$ (n)) , 1+r
1
· ETn [V$ (n + 1)] (2.43)

for the case when the dollar $ is chosen as a reference asset, and
 
VS (n) = max f S ($S (n)) , ESn [VS (n + 1)] (2.44)

for the case when a stock S is chosen as a reference asset. Note that both
expressions are equivalent and they are related by the change of numeraire
formula
V$ (n) = VS (n) · S$ (n).
The option should be exercised the first time the option value coincides
with its intrinsic value. In this case, the continuation value of the option is
smaller or equal to the intrinsic value. Let τ ∗ be the optimal exercise policy.
The exercise time is given by
n o
τ ∗ = min n ≥ 0 : V$ (n) = f $ (S$ (n)) , (2.45)

which is equivalent to
n o
τ ∗ = min n ≥ 0 : VS (n) = f S ($S (n)) . (2.46)

2.2.2 Hedging
The hedging of an American stock option must be done in no-arbitrage
assets. An American stock option is written on two assets, a stock S, and the
80 Stochastic Finance: A Numeraire Approach

dollar $. The hedging portfolio will take positions in a no-arbitrage asset S,

and in a corresponding no-arbitrage proxy asset for $ such as a zero coupon
bond B N . The hedging portfolio Pn takes the same form as in Equation (2.31)
with X replaced by S and Y replaced by B N :
Pn = ∆S (n) · S + ∆T (n) · B T ,
where
VB N (n + 1, H) − VB N (n + 1, T )
∆S (n) = ,
SB N (n + 1, H) − SB N (n + 1, T )
and
VS (n + 1, H) − VS (n + 1, T )
∆T (n) =
BSN (n + 1, H) − BSN (n + 1, T )
V N (n + 1, T ) · SB N (n + 1, H) − VB N (n + 1, H) · SB N (n + 1, T )
= B .
SB N (n + 1, H) − SB N (n + 1, T )
We can express the hedging positions ∆S (n) and ∆T (n) in terms of dollar
prices. Note that
VB N (n + 1, H) − VB N (n + 1, T ) B$N (n + 1)
∆S (n) = ·
SB N (n + 1, H) − SB N (n + 1, T ) B$N (n + 1)
V$ (n + 1, H) − V$ (n + 1, T )
= .
S$ (n + 1, H) − S$ (n + 1, T )
The hedging position ∆T (n) can be expressed either in terms of the prices
with respect to a stock S as
VS (n + 1, H) − VS (n + 1, T ) $B N (n + 1)
∆T (n) = ·
BSN (n + 1, H) − BSN (n + 1, T ) $B N (n + 1)
VS (n + 1, H) − VS (n + 1, T )
= · $ N (n + 1)
$S (n + 1, H) − $S (n + 1, T ) B
VS (n + 1, H) − VS (n + 1, T )
= · (1 + r)N −n−1 ,
$S (n + 1, H) − $S (n + 1, T )
or in terms of the prices with respect to a dollar $ as

∆T (n) =
 
VB N (n + 1, T ) · SB N (n + 1, H) − VB N (n + 1, H) · SB N (n + 1, T )
= ×
SB N (n + 1, H) − SB N (n + 1, T )
" #2
B$N (n + 1)
×
B$N (n + 1)
 
V$ (n + 1, T ) · S$ (n + 1, H) − V$ (n + 1, H) · S$ (n + 1, T )
= ×$B N (n+1)
S$ (n + 1, H) − S$ (n + 1, T )
 
V$ (n + 1, T ) · S$ (n + 1, H) − V$ (n + 1, H) · S$ (n + 1, T )
= ×(1+r)N −n−1 .
S$ (n + 1, H) − S$ (n + 1, T )
 Binomial Models 81

Thus we can write

 
V$ (n + 1, H) − V$ (n + 1, T )
Pn = ·S
S$ (n + 1, H) − S$ (n + 1, T )
 
VS (n + 1, H) − VS (n + 1, T ) N −n−1
+ · (1 + r) · BN .
$S (n + 1, H) − $S (n + 1, T )

Since the interest rate r is assumed to be deterministic, the bond B N is just a

constant multiple of the money market account M . Thus we can alternatively
invest in the money market instead of in the bond B N . Assuming that time
n is the reference time for the money market, we have that Mn = 1$n , and
Mn+1 = (1 + r)$n+1 . The relationship between M and B N is given by Mn =
(1 + r)N −n · BnN . Let ∆M (n) be the hedging position in the money market
account M . Since ∆M (n)·Mn should have the same price as the bond position
∆T (n) · BnN , we must have

∆M (n) = (1 + r)−(N −n) · ∆T (n).

This follows from the relationship

∆M (n) · Mn = ∆M (n) · (1 + r)N −n · BnN = ∆T (n) · BnN .

In this case, we can also express the hedging portfolio as

 
V$ (n + 1, H) − V$ (n + 1, T )
Pn = ·S (2.47)
S$ (n + 1, H) − S$ (n + 1, T )
 
VS (n + 1, H) − VS (n + 1, T ) 1
+ · · M.
$S (n + 1, H) − $S (n + 1, T ) 1+r

2.2.3 Numerical Example

Let us consider a binomial model for S, B N and $ with the following pa-
rameters: u = 2, d = 21 , r = 41 , S$ (0) = 4. Let us consider an American
put option with a payoff (5 · $τ − Sτ )+ . The dollar price of the stock has the
following evolution
82 Stochastic Finance: A Numeraire Approach

S$ (2, HH) = 16
1
2

S$ (1, H) = 8
1
2

1
2
S$ (0) = 4 S$ (2, HT, T H) = 4
1
2

1
2
S$ (1, T ) = 2

1
2
S$ (2, T T ) = 1

The probabilities pT = PT (H) and q T = PT (T ) are given by

1+r−d 1 + 41 − 1
1
pT = = 2
= ,
u−d 2 − 21 2

and
u − (1 + r) 2 − (1 + 41 ) 1
qT = = 1 = .
u−d 2− 2 2

The price process S$ (n) satisfies

S$ (n, ω) = 1
1+r · ETn [S$ (n + 1)] (ω)
 
= 1
1+r · pT · S$ (n + 1, ωH) + q T · S$ (n + 1, ωT )

for n = 0, 1.

Let V denote the American option with a payoff (5 · $τ − Sτ )+ at exercise

time τ , where τ ≤ 2. Let us determine the price of this contract V$ with
respect to the dollar $. The intrinsic value function f $ for the dollar as a
reference asset is
f $ (x) = (5 − x)+ ,

which means that the option pays off

(5 − S$ (τ ))+
 Binomial Models 83

units of a dollar $ at the exercise time τ . The price of the American option is
computed by comparing the intrinsic value and the continuation value of the
option:
 
V$ (n) = max f $ (S$ (n)) , 1+r
1
· ETn [V$ (n + 1)] . (2.48)

The computation of the price is done recursively, starting from the maturity
of the option. At N = 2, there is no continuation possible, and thus the value
of the option is equal to its intrinsic value f $ (S$ (2)):

V$ (2, HH) = 0, V$ (2, HT ) = V$ (2, T H) = 1, V$ (2, T T ) = 4.

We can compute the price V$ (n) for n = 1, 0 using formula (2.48). We first
compute V$ (1) by comparing the intrinsic value and the continuation value of
the option
 
V$ (1, H) = max f $ (S$ (1, H)) , 1+r
1
· ET1 [V$ (2)] (H)
  
= max f $ (S$ (1, H)) , 1+r
1
· pT · V$ (2, HH) + q T · V$ (2, HT )
 
= max (5 − 8)+ , 1 1 · [ 12 · 0 + 2 · 1] = max(0, 52 ) = 25 .
1
1+ 4

The intrinsic value of the option is zero, and thus it is optimal not to exercise
the option in this scenario. On the other hand
 
V$ (1, T ) = max f $ (S$ (1, T )) , 1+r
1
· ET1 [V$ (2)] (T )
  
= max f $ (S$ (1, T )) , 1+r
1
· pT · V$ (2, T H) + q T · V$ (2, T T )
 
= max (5 − 2)+ , 1 1 · [ 12 · 1 + 12 · 4] = max(3, 2) = 3,
1+ 4

and thus it is better to exercise the option since the intrinsic value is larger
than the continuation value. At time n = 0, we have
 
V$ (0) = max f $ (S$ (0)) , 1+r
1
· ET0 [V$ (1)]
  
= max f $ (S$ (0)) , 1+r
1
· pT · V$ (1, H) + q T · V$ (1, T )
 
+
= max (5 − 4) , 1 · [ 2 · 5 + 2 · 3] = max(1, 34
1 1 2 1 34
25 ) = 25
1+ 4

and thus it is better to hold the option. Thus the price of the option is given
by the following binomial tree. Note that in the scenario when ω = T , it is
optimal to exercise the option, and continuation of the option is suboptimal.
Suboptimal continuation is depicted with a dashed line.
84 Stochastic Finance: A Numeraire Approach

V$ (2, HH) = 0
1
2

2
V$ (1, H) = 5
1
2

1
2
34
V$ (0) = 25 V$ (2, HT, T H) = 1
1
2

1
2
V$ (1, T ) = 3

1
2
V$ (2, T T ) = 4

Let us consider the reference asset S. The evolution of the $S price is given
by

1
$S (2, HH) = 16
4
5

1
$S (1, H) = 8
4
5

1
5
1 1
$S (0) = 4 $S (2, HT, T H) = 4
4
5

1
5
1
$S (1, T ) = 2

1
5
$S (2, T T ) = 1
 Binomial Models 85

The probabilities pS = PS (H) and q S = PS (T ) are given by

u 1+r−d 2 1 + 14 − 1
4
pS = · = 1 · 2
= ,
1+r u−d 1+ 4 2 − 12 5

and
1
d u − (1 + r) 2 − (1 + 14 ) 1
qS = · = 2 1 · = .
1+r u−d 1+ 4 2 − 12 5
The price process $S (n) satisfies

$S (n, ω) = (1 + r) · ESn [$S (n + 1)] (ω)

 
= (1 + r) · pS · $S (n + 1, ωH) + q S · $S (n + 1, ωT )

for n = 0, 1.

The intrinsic value of the option in terms of the stock is given by f S (x) =
(5 · x − 1)+ . Similar to the case when the dollar is chosen as a reference asset,
the price of the American option is computed by comparing the intrinsic value
and the continuation value of the option:
 
VS (n) = max f S ($S (n)) , ESn [VS (n + 1)] . (2.49)

The computation of the price is done recursively, starting from the maturity
of the option. At N = 2, there is no continuation possible, and thus the value
of the option is equal to its intrinsic value f S ($S (2)):

VS (2, HH) = 0, VS (2, HT ) = VS (2, T H) = 14 , VS (2, T T ) = 4.

We can compute the price V$ (n) for n = 1, 0 using the formula (2.49). We
first compute VS (1) by comparing the intrinsic value and the continuation
value of the option
 
VS (1, H) = max f S ($S (1, H)) , ES1 [VS (2)] (H)
 
= max f S ($S (1, H)) , pS · VS (2, HH) + q S · VS (2, HT )


= max ( 58 − 1)+ , [ 54 · 0 + 15 · 41 ] = max(0, 20
1 1
) = 20 .

The intrinsic value of the option is zero, and thus it is optimal not to exercise
the option in this scenario. On the other hand
 
VS (1, T ) = max f S ($S (1, T )) , ES1 [VS (2)] (T )
 
= max f S ($S (1, T )) , pS · VS (2, T H) + q S · VS (2, T T )


= max ( 25 − 1)+ , [ 45 · 41 + 51 · 4] = max( 32 , 1) = 23 ,
86 Stochastic Finance: A Numeraire Approach

and thus it is better to exercise the option since the intrinsic value is larger
than the continuation value. At time n = 0, we have
 
VS (0) = max f S ($S (0)) , ES0 [VS (1)]
!

S S S
= max f ($S (0)) , p · VS (1, H) + q · VS (1, T )


= max ( 54 − 1)+ , [ 54 · 1
20 + 1
5 · 23 ] = max( 14 , 50
17
)= 17
50

and thus it is better to continue. Thus the price of the option is given by the
following binomial tree. Note that in the scenario when ω = T , it is optimal to
exercise the option, and continuation of the option is suboptimal. Suboptimal
continuation is depicted with a dashed line.

VS (2, HH) = 0
4
5

1
VS (1, H) = 20
4
5

1
5
17 1
VS (0) = 50 VS (2, HT, T H) = 4
4
5

1
5
3
VS (1, T ) = 2

1
5
VS (2, T T ) = 4

The prices of the American option V with respect to the dollar $ and the
stock S are related by the change of numeraire formula

V$ (n) = VS (n) · S$ (n). (2.50)

Similarly, we have
VS (n) = V$ (n) · $S (n). (2.51)

The reader can check that this is indeed the case.

 Binomial Models 87

The hedging portfolio is given by

 
V$ (1, H) − V$ (1, T )
P0 = ·S
S$ (1, H) − S$ (1, T )
 
VS (1, H) − VS (1, T ) 1
+ · (1 + r) · B 2
$S (1, H) − $S (1, T )
2   1   
−3 − 23 5
= 5 · S + 20 1 1 · · B2
8−2 8 − 2
4
13
= − 30 ·S+ 29
6 · B 2 = − 13
30 · S +
232
75 ·M
at time n = 0, and by
 
V$ (2, HH) − V$ (2, HT )
P1 (H) = ·S
S$ (2, HH) − S$ (2, HT )
 
VS (2, HH) − VS (2, HT )
+ · B2
$S (2, HH) − $S (2, HT )
   
0−1 0− 1
= · S + 1 41 · B 2
16 − 4 16 − 4
1
= − 12 ·S+ 4
3 · B 2 = − 12
1
·S+ 16
15 ·M
at time n = 1. The positions in the money market M correspond to the dol-
lar positions at each time, but the positions in the money market gain interest.

Note that the holder of the option should exercise the contract in the sce-
nario when ω = T . However, if the holder fails to exercise the option in that
scenario, the seller of the option can keep the difference between the intrin-
sic and continuation values. The seller can create a hedging portfolio for the
suboptimal continuation that costs only the continuation value of the option.
The difference of the intrinsic and continuation value is given by

f $ (S$ (1, T )) − 1
· ET1 [V$ (2)] (T ) =
1+r
 
= f $ (S$ (1, T )) − 1+r
1
· pT · V$ (2, T H) + q T · V$ (2, T T ) =
= (5 − 2)+ − 1
1 · [ 12 · 1 + 1
2 · 4] = 3 − 2 = 1
1+ 4

when the reference asset is the dollar $, or

f S ($S (1, T )) − ES1 [VS (2)] (T ) =

 
= f S ($S (1, T )) − pS · VS (2, T H) + q S · VS (2, T T ) =
= ( 25 − 1)+ − [ 45 · 1
4 + 1
5 · 4] = 3
2 −1= 1
2

when the reference asset is a stock S. Thus the seller of the option can keep
this difference
1$ = 21 · S,
88 Stochastic Finance: A Numeraire Approach

and use the remaining funds

2$ = 1 · S
to hedge the option. We get
 
V$ (2, T H) − V$ (2, T T )
P1 (T ) = ·S
S$ (2, T H) − S$ (2, T T )
 
VS (2, T H) − VS (2, T T )
+ · B2
$S (2, T H) − $S (2, T T )
  1 
1−4 −4
= · S + 41 · B2
4−1 4 − 1
= −1 · S + 5 · B 2 = −1 · S + 4 · M.

References and Further Reading

The binomial model was introduced by Cox et al. (1979). An interested
reader may also refer to standard textbooks of Cox and Rubinstein (1985), or
Shreve (2004a).

Exercises
2.1 Consider a one-step binomial model for two no-arbitrage assets X and
Y with parameters 0 < d < 1 < u, and initial price XY (0), meaning that
XY (1, H) = u · XY (0), and XY (1, T ) = d · XY (0).
(a) Compute the price of a contract that pays off V1 = I(ω = H) · Y1 (in
terms of a reference asset Y , and the probability measure PY ).
(b) Find the hedging portfolio P0 for this contract, i.e., determine ∆X (0) and
∆Y (0) such that
P0 = ∆X (0) · X + ∆Y (0) · Y,
with
PY (0) = VY (0), PY (1) = VY (1).
(c) Compute the price of a contract that pays off U1 = I(ω = H) · X1 (in
terms of a reference asset X, and the probability measure PX ).
(d) Find the hedging portfolio P0 for this contract, i.e., determine ∆X (0) and
∆Y (0) such that
P0 = ∆X (0) · X + ∆Y (0) · Y,
 Binomial Models 89

with
PX (0) = UX (0), PX (1) = UX (1).

2.2 Consider a one-step binomial model for two no-arbitrage assets X and
Y with parameters 0 < d < 1 < u, and initial price XY (0) = 1. Find the price
and the hedge of a contract V that pays off

V1 = max(X1 , Y1 ).

Compute the price of the contract V using both martingale measures PY and
PX .

2.3 Consider a two-step binomial model with general u > 1 > d > 0.
(a) Find the price and the hedging portfolio for a contract V that pays off
 
V2 = 12 XY (1) + 12 XY (2) · Y2 .

Note that the price and the hedge do not depend on the choice of pa-
rameters u and d.
(b) How can one lock an arbitrage opportunity if somebody is offering to buy
or to sell V0 for 1.05 × X0 ?
(c) Determine the price and the hedge of a contract A that pays off

A2 = (VY (2) − 2)+ · Y2 ,

using the parameters u = 2, d = 12 , and XY (0) = 4.

2.4 Consider an American contract V that pays off max(5$τ , Sτ ) =

max(5, S$ (τ )) · $τ in a two-step binomial model with parameters u = 2, d = 12 ,
r = 41 , and S$ (0) = 4.
(a) Find the price V$ (n) for n = 0, 1, 2. Determine the optimal stopping
strategy τ ∗ .
(b) Find the hedging portfolio.
(c) What is the price and an optimal stopping strategy for an American
contract that pays off min(5$τ , Sτ )?

2.5 Consider a two-step binomial model with two no-arbitrage assets X and
Y and parameters u > 1 > d > 0. Let A be a contract with a payoff
1
A2 = 3 (XY (0) + XY (1) + XY (2)) · Y2 .

This is an average asset. Assume that u = 2, d = 12 , XY (0) = 4. Consider a

contract that pays off V2 = (A2 − X2 )+ (Asian floating strike option).
90 Stochastic Finance: A Numeraire Approach

(a) Compute the prices VY (n) using the measure PY . Note that

VY (2) = (AY (2) − XY (2))+ .

(b) Compute VX (n) using the measure PX . Note that

VX (2) = (AX (2) − 1)+ .

(c) Determine the hedging strategy for V in the assets X and Y .

(d) The asset A admits a model independent hedge (∆X (0) = 32 , ∆X (1) = 13 ),
and it is a no-arbitrage asset. Therefore there is a probability measure
PA under which the prices of no-arbitrage assets with respect to the
reference asset A are martingales. Determine PA by giving PA (HH),
PA (HT ), PA (T H), PA (T T ). These are functions of parameters u and
d. The easiest way is to employ the Radon–Nikodým derivative with
respect to a martingale measure PY . Recall that

AY (2, ω) Y
PA (ω) = · P (ω).
AY (0)

Determine also PA (H) and PA (T ) (probabilities after one time step)

using a similar formula, and PA (HH|H), PA (HT |H), PA (T H|T ),
PA (T T |T ) (conditional probabilities from time 1 to time 2).
(e) Compute VA (n) using the measure PA . Note that

VA (2) = (1 − XA (2))+ .
Chapter 3
Diffusion Models

This chapter introduces diffusion models. Under very broad conditions, all
no-arbitrage models of a continuous price evolution are diffusion models. In
other words, every continuous evolution of the price can be expressed as an
Ito’s integral. This result is known as a Martingale Representation Theorem.

Diffusion models of price use Brownian motion to represent market noise.

Since the market noise itself can take negative values, it does not serve as a
good model for the prices. However, we can take the corresponding stochastic
exponential which is a positive martingale, and thus is perfectly suitable for a
no-arbitrage model of a price process. The simplest model assumes a constant
volatility that leads to a geometric Brownian motion. While most of the real
price processes do not have constant volatility, this assumption still results
in reasonable models for prices and hedges of complex financial instruments.
Moreover, the prices of many financial products in the geometric Brownian
motion model admit closed form solutions, and thus they are easy to use.

In order to compute the prices of financial derivatives, we need to deter-

mine the martingale measures that correspond to all the assets relevant to
the given contract. For instance, the Black–Scholes formula for the price of
the European call option uses both probability measures PX and PY that are
associated with the assets X and Y . The probability measure PX can be de-
termined from the evolution of the inverse price YX (t), and this price has to
be a PX martingale. It turns out that the evolution of the inverse price YX (t)
is also a geometric Brownian motion, but the market noise W X is associated
with the reference asset X.

Diffusion models have one important property: every no-arbitrage asset

comes with its own market noise. An asset Y has a market noise W Y , and an
asset X has a market noise W X . Although W Y and W X are perfectly cor-
related in the geometric Brownian motion model, we can always identify the
market noise that comes with each individual asset. Even more complicated
assets, such as a power option, or an average asset, come with its own market
noise. This fact will be used for pricing barrier, lookback, and Asian options
in the subsequent chapters.

The first section introduces the geometric Brownian motion model, and

91
92 Stochastic Finance: A Numeraire Approach

studies the evolution of the prices XY and YX under the corresponding mar-
tingale measures PY and PX . We also show that the measures PY and PX
have the interpretation of how costly a given event is if settled in terms of the
asset Y , or in terms of the asset X, respectively. The second section introduces
general European contracts. European contracts are contracts on two assets
that are defined by the payoff function, which can be expressed in terms of
each reference asset Y or X. The two payoff functions are related by a formula
known as a perspective mapping. Some contracts remain the same if the roles
of the assets Y and X is switched in the payoff function; for instance the best
of the two assets defined as max(XT , YT ) is the same as max(YT , XT ). The
best of the two assets naturally leads to European call and put options with
the payoff
(XT − K · YT )+ = max(XT , K · YT ) − K · YT .

We give examples of European call and put options that appear in different
markets: a stock option, a currency option, an exchange option, or a caplet.
Their prices and the hedging portfolios are given by the Black–Scholes for-
mula. We compute all prices in terms of the no-arbitrage assets so that we
can employ the First Fundamental Theorem of Asset Pricing directly. In or-
der to get the prices in terms of a dollar, an arbitrage asset, we can trivially
apply the change of numeraire formula to the prices computed with respect
to no-arbitrage assets.

The price of a contingent claim can be computed by two alternative meth-

ods: by computing the conditional expectation, or by solving the associated
partial differential equation. The case of European options is usually simple
enough to obtain closed form formulas, but both approaches also work for
more complicated products when no close formula is known. The conditional
expectation can be approximated by Monte Carlo methods, and the partial
differential equation can be solved numerically by applying finite difference
techniques.

The primary goal of contingent pricing is to find the dollar price of a given
contract. Our text suggests to compute the price of a contingent
claim with respect to a no-arbitrage asset first, such as a corre-
sponding bond, and then convert it to the dollar price using the
change of numeraire. This approach is valid in general, and it has
clear computational advantages when the contingent claim is more
complex, such as in the case of exotic options. However, the dollar
prices of European claims also satisfy a certain and more complicated partial
differential equation that is obtained by discounting to the dollar prices of the
underlying assets. But this partial differential equation does not hold in gen-
eral, it assumes a deterministic evolution of the interest rate. We mention it in
 Diffusion Models 93

our text since the partial differential equation in terms of dollars is the most
widely used in practice. For simple contracts, such as for European options,
it does not make a difference to compute the prices under different reference
assets (arbitrage or no-arbitrage) since the price of the contract is simple to
determine. The only loss when computing the dollar prices directly from the
corresponding partial differential equation approach is that the approach does
not apply to stochastic interest rates. In that case one should compute the
prices in terms of the bond, and convert it to dollar prices by changing the
numeraire.

For more complex products, such as for barrier, lookback, or Asian options,
using the no-arbitrage asset as a numeraire leads to significant computational
advantages. On the other hand, American options have to use dollar values
in order to compare the intrinsic and the continuation values, and the par-
tial differential equation in terms of dollars has to be used. In the case of
the American option, it is the setup of the contract that forces us to use the
partial differential equation in terms of a dollar.

We also discuss how to construct the hedging portfolios for European con-
tracts. The hedging must always be done in the two underlying no-arbitrage
assets. We determine the hedging positions in both assets. We can also get a
similar expression for the hedging positions in terms of the dollar price func-
tions. The hedging positions for European call options are bounded in both
assets; the position in the asset X is always between zero and one, and the
hedging position in the asset Y is always between minus the strike K and zero.

We also briefly introduce stochastic volatility models. The price of the con-
tract is still considered to be Markov, but it depends on two parameters: the
price XY (t) of the asset X and stochastic volatility ξ(t). The resulting partial
differential equation for the price of the derivative security becomes two di-
mensional in space. The chapter is concluded with an example of a European
option contract in the foreign exchange market which is just a special case of
the general approach presented in the previous text.

3.1 Geometric Brownian Motion

Assume that the two assets X and Y are no-arbitrage assets. We have
seen that the price XY (t) must be a PY martingale in order to prevent any
arbitrage opportunity. In continuous time, a general martingale can be written
as a sum of a martingale with continuous paths and a purely discontinuous
94 Stochastic Finance: A Numeraire Approach

martingale:
M(t) = Mc (t) + Md (t). (3.1)
Continuous martingales adapted to a filtration FtW generated by a Brownian
motion W are in fact diffusions; they can be represented as stochastic integrals
with respect to Brownian motion. Thus
Z t
Mc (t) = Mc (0) + φ(s)dW (s), (3.2)
0

where φ(t) is adapted to FtW . This result is known as a Martingale Repre-

sentation Theorem (Theorem A.3).

This chapter focuses on price models with continuous paths. The process
XY (t) must have the form

dXY (t) = φ(t)dW (t).

Let us start with the simple but very popular model when

φ(t) = σXY (t).

The price process XY (t) follows

dXY (t) = σXY (t)dW Y (t), (3.3)

which is known as a geometric Brownian motion. The parameter σ is

referred to as volatility. Volatility is inherent to diffusion models. Similar to
price, volatility is a pairwise relationship between two assets X and Y . The
price XY of the asset X with respect to a reference asset Y may have very
different volatility than the price XZ with respect to a different reference asset
Z. For instance, a typical dollar stock price S$ is more volatile than a stock
price SI taken with respect to a market index I. Sometimes we will denote
by σxy the volatility that corresponds to the assets X and Y .

A natural question is how the measure PY is determined. Under PY , the

driving process W Y (t) is a Brownian motion. Also the above stochastic dif-
ferential equation has the following solution:


XY (t) = XY (0) · exp σW Y (t) − 12 σ 2 t . (3.4)

Note that XY (t) is a PY martingale.

In order to compute the prices of European options and other derivative

securities, we also need to determine the probability measure PX . The role
 Diffusion Models 95

of X and Y should be exchangeable in models that preserve the symmetry

between both assets. Mathematically, this requirement translates to
   
Y XY (T ) X YX (T )
Lt = Lt , (3.5)
XY (t) YX (t)
XY (T )
meaning that the price increment XY (t) under the probability measure PY
should have the same distribution as the price increment YYXX(T )
(t) under the
probability measure PX . Therefore we need to have a description of the dy-
namics of the inverse price, YX (t), that would be analogous to the dynamics
of the original price XY (t). Ideally, the evolution of this price should have the
same form as (3.3), but the dynamics are already determined by Ito’s formula
(see Appendix):

dYX (t) = dXY (t)−1 = −XY (t)−2 dXY (t) + 1

2 · 2XY (t)−3 d2 XY (t)
= −σYX (t)dW Y (t) + σ YX (t)dt
2


= σYX (t) · −dW Y (t) + σdt . (3.6)

Given the exchangeability argument of X and Y , we should also have

dYX (t) = σYX (t)dWtX , (3.7)

which is the same stochastic differential equation as (3.3) with X and Y

flipped, and with a different Brownian motion W X (t) under the measure PX .
The solution of the above stochastic differential equation is given by


YX (t) = YX (0) · exp σW X (t) − 21 σ 2 t . (3.8)

In diffusion models, each reference asset Y has its own market noise that
is represented by one or several Brownian motions W i,Y (t). Other reference
assets, such as an asset X, have different market noise that is represented
by Brownian motions W i,X (t). Obviously, the Brownian motions W Y and
W X are related. In the above case, we just have one Brownian motion for
each asset, and the relationship between W X (t) and W Y (t) follows from the
equation


dYX (t) = σYX (t) · −dW Y (t) + σdt = σYX (t)dW X (t). (3.9)

Thus we must have

dW X (t) = −dW Y (t) + σdt,
or in other words,
W X (t) = −W Y (t) + σt. (3.10)
96 Stochastic Finance: A Numeraire Approach

Note that a symmetric relationship holds as well

W Y (t) = −W X (t) + σt. (3.11)

REMARK 3.1 Some authors define dW X (t) as dW Y (t) + σdt which is

an equivalent definition since the Brownian motion is symmetric and thus
dW Y (t) has the same distribution as −dW Y (t). However, such a definition
would break the symmetry of the price formulas for X and Y , and thus it is
more appropriate to use dW X (t) = −dW Y (t) + σdt.

The two Brownian motions W Y (t) and W X (t) are perfectly correlated with
a correlation coefficient of -1:
dW Y (t) · dW X (t) = −1 · dt.
This makes sense since when XY (t) goes up, the inverse price YX (t) goes
down, and vice versa.

From the financial representation of the Radon–Nikodým derivative we have

dPXt XY (t)


Z(t) = Y
= = exp σW Y (t) − 21 σ 2 t = exp −σW X (t) + 12 σ 2 t .
dPt X Y (0)
(3.12)
The concept of equivalent treatment of both X and Y is also supported by
the following theorem.

THEOREM 3.1 (Girsanov).

Let W Y (t) be a PY Brownian motion. Then W X (t) = −W Y (t) + σt is a PX
dPX XY (t) Y 1 2


Brownian motion, where Z(t) = dPtY = X Y (0) = exp σW (t) − 2 σ t .
t

REMARK 3.2 The two measures PY and PX may disagree on the drift
of the Brownian motion. More specifically,
EY [W X (t)] = EY [−W Y (t) + σt] = σt,
but
EX [W X (t)] = 0.
The last statement can be proved by a change of the measure argument (1.66)
EX [W X (t)] = EX [−W Y (t) + σt] = EY [(−W Y (t))Z(t)] + σt


= EY [−W Y (t) · exp σW Y (t) − 12 σ 2 t ] + σt


= − exp(− 21 σ 2 t) · EY [W Y (t) · exp σW Y (t) ] + σt


= − exp(− 12 σ 2 t) · dσ
d Y
E [exp σW Y (t) ] + σt
= − exp(− 21 σ 2 t) · d 1 2
dσ [exp( 2 σ t)] + σt
= 0.
 Diffusion Models 97

Note that we have

d2 XY (t)
= σ 2 dt, (3.13)
XY (t)2
as well as
d2 YX (t)
= σ 2 dt, (3.14)
YX (t)2
and thus the volatility of XY (t) is the same as the volatility of YX (t). It does
not matter which of the two assets, X or Y , is chosen as a numeraire. For
instance the volatility of the dollar/euro exchange rate is the same as the
volatility of the euro/dollar exchange rate. This is true even when the volatil-
ity is stochastic.

Having the closed form expressions for the price XY (T ) from Equation (3.3)
and for the price YX (T ) from Equation (3.8), we can determine the prices of
Arrow–Debreu securities that pay off either IA (ω) units of an asset Y at time
T , or IA (ω) units of an asset X at the same time. Let us consider events A of
the form
A = {ω ∈ Ω : XY (T, ω) ≥ K}
for a given constant K. A is a set of scenarios where the terminal price of
XY (T ) exceeds a level K. Let us determine the price of a contract U that
pays off
UT = IA (ω) · YT .
Since the price of this contract is a martingale under the PY measure, we have

UY (t) = EYt [IA (ω)] = PYt (A).

The event
A = {XY (T ) ≥ K}
is equivalent to


XY (t) · exp σ(W Y (T ) − W Y (t)) − 21 σ 2 (T − t) ≥ K,

or in other words
W Y (T ) − W Y (t) 1   √
− √ ≤ √ · log XYK(t) − 12 σ T − t.
T −t σ T −t
Y Y
Since − W (T√)−W
T −t
(t)
has a normal distribution with zero mean and a unit
variance N (0, 1) under the probability measure PY , the probability of the
event A is given by
   √ 
PYt (A) = PYt (XY (T ) ≥ K) = N σ√1T −t · log XYK(t) − 12 σ T − t , (3.15)
98 Stochastic Finance: A Numeraire Approach

where N (·) is a cumulative distribution function of a standard normal variable

Z x y2
N (x) = √1
2π
· e− 2 dy.
−∞

We can determine the price of the Arrow–Debreu security V that pays off
IA (ω) units of X at time T in a similar fashion. At time T we have

VT = IA (ω) · XT .

The price VX (t) is a PX martingale, and thus

VX (t) = EX X
t [IA (ω)] = Pt (A).

The event
A = {XY (T ) ≥ K}
is equivalent to


XY (t) · exp −σ(W X (T ) − W X (t)) + 21 σ 2 (T − t) ≥ K.

Here we used the fact that

1 1
XY (T ) = =

YX (T ) YX (t) · exp σ(W (T ) − W X (t)) − 21 σ 2 (T − t)
X


= XY (t) · exp −σ(W X (T ) − W X (t)) + 12 σ 2 (T − t) .

We need to express the price of XY (T ) in terms of the Brownian motion

W X (t) in order to determine the probability of the event A using the PX
measure. The event A is equivalent to

W X (T ) − W X (t) 1   √
√ ≤ √ · log XYK(t) + 21 σ T − t.
T −t σ T −t
Therefore
   √ 
XY (t)
PX X
t (A) = Pt (XY (T ) ≥ K) = N
√1
σ T −t
· log K + 12 σ T − t . (3.16)

The Arrow–Debreu securities UT = IA (ω) · XT and VT = IA (ω) · YT are also

known as digital options. Determination of their hedging portfolios is the
subject of Exercise 3.4.

REMARK 3.3
It is interesting to note that when K = XY (0), we have
 √ 
PY (XY (T ) ≥ XY (0)) = N − 21 σ T < 21 ,
 Diffusion Models 99

and  √  1
PX (XY (T ) ≥ XY (0)) = N 1
2σ T > 2.
A delivery of a unit of Y when the price XY of the asset X moves up requires
less than a 21 unit of Y to start with. On the other hand, a delivery of a unit
of X on the same event requires more than a 21 unit of an asset X. In this
sense, the asset Y is “cheaper” (it requires a smaller fraction of the underlying
asset) to deliver than the asset X on the up movement of the price XY .

3.2 General European Contracts

A general European-type contract pays off either f Y (XY (T )) units of an
asset Y , or f X (YX (T )) units of an asset X at time T . In order that these two
payoffs correspond to the same contract, we must have
f Y (XY (T )) · Y = f X (YX (T )) · X
or in other words,
 
1
f Y (XY (T )) · Y = f X · XY (T ) · Y.
XY (T )
Therefore the two payoff functions f Y and f X are linked by the following
symmetric relationship



f Y (x) = f X x1 · x, or f X (x) = f Y x1 · x, (3.17)

which is valid for 0 < x < ∞, meaning that neither the asset X nor the
asset Y is worthless. Note that the payoff function depends on a choice of
the reference asset. The formulas that link functions f Y and f X are known
as a perspective mapping. A financial contract with a non-negative payoff
function f Y (x) is known as an option. Note that f Y (x) ≥ 0 is equivalent to
f X (x) ≥ 0, so the definition of the option does not depend on the choice of
the reference asset. An option of this type is also known as a plain vanilla
option. The perspective mapping also preserves convexity; f Y (x) is convex
if and only if f X (x) is convex (see Exercise 3.1).

Example 3.1 The best asset and the worst asset

The simplest contract on two assets one can think of is the best of the two
assets, or the worst of the two assets. The best of the two assets contract
pays off max(XT , YT ) at time T ; the worst of the two assets contract pays off
min(XT , YT ) at time T . These contracts are completely symmetric since
max(XT , YT ) = max(YT , XT ),
100 Stochastic Finance: A Numeraire Approach

and
min(XT , YT ) = min(YT , XT ).
When the best of the two assets contract is settled in the asset Y , the contract
pays off max (XY (T ), 1) units of Y . Similarly, when the best of the two assets
contract is settled in the asset X, the contract pays off max (YX (T ), 1) units
of X. The payoff functions for the best of the two assets are thus given by

f Y (x) = max(x, 1),

and
f X (x) = f Y ( x1 ) · x = max( x1 , 1) · x = max(1, x).
Note that we have f X (x) = f Y (x). Analogously, the payoff functions for the
worst of the two assets are given by

f Y (x) = min(x, 1),

f X (x) = f Y ( x1 ) · x = min(x, 1).

Note that the payoff of the best asset contract can be re-expressed in the
following form

max(XT , YT ) = (XT − YT )+ + YT = (YT − XT )+ + XT ,

where x+ = max(x, 0), leading us to contracts known as the call and the put
options.

The most typical traded contract that has the feature of paying the best
asset is a convertible bond. One of the payments of the convertible bond is
max(ST , K · BTT ), so the holder of this contract can choose between the equity
position in the asset S, and K units of the bond B T at the expiration time
T.

However, the logic of the financial markets is to allow for maximal leverage,
and in this respect, the contract that delivers the best asset is not ideal as it
ties down a portion of the capital of the investor that can be used otherwise.
Instead, one can trade just the differences between the best asset and the asset
itself, which requires significantly less capital. The contract on the difference
of the best asset and the asset itself is known as a call option. Formally, a
European call option V EC (X, K · Y, T ) is a contract that pays off

(XT − K · YT )+ , (3.18)

where X and Y are two assets. The constant K is known as the strike. The
relationship between the European call option and the contract that delivers
the best asset is given by

max(XT , K · YT ) = (XT − K · YT )+ + K · YT .
 Diffusion Models 101

In the contract that delivers we may rescale one of the assets by a factor of
K to achieve a better proportionality of the assets X and Y . Note that the
European call option is a combination of two Arrow–Debreu securities

(XT − K · YT )+ = I(XY (T ) ≥ K) · X − K · I(XY (T ) ≥ K) · Y. (3.19)

The first Arrow–Debreu security pays off I(XY (T ) ≥ K) units of the asset X,
the second Arrow–Debreu security pays off I(XY (T ) ≥ K) units of the asset Y .

A closely related contract to a European call option is a European put

option V EP (K · Y, X, T ) with a payoff

(K · YT − XT )+ . (3.20)

The put option is also related to the contract that delivers the best asset by

max(XT , K · YT ) = (K · YT − XT )+ + XT .

The only difference between the call and the put option is which of the two
available assets is chosen to be subtracted from the payoff of the contract
on the best asset. Since this choice is arbitrary, the call option on assets X
and K · Y is the same contract as a put option on assets K · Y and X. This
relationship is known as the put-call duality:

V EC (X, K · Y, T ) = V EP (K · Y, X, T ) = K · V EP (Y, X
K , T ). (3.21)

Another simple relationship between European call and European put op-
tions is a put-call parity. Note that

XT − K · YT = (XT − K · YT )+ − (K · YT − XT )+ , (3.22)

where XT − K · YT is a payoff of a forward contract F (X, K · Y, T ). The

relationship between the forward contract and the corresponding call and put
options holds at all times t ≤ T :

F (X, K · Y, T ) = V EC (X, K · Y, T ) − V EP (X, K · Y, T ). (3.23)

A European call option payoff can be written in the following equivalent

ways
+ +
(XT − K · YT )+ = (XY (T ) − K) · YT = (1 − K · YX (T )) · XT . (3.24)
102 Stochastic Finance: A Numeraire Approach

When the European call option is settled in the asset Y , the payoff is given
by
(XY (T ) − K)+ · Y (3.25)

which corresponds to a payoff function f Y (x) = (x−K)+ . The holder receives

(XY (T )−K)+ units of Y at time T . Similarly, the European call option settled
in the asset X has the payoff

(1 − K · YX (T ))+ · X (3.26)

which corresponds to a payoff function f X (x) = (1 − K · x)+ . Note that

f X ( x1 ) · x = (1 − K · x1 )+ · x = (x − K)+ = f Y (x). The holder receives
+
(1 − K · YX (T )) units of X at time T .

European-type contracts can always be expressed in terms of two no-

arbitrage assets.

REMARK 3.4 European option as a contract on two no-arbitrage

assets
A European option can always be expressed as a contract on two no-arbitrage
assets. The payoff of a European option is defined as f Y (XY (T )) units of
an asset Y , or f X (YX (T )) units of an asset X at time T for general assets
with positive price X and Y . When X or Y is an arbitrage asset, such as the
dollar $, we can substitute an arbitrage asset X (or Y ) with a corresponding
no-arbitrage asset U or V that delivers a unit of an asset X or an asset Y at
time T . In particular, we have

U T = XT , VT = YT .

Thus the European option payoff can be re-expressed as f Y (UV (T )) =

f V (UV (T )) units of an asset V , or f X (VU (T )) = f U (VU (T )) units of an
asset U at time T for two no-arbitrage assets U and V . This substitution is
not possible when there is no fixed delivery of the option payoff such as in the
case of American options.

Example 3.2 European call option in different markets

Stock option When the asset X is a stock S, and the asset Y is a dollar $,
we have a European stock option

(ST − K · $T )+ . (3.27)

Note that the existing literature typically omits the fact that the strike
is in fact multiplied by the dollar $. This notation means that the holder
of the option has the right to increase his position in the stock S by one
 Diffusion Models 103

unit, and decrease his position in the dollar $ by K units at time T .

Should the contract be settled in dollars, one can write the payoff as
+
(S$ (T ) − K) · $T . (3.28)

The holder receives (S$ (T ) − K)+ units of the dollar $ at time T . As

noted earlier, the European option on a stock may also be settled in
terms of a bond B T , a contract that delivers 1 $ at time T , so that
BTT = 1$T . In this case, the payoff of the option may be written as

(ST − K · BTT )+ . (3.29)

This fact is useful in the pricing of this option. In contrast to the dollar,
the bond does not create arbitrage opportunities in time. Therefore it
can be used as a natural reference asset for pricing this option. The
option can be settled entirely in the bond
+
(SB T (T ) − K) · BTT , (3.30)

or in the stock
+
1 − K · BST (T ) · ST . (3.31)

Exchange option When the asset X is a stock S 1 , and the asset Y is another
stock S 2 , the corresponding European call option

(ST1 − K · ST2 )+ (3.32)

is known as an exchange option. The natural reference asset for pric-

ing this option is either the stock S 1 , or the stock S 2 . Adding another
reference asset, such as a dollar $, for pricing this option would only
increase the dimensionality of the problem.
Currency option When the asset X is a euro e, and the asset Y is a dollar
$ (or any other currencies), we have a European currency option

(eT − K · $T )+ . (3.33)

A European currency option can be settled in the dollar or in the euro

only

(eT − K · $T )+ = (e$ (T ) − K)+ · $T = (1 − K · $e (T ))+ · eT .

In order to express the payoff in terms of no-arbitrage assets only, we

can take a foreign bond B e,T that delivers a unit of a foreign currency
e at time T , and a domestic bond B T that delivers a unit of a domestic
currency $ at time T . The payoff of the currency option is equivalent to

(eT − K · $T )+ = (BTe,T − K · BTT )+ .

104 Stochastic Finance: A Numeraire Approach

Caplet A caplet is an option on a LIBOR that pays off

(L(T, T ) − K)+ · $T +δ . (3.34)

The LIBOR L(T, T ) is observed at time T , but the contract is settled

at a later time T + δ in a dollar $. Here it is not entirely obvious what
the corresponding assets X and Y should be. But from the definition of
the LIBOR
B$T (T ) − B$T +δ (T )
L(T, T ) = = [B T − B T +δ ]δB T +δ (T ),
δB$T +δ (T )

and using the fact that BTT +δ

+δ
= $T +δ , we can rewrite the payoff as

(L(T, T ) − K)+ · $T +δ
 +
= [B T − B T +δ ]δB T +δ (T ) − K · BTT +δ
+δ

 +
= δ1 · [B T − B T +δ ]T − K · δBTT +δ
+δ
.

Thus the asset X is a combination of two bonds [B T − B T +δ ], and the

asset Y is δB T +δ .

REMARK 3.5 European call option price

We have already seen that a European call option is just a combination of
two Arrow–Debreu securities
+
(XT − K · YT ) = I(XY (T ) ≥ K) · XT − K · I(XY (T ) ≥ K) · YT . (3.35)

The first Arrow–Debreu security costs PXt (XY (T ) ≥ K) units of the asset X,
the second Arrow–Debreu security costs PY (XY (T ) ≥ K) units of the asset
Y . Therefore we have the following result:

THEOREM 3.2 Black–Scholes formula

The price of a European option contract V EC (X, K · Y, T ) with the payoff
(XT − K · YT )+ is given by

V EC (X, K · Y, T ) = PX Y
t (XY (T ) ≥ K) · X − KPt (XY (T ) ≥ K) · Y. (3.36)

Recall from the previous section (Equations (3.16) and (3.15)) that for the
geometric Brownian motion model, we have
   √ 
XY (t)
PX √1
t (XY (T ) ≥ K) = N σ T −t · log K + 1
2 σ T − t ,
 Diffusion Models 105

and    √ 
XY (t)
PYt (XY (T ) ≥ K) = N √1
σ T −t
· log K − 21 σ T − t .
Thus in the geometric Brownian motion model, the Black–Scholes formula
simplifies to

V EC (X, K · Y, T ) = [N (d+ )] · Xt + [−K · N (d− )] · Yt , (3.37)

where

√
d± = √1 · log 1
· XY (t) ± 21 σ T − t. (3.38)
σ T −t K

REMARK 3.6 Option on a dividend-paying asset

When one of the underlying assets is a stock S that pays dividends, which is
an arbitrage asset, we can transform the problem using a no-arbitrage asset Se
that represents the stock S plus the dividends. Assume for instance a constant
continuous dividend yield δ = a(t), in which case the relationship between S
and Se in (1.31) simplifies to

SeT = eδT ST . (3.39)

Consider a European call option V with a payoff

VT = (ST − K · $T )+ ,

which is written in terms of two arbitrage assets: the dividend paying stock
S and the dollar $. Given the relationship between S and Se in (3.39), we can
rewrite the option payoff in terms of two no-arbitrage assets as

VT = (e−δT SeT − K · BTT )+ .

The Black–Scholes formula now applies, and the price of the call option at
time t = 0 is given

PS̃ (SeB T (T ) ≥ eδT K) · e−δT Se0 − KPTt (SeB T (T ) ≥ eδT K) · B0T

= PS̃ (SeB T (T ) ≥ eδT K) · e−δT S0 − KPTt (SeB T (T ) ≥ eδT K) · B0T .

The only difference is that the dividend yield modifies the strike from K to
eδT K, and the stock S is discounted by the factor e−δT . In the geometric
Brownian motion model we get at time t

Vt = [N (d+ )] · eδ(T −t) St + [−K · N (d− )] · BtT , (3.40)

where
√
d± = √1 · log 1
· XY (t) + (−δ ± 21 σ) T − t. (3.41)
σ T −t K
106 Stochastic Finance: A Numeraire Approach

REMARK 3.7 Money as a reference asset

We have seen that European-type contracts can be expressed in terms of
two no-arbitrage assets X and Y which also serve as natural reference assets
for pricing a given European option V . Thus for pricing a general European
contract, one first determines the price VY (t) or VX (t) in terms of the no-
arbitrage assets Y or X. The dollar price V$ (t) follows immediately from the
change of numeraire formula

V$ (t) = VY (t) · Y$ (t) = VX (t) · X$ (t).

Let us illustrate how to compute the dollar price of a European call option on
a stock and a dollar with a payoff

VTEC = (ST − K · $T )+ .

Since a dollar does not have a martingale measure P$ , we have to compute

the price of the European call option using the First Fundamental Theorem
of Asset Pricing either in terms of a stock S, or a bond B T . This leads to the
Black–Scholes formula (3.37), which takes the following form:

VtEC = [N (d+ )] · St + [−K · N (d− )] · BtT ,

where
√
d± = √1 · log 1
· SB T (t) ± 12 σ T − t.
σ T −t K

We can rewrite the Black–Scholes formula in terms of prices with respect to

a bond B T as

VBEC
T (t) = N (d+ ) · SB T (t) − K · N (d− ) .

Multiplying the above equation by the dollar price of the bond B$T (t) and using
the change of numeraire formula, we obtain the dollar price of the European
call option

V$EC (t) = VBEC T

T (t) · B$ (t)

= N (d+ ) · SB T (t) · B$T (t) − K · N (d− ) · B$T (t)

= N (d+ ) · S$ (t) − K · N (d− ) · B$T (t).

The formula for d± can also be expressed in terms of dollar prices as

√
d± = σ√1T −t · log K
1
· S$ (t) · $B T (t) ± 12 σ T − t.

If we further assume a deterministic term structure evolution with a constant

interest rate r,
BtT = e−r(T −t) · $t ,
 Diffusion Models 107

the above relationships simplify to

V$EC (t) = S$ (t) · N (d+ ) − K · e−r(T −t) · N (d− ) , (3.42)

with



d± = √1
σ T −t
· log 1
K · S$ (t) + r ± 21 σ 2 (T − t) . (3.43)

This is the Black–Scholes formula expressed in terms of the dollar prices. Note
that we had to assume a deterministic interest rate r in order to simplify the
Black–Scholes formula (3.37) that applies also to stochastic interest rates.

Table 3.1 summarizes payoffs of various contracts. Options with the power
and the logarithmic payoff do not appear directly on the market, but they are
related to barrier and lookback options as we will see in the following text.
Note that the payoff function f Y (x) that corresponds to Y being chosen as
a reference asset may have a different form than the payoff function f X (x)
that corresponds to X being chosen as a reference asset. But the two payoff
functions f Y and f X represent the same contract. One can think of switch-
ing roles of the assets X and Y , in which case we would get a new contract
with a payoff f X (XY (T )) units of Y . This is a dual contract to the original
contract that pays off f Y (XY (T )) units of Y . When we know the price of an
original contract, we also know the price of the dual contract by switching the
roles of X and Y .

We have already seen that the call option with a payoff f Y (x) = (x − K)+
is a dual contract to the put option with a payoff f X (x) = (1 − K · x)+ .
The contract that pays off the best asset max(XT , YT ) is dual to itself as
f Y (x) = f X (x) = max(x, 1). The role of X and Y can be switched and it
does not change the contract as max(XT , YT ) = max(YT , XT ). Similarly, the
worst asset min(XT , YT ) is also dual to itself. The following example illus-
trates the concept of the dual contracts of the power options.

TABLE 3.1: Contracts on two assets.

Contract Payoff f Y (x) f X (x)
Digital IA (XY (T )) · YT IA (x) IA ( x1 ) · x
Best Asset max(XY , K · YT ) max(x, K) max(K · x, 1)
Worst Asset min(XY , K · YT ) min(x, K) min(K · x, 1)
Call (XT − K · YT )+ (x − K)+ (1 − K · x)+
Put (K · YT − XT )+ (K − x)+ (K · x − 1)+
Forward XT − K · YT x−K 1−K ·x
Power [XY (T )]α · YT xα x1−α
Logarithm log(XY (T )) · YT log(x) −x · log(x)
108 Stochastic Finance: A Numeraire Approach

Example 3.3 Dual contracts of power options

A power option Rα pays off

RTα = [XY (T )]α · YT .

Power options are useful in pricing barrier and lookback options. The dual
contract switches the roles of the assets X and Y ; it pays off

[YX (T )]α · XT .

This can be rewritten as

[YX (T )]α · XT = [XY (T )]−α · XY (T ) · YT = [XY (T )]1−α · YT .

Thus the payoff function xα has a dual payoff function x1−α .

Note that when α = 0, the corresponding power option R0 coincides with

the asset Y . When α = 1, the corresponding power option R1 is the asset
X. This suggests that for α ∈ (0, 1), the resulting power option Rα creates
an asset that is a combination of the assets X and Y . When α > 1, the
power option Rα leverages the position in the asset X. Similarly, when α < 0,
the power option Rα leverages the position in the asset Y . This is supported
by the following argument. When X is comparable to Y in terms of price,
meaning XY (T ) ≈ 1, we have

[XY (T )]α ≈ 1 + α(XY (T ) − 1) = (1 − α) + αXY (T )

according to the first order Taylor expansion around 1. Rewriting this rela-
tionship in terms of the assets, we have

RTα ≈ (1 − α) · YT + α · XT . (3.44)

Clearly, when α ∈ (0, 1), the power option is approximately a linear combi-
nation of the assets X and Y with positive weights. In particular, the power
option R1/2 corresponding to a square root asset is approximately just an
average of the two assets X and Y . The square root asset is the only power
option that is dual to itself, meaning that one can swap the roles of the assets
X and Y without changing the contract.

When α > 1, the power option corresponds to having a long position in

the asset X, and a short position in the asset Y . When α < 0, the situation
is reversed, and the power option represents a long position in the asset Y ,
and a short position in the asset X. The hedging position ∆X (t) of the power
option indeed has the same sign as α, and the hedging position ∆Y (t) has
the same sign as 1 − α. This confirms that the approximation from (3.44) is
reasonable. The reader should determine the price and the hedging portfolio
of the power options in Exercise 5.1.
 Diffusion Models 109

3.3 Price as an Expectation

For pricing a general European claim V , we can use either reference asset
Y or X in order to determine the price of V :

V = VY (t) · Y = VX (t) · X.

In Markovian models, which include geometric Brownian motion, we can ex-

press these prices in terms of the price functions uY and uX defined as

VY (t) = uY (t, XY (t)), VX (t) = uX (t, YX (t)).

The functions uY and uX are linked by

uY (t, XY (t)) = uX (t, YX (t)) · XY (t),

or by
uX (t, YX (t)) = uY (t, XY (t)) · YX (t).
Therefore we have the following symmetric relationship

uY (t, x) = uX (t, x1 ) · x, or uX (t, x) = uY (t, x1 ) · x (3.45)

for 0 < x < ∞, which is known as a perspective mapping. Note that

we have uY (T, x) = f Y (x), and uX (T, x) = f X (x), so the terminal price
of the contract agrees with the payoff function. We have already seen that
f Y (x) = f X ( x1 ) · x, and f X (x) = f Y ( x1 ) · x, which is just a special case of the
relationship between the prices uY (t, x), and uX (t, x).

Recall that the payoff of European options can always be written in terms
of two no-arbitrage assets: U that agrees to deliver an asset X at time T ,
and V that agrees to deliver an asset Y at time T . It is easy to see that the
contract to deliver a no-arbitrage asset is the asset itself, so the substitution
of the underlying for a no-arbitrage asset makes sense only when one of the
underlying assets is an arbitrage asset, such as in the case of the dollar or
other currencies. Therefore without loss of generality, we may assume that
the European option is settled in terms of two no-arbitrage assets.

Given that European options can be expressed in terms of two no-arbitrage

assets, the First Fundamental Theorem of Asset Pricing states that the price
of V in terms of the reference asset Y is a PY martingale, and the price of V in
terms of the reference asset X is a PX martingale. This gives us a stochastic
representation of the contingent claim price
 
VY (t) = EYt [VY (T )] = EYt f Y (XY (T )) , (3.46)
110 Stochastic Finance: A Numeraire Approach

when the asset Y is used as a numeraire, and

 X 
VX (t) = EX X
t [VX (T )] = Et f (YX (T )) , (3.47)

when the asset X is used as a numeraire. The number of units

EYt f Y (XY (T )) of Y that is needed in order to acquire the contract V
is the price of the contract
 X in terms
 of the reference asset Y . Similarly, the
number of units EX t f (Y X (T )) of X that is needed in order to acquire the
contract V is the price of the contract in terms of the reference asset X.

When the prices VY (t) and VX (t) are Markovian in the prices XY (t) and
YX (t), the price functions uY and uX have the following representations
 
uY (t, x) = EY f Y (XY (t)) |XY (t) = x , (3.48)

and  
uX (t, x) = EX f X (YX (t)) |YX (t) = x . (3.49)
The price processes VY and VX are indeed Markovian in the geometric Brow-
nian motion model.

When the price processes XY (t) and YX (t) are geometric Brownian motions,
we can compute the price functions uY and uX directly by computing the
conditional expected value. For the function u we have
 
uY (t, x) = EY f Y (XY (T )) |XY (t) = x (3.50)
Y
 Y Y 1 2



= E f XY (t) · exp σW (T − t) − 2 σ (T − t) |XY (t) = x



= EY f Y x · exp σW Y (T − t) − 21 σ 2 (T − t) |XY (t) = x
Z ∞   √   2
= f Y x · exp σy T − t − 12 σ 2 (T − t) · √12π exp − y2 dy.
−∞

We have used the fact that

XY (T ) = XY (t) · exp σW Y (T − t) − 21 σ 2 (T − t) ,

W Y (T −t)
and that √
T −t
has a normal distribution N (0, 1).

Similarly, the function uX can be determined from the following formula:

 
uX (t, x) = EX f X (YX (T )) |YX (t) = x (3.51)
X
 X X 1 2



= E f YX (t) · exp σW (T − t) − 2 σ (T − t) |YX (t) = x



= EX f X x · exp σW X (T − t) − 21 σ 2 (T − t) |YX (t) = x
Z ∞   √   2
= f X x · exp σy T − t − 12 σ 2 (T − t) · √12π exp − y2 dy.
−∞
 Diffusion Models 111

Example 3.4
Consider a European call option with a payoff (XT − K · YT )+ . When Y is
chosen as a reference asset, the payoff function is given by f Y (x) = (x − K)+ .
Thus we have
Z ∞   √   2
uY (t, x) = f Y x · exp σy T − t − 21 σ 2 (T − t) · √12π exp − y2 dy
−∞
Z ∞  √  +
= x · exp σy T − t − 12 σ 2 (T − t) − K
−∞
 2
× √12π exp − y2 dy
 
1 2 
= x · N σ√1T −t · log Kx
+ 2 σ (T − t)
 
1 2 
−K · N σ√1T −t · log K x
− 2 σ (T − t) .

When X is chosen as a reference asset, the payoff function is given by f X (x) =

f Y ( x1 ) · x = (1 − K · x)+ , and thus we have
Z ∞   √   2
X
u (t, x) = f X x · exp σy T − t − 12 σ 2 (T − t) · √1
2π
· exp − y2 dy
−∞
Z∞   √ +
= 1 − K · x · exp σy T − t − 21 σ 2 (T − t)
−∞
 2
× √12π exp − y2 dy
 
1 2 
= N σ√1T −t · log K·x
1
+ 2 σ (T − t)
 
1 2 
1
−K · x · N σ√1T −t · log K·x − 2 σ (T − t) .

The reader may check that the price functions uY and uX indeed satisfy
uX (t, x) = uY (t, x1 ) · x.

3.4 Connections with Partial Differential Equations

Let us assume that the price XY (t) follows the geometric Brownian motion
model
dXY (t) = σXY (t)dW Y (t).

We point out in this section that the price functions uY and uX satisfy a
certain partial differential equation.
112 Stochastic Finance: A Numeraire Approach

THEOREM 3.3  
The price function uY (t, x) = EY f Y (XY (T )) |XY (t) = x satisfies the par-
tial differential equation

uYt (t, x) + 21 σ 2 x2 uYxx (t, x) = 0 (3.52)

with the terminal condition

uY (T, x) = f Y (x), (3.53)
and the boundary condition
uY (t, 0) = f Y (0). (3.54)
 
The price function uX (t, x) = EX f X (YX (T )) |YX (t) = x satisfies the par-
tial differential equation

uX 1 2 2 X
t (t, x) + 2 σ x uxx (t, x) = 0 (3.55)

with the terminal condition

uX (T, x) = f X (x), (3.56)
and the boundary condition
uX (t, 0) = f X (0). (3.57)

REMARK 3.8
The partial differential equations (3.52) and (3.55) are also known as the
Black–Scholes partial differential equations.

PROOF Let
 
uY (t, x) = EY f Y (XY (T )) |XY (t) = x
be the price of the contract with respect to the reference asset Y . According
to Ito’s formula, the option price has the following dynamics:
duY (t, XY (t)) = uYt (t, XY (t)) dt + uYx (t, XY (t)) dXY (t)
+ 12 uYxx (t, XY (t)) d2 XY (t)
 
= uYt (t, XY (t)) + 12 σ 2 XY (t)2 uYxx (t, XY (t)) dt
+uYx (t, XY (t)) dXY (t).
Since uY (t, XY (t)) is a PY martingale, the dt term of this equation must van-
ish for all values of XY (t), and thus the following partial differential equation
for the price of the option must hold:
uYt (t, x) + 12 σ 2 x2 uYxx (t, x) = 0,
 Diffusion Models 113

with the terminal condition

uY (T, x) = f Y (x).

The case when x = 0 represents the situation when XY (t) = 0 (the asset X
becomes worthless), and thus the value of XY (T ) will also be zero. Thus the
payoff of the option will be f Y (0) units of an asset Y at time T . Thus the
value of the contract at time t is uY (t, 0) = f Y (0).

We can apply the same technique using the no-arbitrage asset X as a nu-
meraire when the payoff of the contract is f X (YX (T )) units of an asset X,
leading to the partial differential equation (3.55).

REMARK 3.9 The prices of X and Y satisfy the Black–Scholes

partial differential equation
Partial differential equation (3.52)

uYt (t, x) + 12 σ 2 x2 uYxx (t, x) = 0

has two trivial solutions that correspond to the payoff functions f Y (x) = 1
and f Y (x) = x. When the payoff function is f Y (x) = 1, the price function
uY (t, x) is also identically equal to one, and the partial differential equation
(3.52) is satisfied. In financial terms, the payoff function f Y (x) = 1 corre-
sponds to the delivery of a unit of an asset Y at time T . This is a contract
to deliver an asset Y , and its price at any given time t ≤ T is one unit of an
asset Y . Thus we have uY (t, x) = 1 as a solution. When the payoff function
is f Y (x) = x, the price function uY (t, x) is also equal to x, and the partial
differential equation (3.52) is satisfied. In financial terms, the payoff function
f Y (x) = x corresponds to the delivery of a unit of an asset X at time T (it is
XY (t) units of an asset Y ). This is a contract to deliver an asset X at time T
and its price at any given time t ≤ T is one unit of an asset X. Thus we have
uY (t, x) = x as a solution.

Similarly, the partial differential equation (3.55)

uX 1 2 2 X
t (t, x) + 2 σ x uxx (t, x) = 0

also has two trivial solutions that correspond to the payoff functions f X (x) =
1 and f X (x) = x. In financial terms, the payoff function f X (x) = 1 corre-
sponds to the delivery of an asset X, the payoff function f X (x) = x corre-
sponds to the delivery of an asset Y .

Example 3.5
The European option V with a payoff VT = (XT − K · YT )+ has an associated
payoff function f Y (x) = (x − K)+ , or f X (x) = (1 − K · x)+ . The VY (t) =
114 Stochastic Finance: A Numeraire Approach

uY (t, XY (t)) price satisfies the partial differential equation (3.52) and the
VX (t) = uX (t, YX (t)) price satisfies the partial differential equation (3.55).
When the asset X becomes worthless, or in other words when XY (t) = 0, the
option will also be worthless as f Y (0) = 0, giving us the boundary condition
uY (t, 0) = 0. The asset X will not serve as a reference asset in this case, but
the price of the contract can still be expressed in terms of the asset Y . On
the other hand, when the asset Y becomes worthless, YX (t) = 0, the option
will pay off a unit of the asset X, which corresponds to f X (0) = 1. This
gives the boundary condition uX (t, 0) = 1, the asset X can still be used as a
numeraire. Note that the boundary conditions when one of the prices is zero
do not have a perspective mapping counterpart as the perspective mapping
applies only to cases when the prices are positive. When one of the assets
becomes worthless, it still makes sense to use the remaining asset with a
positive price as a numeraire, but the pricing problem cannot be solved using
the worthless asset.

3.5 Money as a Reference Asset

It is also possible to write the Black–Scholes partial differential equation in
terms of the dollar $ as a reference asset. Let X be a stock S, and Y be a
bond B T . A contract V that pays off f T (SB T (T )) units of a bond B T at time
T can equivalently be expressed as

VT = f T (SB T (T )) · BTT = f $ (S$ (T )) · $T ,

a contract that pays off f $ (S$ (T )) units of a dollar $ at time T . The payoff
functions in terms of a bond B T and a dollar $ agree: f T (x) = f $ (x). The
contract V at time t can be also expressed in the following equivalent ways:

Vt = VB T (t) · BtT = V$ (t) · $t = VS (t) · St .

Let uT (t, SB T (t)) = VB T (t) be the price of the contract V in terms of a bond
B T , and let
v $ (t, S$ (t)) = V$ (t) (3.58)
be the price of the contract V in terms of a dollar $. We are using a different
letter v for the dollar price in order to distinguish it from the prices u that
use only no-arbitrage assets. Let us also assume B$T (t) = e−r(T −t) . Since

Vt = uT (t, SB T (t)) · BtT = v $ (t, S$ (t)) · $t ,

we get the following relationship between uT and v $ :

v $ (t, x) = e−r(T −t) · uT (t, er(T −t) x), (3.59)

 Diffusion Models 115

and
uT (t, x) = er(T −t) · v $ (t, e−r(T −t) x). (3.60)

We have seen that the price function uT satisfies the partial differential
equation
uTt (t, x) + 12 σ 2 x2 uTxx (t, x) = 0.

Using the relationship between the functions uT and v $ , we find that

uTt (t, x) = er(T −t) · −rv $ (t, e−r(T −t) x) + vt$ (t, e−r(T −t) x)
!
 
−r(T −t) $ −r(T −t)
+r e x vx (t, e x) ,

and
 2
uTxx (t, x) = er(T −t) e−r(T −t) vxx
$
(t, e−r(T −t) x).

After substitution of x for e−r(T −t) x, we conclude that v $ satisfies the follow-
ing partial differential equation

−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx

$
(t, x) = 0. (3.61)

The terminal condition is given by

v $ (T, x) = f T (x) = f $ (x), (3.62)

and the boundary condition is

v $ (t, 0) = e−r(T −t) · uT (t, 0) = e−r(T −t) · f T (0). (3.63)

The Black–Scholes partial differential equation in the form of (3.61) is widely

used since it directly determines the price of a contract in terms of a dollar.
However, the partial differential equation (3.61) has two limitations.
First, it applies only when the interest rate r is deterministic. Sec-
ond, its form is more complicated than the Black–Scholes partial
differential equation (3.52) obtained for two no-arbitrage assets S
and B T . The pricing of European options is still relatively straightforward,
so the advantage of using no-arbitrage assets for pricing is small. Therefore
using no-arbitrage assets in pricing is more important for complex financial
products, such as exotic options.
116 Stochastic Finance: A Numeraire Approach

Example 3.6
We have seen that the price of the European call option with a payoff (ST −
K · BTT )+ is given by
 

uT (t, x) = x · N √1
σ T −t
· log+ 21 σ 2 (T − t)
x
K
 
1 2 
− K · N σ√1T −t · log K x
− 2 σ (T − t) . (3.64)

Using the relationship v $ (t, x) = e−r(T −t) · uT (t, er(T −t) x), we can express the
dollar price of the option as
 

v $ (t, x) = x · N √1 x
· log K
σ T −t
+ (r + 21 σ 2 )(T − t)
 

− Ke−r(T −t) · N σ√1T −t · log Kx
+ (r − 12 σ 2 )(T − t) . (3.65)

This is the best-known form of the Black–Scholes formula. One can verify that
v $ (t, x) from (3.65) satisfies the Black–Scholes partial differential equation
(3.61).

Similarly, we can define the price function v S in terms of $ and S as a

reference asset by
Vt = v S (t, $S (t)) · St . (3.66)
The relationship between V S and the price function uS defined as

Vt = uS (t, BST (t)) · St (3.67)

is given by

v S (t, x) = uS (t, e−r(T −t) x), uS (t, x) = v S (t, er(T −t) x). (3.68)

Using the relationship between the price functions v S and uS , we can obtain
a partial differential equation for v S . Since uS satisfies the partial differential
equation
uSt (t, x) + 12 σ 2 x2 uSxx (t, x) = 0,
the function v S satisfies the partial differential equation

vtS (t, x) − rxvxS (t, x) + 12 σ 2 x2 vxx

S
(t, x) = 0. (3.69)

The terminal condition is

v S (T, x) = uS (T, x) = f S (x), (3.70)

and the boundary condition is

v S (t, 0) = uS (t, 0) = f S (0). (3.71)

 Diffusion Models 117

3.6 Hedging
Let us determine the hedging portfolio for a general European option con-
tract V .

THEOREM 3.4
The hedging portfolio Pt of the European option is given by
h i h i
Pt = uYx (t, XY (t)) · X + uY (t, XY (t)) − uYx (t, XY (t)) · XY (t) · Y,
(3.72)
or equivalently by
h i h i
Pt = uX (t, YX (t)) − uX X
x (t, YX (t)) · YX (t) · X + ux (t, YX (t)) · Y.

(3.73)

PROOF The hedging portfolio is in the form

Pt = ∆X (t) · X + ∆Y (t) · Y,
and has dynamics of the form
dPY (t) = ∆X (t, XY (t)) dXY (t).
We also have
dVY (t) = duY (t, XY (t)) = uYx (t, XY (t)) dXY (t).
In order to have
Pt = Vt ,
at all times, the hedge of this contract must satisfy

∂VY (t)
∆X (t, XY (t)) = uYx (t, XY (t)) = . (3.74)
∂XY (t)

The hedging position ∆X in the asset X is the sensitivity of the price of

the contract VY (t) to the changes of the underlying price XY (t). The hedge
position ∆Y in the asset Y follows from
∆Y (t) = PY (t) − ∆X (t) · XY (t) = uY (t, XY (t)) − uYx (t, XY (t)) · XY (t).
When X is chosen as a reference asset, the price dynamics of the hedging
portfolio P are given by
dPX (t) = ∆Y (t, YX (t)) dYX (t).
118 Stochastic Finance: A Numeraire Approach

We also have

dVX (t) = duX (t, YX (t)) = uX

x (t, YX (t)) dYX (t),

and thus in order to have

Pt = Vt ,
Y
the hedging position ∆ must satisfy

∂VX (t)
∆Y (t, YX (t)) = uX
x (t, YX (t)) = . (3.75)
∂YX (t)

The hedging position ∆X (t) in the asset X follows from

∆X (t) = PX (t) − ∆Y (t) · YX (t) = uX (t, YX (t)) − uX

x (t, YX (t)) · YX (t).

Recall that the prices in terms of the functions uY and uX are related by
the following symmetric relationship known as a perspective mapping:

uY (t, x) = uX (t, x1 ) · x, or uX (t, x) = uY (t, x1 ) · x.

We can connect the pricing partial differential equations for uY and uX

through the above relationship. The function uY solves Equation (3.52):

uYt (t, x) + 21 σ 2 x2 uYxx (t, x) = 0.

We can rewrite this partial differential equation in terms of uX using the

following identities:

uYt (t, x) = uX 1
t (t, x ) · x,
uYx (t, x) = uX (t, x1 ) − 1 X 1
x · ux (t, x ),
uYxx (t, x) = − x12 · uX 1 1 X 1
x (t, x ) + x2 · ux (t, x ) + 1
x3 · uX 1
xx (t, x )
1 X 1
= x3 · uxx (t, x ).

Substituting for uYt (t, x) and uYxx (t, x) in (3.52) we get

uYt (t, x) + 12 σ 2 x2 uYxx (t, x) = uX 1 1 2 2 1 X 1

t (t, x ) · x + 2 σ x x3 · uxx (t, x ) = 0,

which leads to
uX 1 1 2 1 X 1
t (t, x ) + 2 σ x2 · uxx (t, x ) = 0.

After making the substitution x1 → x, we can rewrite the above partial differ-
ential equation as
uX 1 2 2 X
t (t, x) + 2 σ x uxx (t, x) = 0,
 Diffusion Models 119

which is Equation (3.55). This is an independent derivation of this partial

differential equation using the relationship between uY and uX . Note that
the partial differential equation for uY and uX takes the same form, so it is
completely symmetric with respect to the choice of the reference asset. This
is not the case for more complex products, such as for Asian options.

We have previously seen that the hedging portfolio is given by

   
Pt = ∆X (t) · X + ∆Y (t) · Y = uYx (t, XY (t)) · X + uXx (t, YX (t)) · Y,

when using both price functions uY and uX , or in other words,

   
∂VY (t) ∂VX (t)
Pt = ·X+ · Y. (3.76)
∂XY (t) ∂YX (t)

Using the relationship between uY and uX :

uYx (t, x) = uX (t, x1 ) − 1

x · uX 1
x (t, x ),

or
uX Y 1
x (t, x) = u (t, x ) −
1
x · uYx (t, x1 ),
we can also write
h i h i
Pt = uYx (t, XY (t)) · X + uY (t, XY (t)) − uYx (t, XY (t)) · XY (t) · Y,

or equivalently
h i h i
Pt = uX (t, YX (t)) − uX
x (t, YX (t)) · YX (t) · X + u X
x (t, YX (t)) · Y.

This confirms Theorem 3.4.

Example 3.7 Hedging of the forward contract

The forward contract pays off XT − K · YT , which corresponds to the payoff

functions f Y (x) = x − K, and f X (x) = 1 − K · x. The price of the forward
contract is trivially given by uY (t, x) = x − K, and uX (t, x) = 1 − K · x.
Therefore the hedging portfolio is given by
   
Pt = uYx (t, x) · Xt + uXx (t, x) · Yt = Xt − K · Yt .

The hedge is static; one buys one unit of the asset X and sells K units of
Y . The forward contract can be thought of as a combination of two contracts
to deliver: one that delivers a unit of an asset X and one that delivers −K
units (or in other words shorts K units) of an asset Y . A contract to deliver
an asset X at time T is trivial: it is the asset X itself. One simply buys the
120 Stochastic Finance: A Numeraire Approach

asset and holds it until expiration. A similar argument applies to the asset Y .
Note that the hedge of the forward contract is model independent; it does not
depend on the evolution of the price XY (t).

Example 3.8 Hedging of the European call option

We have seen that the hedging position of a general European option in the
asset X is given by
∆X Y
t (t, XY (t)) = ux (t, XY (t)) .

This further simplifies when the payoff function is given by f Y (x) = (x−K)+ .
We have that
+
uYx (t, x) = d Y
(XY (T ) − K)
dx Et
 + 
d Y
= dx E x· X Y (T )
XY (t) − K |X Y (t) = x
 + 
d X
= dx E x − K · YYXX(T
(t)
)
|X Y (t) = x

= PX
t (XY (T ) ≥ K).

Thus we have

√ 
∆X (t) = PX
t (XY (T ) ≥ K) = N
√1
σ T −t
· log 1
K · XY (t) + 21 σ T − t .
(3.77)
Similarly we get

∆Y (t) = −K · PYt (XY (T ) ≥ K)


√ 
= −K · N σ√1T −t · log 1
K · XY (t) − 21 σ T − t . (3.78)

Hedging of an option that has a dollar as an underlying asset has to be

done in a stock S and in the bond B T (or equivalently in the money market
M ). Thus the hedging portfolio Pt is in the form
Pt = ∆S (t) · S + ∆T (t) · B T .
We have already seen that


∆S (t) = uTx (t, SB T (t)) , and ∆T (t) = uSx t, BST (t) .

We can also express the hedging positions in terms of the price functions v $
and v S . Since uT (t, x) = er(T −t) · v $ (t, e−r(T −t) x), we have
uTx (t, x) = vx$ (t, e−r(T −t) x),
 Diffusion Models 121

and thus

∆S (t) = uTx (t, SB T (t)) = uTx (t, S$ (t) · $B T (t))

 
= uTx t, er(T −t) · S$ (t) = vx$ (t, S$ (t)) .

The hedging position in the bond B T can be obtained from the dollar price
of the hedging portfolio
P$ (t) = ∆S (t) · S$ (t) + ∆T (t) · B$T (t),
or in other words,
v $ (t, S$ (t)) = vx$ (t, S$ (t)) · S$ (t) + ∆T (t) · e−r(T −t) .
Thus we have
h i
∆T (t) = er(T −t) · v $ (t, S$ (t)) − vx$ (t, S$ (t)) · S$ (t) .

Similarly, we can express the hedging portfolio in terms of the price function
v S . Since uS (t, x) = v S (t, er(T −t) x), we have
uSx (t, x) = vxS (t, er(T −t) x) · er(T −t) ,
and thus



∆T (t) = uSx t, BST (t) = uSx t, $S (t) · B$T (t)
 
= uSx t, e−r(T −t) · $S (t) = er(T −t) · vxS (t, $S (t)) .

The hedging position ∆S (t) can be obtained from

∆S (t) = PS (t) − ∆T (t) · BST (t)
= v S (t, $S (t)) − er(T −t) · vxS (t, $S (t)) · $S (t) · B$T (t)
= v S (t, $S (t)) − vxS (t, $S (t)) · $S (t).

COROLLARY 3.1
The hedging portfolio is given by
h i
Pt = vx$ (t, S$ (t)) · S
h h ii
+ er(T −t) · v $ (t, S$ (t)) − vx$ (t, S$ (t)) · S$ (t) · B T , (3.79)
or
h i
Pt = v S (t, $S (t)) − vxS (t, $S (t)) · $S (t) · S
h i
+ er(T −t) · vxS (t, $S (t)) · B T . (3.80)
122 Stochastic Finance: A Numeraire Approach

Assuming that M$ (t) = 1, or equivalently stated, Mt = er(T −t) · BtT , we can

also express the hedging portfolio in term of the stock S and the money market
M as
h i h i
Pt = vx$ (t, S$ (t)) · S + v $ (t, S$ (t)) − vx$ (t, S$ (t)) · S$ (t) · M, (3.81)

or
h i h i
Pt = v S (t, $S (t)) − vxS (t, $S (t)) · $S (t) · S + vxS (t, $S (t)) · M. (3.82)

3.7 Properties of European Call and Put Options

An option is in the money at time t if f Y (XY (t)) > 0. If the option were
to expire immediately at time t, its holder would collect a positive payoff. An
option is deep in the money if it is in the money and f Y (XY (T )) > 0 with
high probability, meaning that the option is likely to expire with a positive
payoff. An option is out of the money at time t if f Y (XY (t)) = 0. An option
is deep out of the money if it is out of the money, and f Y (XY (T )) = 0
with high probability, meaning that the option is likely to expire worthless.
An option is at the money if f Y (XY (t) + ǫ) > 0 and f Y (XY (t) − ǫ) = 0 for
ǫ > 0. An at the money option is a boundary case between in the money and
out of the money option.

Given the hedge representation for a European call option

∆X (t) = PX
t (XY (T ) ≥ K),

and
∆Y (t) = −K · PYt (XY (T ) ≥ K),
we can see that

0 ≤ ∆X (t) ≤ 1, and − K ≤ ∆Y (t) ≤ 0.

Moreover, if the option is deep out of the money, the option is almost worth-
less, and the corresponding hedge is ∆X (t) ≈ 0, and ∆Y (t) ≈ 0. On the other
hand, if the option is deep in the money, ∆X (t) ≈ 1, ∆Y (t) ≈ −K, and the
European option contract is close to a forward Xt − K · Yt .

Another interesting observation is to see what happens when the maturity of

the option approaches infinity, or equivalently, when the volatility approaches
infinity. Recall that the price of a European call option is given by

V EC (X, K · Y, T ) = N (d+ ) · X − K · N (d− ) · Y,

 Diffusion Models 123

FIGURE 3.1: The price VY (t) of a European option contract with a payoff
(XT − K · YT )+ with parameters K = 21 , σ = 0.2, as a function of price XY (t),
and time to maturity T . We have considered unrealistically large maturities
in order to show the limiting behavior of the option price.

where
1 1

1 √
d± = √
σ T
· log ·
K X Y (0) ± 2σ T,
√
and so the price is a function of a factor σ T . For instance, doubling the
volatility has the same effect on the option price as quadrupling time. When
T → ∞, we simply have
lim VYEC = XY
T →∞

since d+ → ∞, and d− → −∞. Therefore for large T , VY (0) ≈ XY (0), and

the hedge is to hold a unit of an asset X and have no position in the asset
Y . Figure 3.1 shows the price VY of a European call option with a payoff
(XT − 21 YT )+ as a function of the price XY (t) of the underlying asset X, and
time to maturity T − t. Note that when t = T , the price of the contract is
simply the payoff (x− 12 )+ . On the other hand, for large maturities the price of
the contract is approximately XY , so the price of VY becomes approximately
linear in XY .

Figures 3.2 and 3.3 show the corresponding hedging positions in the under-
lying assets X and Y as a function of the price XY (t) and time to maturity
T − t. Note that the hedging position in the asset X is between 0 and 1, and
the hedging position in the asset Y is between − 21 and 0. For short maturities,
the hedging position in the asset X should be close to 1 when the option is in
124 Stochastic Finance: A Numeraire Approach

FIGURE 3.2: The hedging position in the asset X for the European option
contract (XT − K · YT )+ with parameters K = 12 , σ = 0.2, as a function of
the price XY (t) and time to maturity T − t. Note that the hedging position
in the asset X is between 0 and 1.

the money, but it should be close to 0 when the option is out of the money.
There is a jump in the hedging position at the strike price at the time of
maturity. For large maturities, the hedging position in the asset X should be
close to 1.

Similarly, for short maturities, the hedging position in the asset Y should
be close to − 21 when the option is in the money, and it should be close to 0
when the option is out of the money. For long maturities, the hedging position
in the asset Y should be close to 0.

Figure 3.4 shows a sample path of XY in a geometric Brownian motion

model, and the corresponding price of the European option VY . Figure 3.5
shows the corresponding hedging position in the underlying assets X and
Y . Note that the hedging positions start to change dramatically when the
time is close to maturity. The reason is that the price of the underlying asset
happens to be near the strike price when the option is close to maturity, and
the corresponding hedging position in the asset X takes the values close to 0
or 1 depending whether the option is out of the money or in the money. We
observe a similar behavior for the hedging position in the asset Y .
 Diffusion Models 125

FIGURE 3.3: The hedging position in the asset Y for the European option
contract (XT − K · YT )+ with parameters K = 21 , σ = 0.2, as a function of
the price XY (t) and time to maturity T − t. Note that the hedging position
in the asset Y is between − 21 and 0.

0.7

0.6

0.5

0.4
Price

0.3

0.2

0.1

0
0 1/4 1/2 3/4 1
Time

FIGURE 3.4: The price XY (t) of an asset X in terms of the reference

asset Y (top), and the price VY (t) of a European option contract with a
payoff (XT − K · YT )+ with parameters XY (0) = 21 , K = 12 , σ = 0.2, T = 1
(bottom).
126 Stochastic Finance: A Numeraire Approach

0.5
Hedge

−0.5
0 1/4 1/2 3/4 1
Time

FIGURE 3.5: The hedging position in the asset X (top) and Y (bottom)
for the European option contract (XT − K ·YT )+ with parameters XY (0) = 21 ,
K = 21 , σ = 0.2, T = 1. Note that the hedging position in the asset X is
between 0 and 1, and the hedging position in the asset Y is between − 12 and
0.

REMARK 3.10 Greeks

Greeks measure sensitivities of the prices of the portfolio (or in particular a
single financial contract) to the changes of the parameters of the model. They
describe how the price of the portfolio would change if the parameters change.
Note that the price of the portfolio is given relative to the reference asset, so
one can define portfolio sensitivities for any price function. The traditional
definition of greeks applies to the price function v $ , but it would make even
better sense to apply it to the price function uY , or uX . The assets Y and
X have no time value (in contrast to a dollar $), and thus the corresponding
greeks would not be influenced by the time decay of the reference asset.

Delta is the sensitivity of the price uY with respect to the price of the un-
derlying XY :
∂uY (t, XY (t))
∆(t) = uYx (t, XY (t)) = . (3.83)
∂XY (t)
Gamma is the sensitivity of ∆ with respect to the price of the underlying
XY , which is the same as the second derivative of uY with respect to XY :

∂ 2 uY (t, XY (t))
Γ(t) = uYxx (t, XY (t)) = . (3.84)
∂XY2 (t)
 Diffusion Models 127

Theta is the sensitivity of the price uY with respect to time t:

∂uY (t, XY (t))
Θ(t) = uYt (t, XY (t)) = . (3.85)
∂t
Vega is the sensitivity of the price uY with respect to the volatility σ:
∂uY (t, XY (t))
ν(t) = uYσ (t, XY (t)) = . (3.86)
∂σ
Rho is the sensitivity of the price uY with respect to the interest rate r:
∂uY (t, XY (t))
ρ(t) = uYr (t, XY (t)) = . (3.87)
∂r

Example 3.9
Consider an option with a payoff (XT − K · YT )+ . Its price is given by the
Black–Scholes formula
   √ 
uY (t, XY (t)) = xN σ√1T −t · log XYK(t) + 12 σ T − t
   √ 
− KN σ√1T −t · log XYK(t) − 12 σ T − t .

The corresponding greeks are given by

   √ 
∆(t) = N σ√1T −t · log XYK(t) + 12 σ T − t ,
1    √ 
Γ(t) = √ · φ σ√1T −t · log XYK(t) + 12 σ T − t ,
XY (t)σ T − t
σXY (t)    √ 
θ(t) = − 12 · √ · φ σ√1T −t · log XYK(t) + 21 σ T − t ,
T −t
√    √ 
ν(t) = XY (t) T − t · φ σ√1T −t · log XYK(t) + 12 σ T − t ,
ρ(t) = 0.
The sensitivity ρ turns out to be zero since the price evolution XY is not
influenced by the changes of the interest rate (assets X and Y have no time
value). The changes of the interest rate would influence contracts that depend
on the assets with time value, such as a dollar $.

3.8 Stochastic Volatility Models

When we have a contingent claim V whose payoff depends on the assets X
and Y , its price VY (t) can depend on the entire price evolution XY (s) up to
128 Stochastic Finance: A Numeraire Approach

time t. It can also depend on several additional external processes ξ i (s), such
as a random process that represents stochastic volatility. In this case we can
write
VY (t) = uY (t, {XY (s)}ts=0 , {ξ i (s)}ts=0 ).
While this expression would explain the price process VY (t) in full, it would
be prohibitively complicated to model the price of VY (t) using infinitely many
possible values from {XY (s)}ts=0 and {ξ i (s)}ts=0 . Thus it is desirable to ex-
press such dependence using only a small number of factors that would explain
the price evolution VY (t) sufficiently well.

A common approach to price modeling is to use the Markov property:

VY (t) = uY (t, {XY (s)}ts=0 , {ξ i (s)}ts=0 ) = uY (t, XY (t), ξ i (t)),

which says that the only relevant information about the future evolution of
the process VY (t) is given by the present values of the underlying processes
XY (t) and ξ i (t).

The simplest models that we considered in the previous text assume no

external processes ξ i (t), and the price of the contract V can be written as

V (t) = uY (t, XY (t)) · Y = uX (t, YX (t)) · X.

More general models of the asset prices consider a stochastic evolution of

volatility. The price of a contract V depends on the price of the underlying
asset XY (t), and on a process ξ(t) that represents the volatility

V (t) = uY (t, XY (t), ξ(t)) · Y = uX (t, YX (t), ξ(t)) · X.

This model has two sources of uncertainty, and it is not possible in general to
hedge such contracts perfectly with only two assets X and Y . A general rule
for a complete market is to have n+1 assets for n sources of noise, which is not
the case here. Thus stochastic volatility models are not complete in general
and a perfect replication of an arbitrary contingent claim may no longer be
possible. As mentioned earlier, the volatility is the same for both XY (t) and
YX (t).

Let us assume that the price process follows

dXY (t) = g(t, ξ(t))XY (t)dW Y (t), (3.88)

where ξ(t) is a stochastic process in the form

dξ(t) = α(t, ξ(t))dt + β(t, ξ(t))dW ξ (t). (3.89)

We assume that the two Brownian motions W Y and W ξ are correlated:

dW Y (t) · dW ξ (t) = ρdt.

 Diffusion Models 129

Note that the price process XY (t) is a PY martingale. The process ξ(t) is a
parameter of the model, and as such it can have an arbitrary evolution. In
particular, it does not need to be a martingale.

Example 3.10
A popular stochastic volatility model is the Heston model, which is given
by the following choice of the functions g, α, and β:
p p
g(t, ξ) = ξ, α(t, ξ) = a − b · ξ, β(t, ξ) = σ ξ.

In this case we can write

p
dXY (t) = ξ(t) · XY (t)dW Y (t),

and p
dξ(t) = (a − b · ξ(t))dt + σ ξ(t)dW ξ (t).

Let V be a contingent claim whose price VY (t) depends only on XY (t) and
on ξ(t). We can write

VY (t) = uY (t, XY (t), ξ(t)).

Since VY (t) is a PY martingale, we can obtain a partial differential equation

for the price function uY . We have

duY (t, XY (t), ξ(t)) = uYt dt + uYx dXY (t) + uYξ dξ(t)
+ 1 uYxx d2 XY (t) + uYxξ dXY (t)dξ(t) + 21 uYξξ d2 ξ(t)
h 2
= uYt + α(x, ξ)uYξ + 21 g 2 XY (t)2 uYxx
i
+ρβgXY (t)uYxξ + 21 β 2 uYξξ dt
+gXY (t)uYx + βuYξ dW ξ (t).

Since the dt term must be zero, we get a partial differential equation for uY :

uYt (t, x, ξ) + α(t, ξ)uYξ (t, x, ξ) + 12 g(t, ξ)2 x2 uYxx (t, x, ξ)

+ ρβ(t, ξ)g(t, ξ)xuYxξ (t, x, ξ) + 12 β(t, ξ)2 uYξξ (t, x, ξ) = 0. (3.90)

Similarly, we can study the evolution of the inverse price that takes the
same form
dYX (t) = g(t, ξ(t)) · YX (t)dW X (t),
where
dW X (t) = −dW Y (t) + g(t, ξ(t))dt.
130 Stochastic Finance: A Numeraire Approach

This follows from Ito’s formula

dYX (t) = dXY (t)−1 = −YX (t)2 dXY (t) + 1

2 · 2YX (t)3 d2 XY (t)
= −g(t, ξ(t)) · YX (t)dW Y (t) + g(t, ξ(t))2 · YX (t)dt
= g(t, ξ(t)) · YX (t)dW X (t).

The correlation between W X (t) and W ξ (t) is given by

dW X (t) · dW ξ (t) = (−dW Y (t) + g(t, ξ(t))dt) · dW ξ (t) = −ρdt.

The only difference is that the correlation coefficient takes an opposite sign.
Thus if we have
VX (t) = uX (t, YX (t), ξ(t)),
the partial differential equation for uX differs only in the sign that corresponds
to the correlation coefficient. Therefore uX satisfies

uX X 1 2 2 X
t (t, x, ξ) + α(x, ξ)uξ (t, x, ξ) + 2 g(t, ξ) x uxx (t, x, ξ)

− ρβ(t, ξ)g(t, ξ)xuX 1 2 X

xξ (t, x, ξ) + 2 β(t, ξ) uξξ (t, x, ξ) = 0. (3.91)

3.9 Foreign Exchange Market

This section studies contracts traded on foreign exchange markets. Let an
asset X be the domestic currency $, and an asset Y be the foreign currency
e. Let
e$ (t)
denote the amount of a domestic currency that is needed to acquire a unit of
a foreign currency at time t. The quantity e$ (t) is known as an exchange
rate, but in fact this is just a special case of the price XY (t), where X = e,
and Y = $. Thus we can apply the results we have obtained in the previous
sections for the case of the foreign exchange market.

The foreign exchange market is an excellent example to illustrate the rela-

tive concept of prices since both the domestic and the foreign currencies are
legitimate choices for the reference asset. Whether a currency is domestic or
foreign depends on which country one lives in. For some people $ is the do-
mestic currency and e is the foreign currency, but for other people e is the
domestic currency and $ is the foreign currency. Thus it makes perfect sense
to study the inverse exchange rate
1
= $e (t).
e$ (t)
 Diffusion Models 131

Note that e$ (t) and $e (t) are prices, not assets that could be bought or sold.
Moreover, the currencies themselves are arbitrage assets, and thus one needs
to immediately acquire a suitable no-arbitrage asset for it in order not to lose
value, such as a bond that is denominated in the corresponding currency.

3.9.1 Forwards
Let us consider first a forward contract on the foreign exchange with a
payoff
eT − K · $T = (e$ (T ) − K) · $T = (1 − $e (T )) · eT .
at time T . Let us write this contract in terms of no-arbitrage assets. There is
a corresponding foreign bond that delivers one e at time T . We will denote
this no-arbitrage asset by B e,T . Similarly, there is a domestic bond B T that
delivers one $ at time T . Therefore the forward contract is equivalent to the
contract with a payoff
BTe,T − K · BTT .
Let us denote the forward contract by F (B e,T , K · B T , T ), and let us com-
pute its price. The contract depends on two no-arbitrage assets, namely on
B e,T , and B T . A possible numeraire for pricing is B T , a domestic bond ma-
turing at time T . We have

 e,T
FB T (t) = ETt B e,T − K · B T B T (T ) = BB T (t) − K. (3.92)

The last identity follows from the fact that the price of the bond B e,T in
terms of the bond B T is a martingale under the T-forward measure that
corresponds to B T as a reference asset. Thus we conclude that the forward
contract is equal to

Ft (B e,T , K · B T , T ) = Bte,T − K · BtT . (3.93)

Note that this is a model-independent formula (we have not assumed any
particular dynamics). A forward exchange rate is a choice of K̄ that makes
the forward contract equal to zero:

F0 (B e,T , K̄ · B T , T ) = 0. (3.94)

Solving this equation, we get

e,T
K̄ = BB T (0). (3.95)

If we assume constant interest rates for both the domestic and the foreign
zero coupon bond, we can express the above relationship in terms of the ex-
change rate e$ (0). The domestic bond price in terms of the domestic currency
is
BtT = e−r(T −t) · $t ,
132 Stochastic Finance: A Numeraire Approach

and the foreign bond price in terms of the foreign currency is

F
Bte,T = e−r (T −t)
· et .

Thus we can write

e,T e,T
BB T (t) = Be (t) · e$ (t) · $B T (t)
F F
= e−r (T −t)
· e$ (t) · er(T −t) = e(r−r )(T −t)
· e$ (t).

In other words,
e,T F
(r−r )T
K̄ = BB T (0) = e · e$ (0). (3.96)

We can consider a similar contract on the inverse exchange rate $e (T )

$T − K f · eT = $e (T ) − K f · eT = (1 − K f · e$ (T )) · $T .

The corresponding no-arbitrage assets are B T and B e,T . We can rewrite the
payoff of the contract as
BTT − K f · BTe,T .

If we denote this contract by F (B T , K f ·B e,T , T ), and choose the correspond-

ing foreign bond B e,T as a numeraire, we get


FB e,T (t) = Ee,T
t B T − K f · B e,T B e,T
T
(T ) = BB f
e,T (t) − K . (3.97)

The last equation follows from the fact that the price of B T in terms of B e,T
is a martingale under the measure that corresponds to B e,T as a numeraire.
The forward contract is equal to

Ft (B T , K f · B e,T , T ) = BtT − K f · Bte,T . (3.98)

The corresponding forward exchange rate from the point of view of the foreign
currency is a choice of K̄ f that makes the value of the forward contract zero:

F0 (B T , K̄ f · B e,T , T ) = 0. (3.99)

Solving for K̄ f , we get

F 1
K̄ f = BB
T
e,T (0) = e
(r −r)T
· $e (0) = . (3.100)
K̄

Note that the forward exchange rates as seen from the domestic currency and
from the foreign currency point of view are linked through K̄ = K̄1f .
 Diffusion Models 133

3.9.2 Options
European-type contracts on foreign exchange are special cases of general
European contracts where the roles of the no-arbitrage assets X and Y are
played by no-arbitrage assets B e,T and B T . For instance, a call option with
payoff
(eT − K · $T )+
can be rewritten in terms of the no-arbitrage assets as

(BTe,T − K · BTT )+ .

We have a special case of the Black–Scholes formula that is also known as

Garman–Kohlhagen formula.

REMARK 3.11 Garman–Kohlhagen formula

The value VtEC (B e,T , KB T , T ) of a European option contract with a payoff

+
B e,T − K · BTT is given by

V EC (B e,T , KB T , T ) = Pe,T
t
e,T
(BB e,T
T (T ) ≥ K) · Bt

e,T
−K · PTt (BB T
T (T ) ≥ K) · Bt (3.101)
f
= Pe,T
t (e$ (T ) ≥ K) · e−r (T −t)
· eT
−K · PTt (e$ (T ) ≥ K) · e−r(T −t) · $T .

Moreover, the corresponding deltas are given in the geometric Brownian mo-
tion model by

∆e,T (t) = Pe,T

t
e,T
(BB e,T
T (T ) ≥ K) = Pt (e$ (T ) ≥ K), (3.102)

and
e,T
∆T (t) = −K · PTt (BB T
T (T ) ≥ K) = −K · Pt (e $ (T ) ≥ K). (3.103)

References and Further Reading

The first introduction of Brownian motion to finance is by Bachelier (1900).
He used it as a model for stock prices, although Brownian motion can take
negative values. The general theory of stochastic calculus was developed by
Ito (1944). Merton (1969) was the first person who used it in finance. Samuel-
son (1965, 1973) argued that geometric Brownian motion is a good model
134 Stochastic Finance: A Numeraire Approach

for stock prices. The first derivation of the Black–Scholes formula appears in
Black and Scholes (1973). A similar result appears in the independent work of
Merton (1973). The Black–Scholes formula is quite robust to model misspec-
ifications as shown in El Karoui et al. (1998). Garman and Kohlhagen (1983)
found the analogous formula to the Black–Scholes for currency options.

Girsanov’s theorem is due to Girsanov (1960), although the result for the
constant σ appeared already in Cameron and Martin (1944). The principle of
exchangeability of the reference assets appears implicitly already in Carr and
Bowie (1994), and more recently the symmetries between the pricing martin-
gale measures are explored in detail in Carr and Lee (2009). Hoogland and
Neumann (2001a,b) explored the symmetries in pricing with respect to dif-
ferent reference assets using the partial differential equation approach. They
noticed the advantages of using no-arbitrage assets for the numeraire. Wystup
(2008) applied symmetry analysis in the foreign exchange market and showed
various relationships of greeks for specific options. Preservation of convex-
ity by the perspective mapping is shown for instance in Hiriart-Urruty and
Lemarechal (1993).

Books that explain the numerical implementation of partial differential

equations in detail are, for instance, Tavella and Randall (2000) or Duffy
(2006). Monte Carlo methods for pricing financial derivatives are employed
for instance in the papers of Boyle (1977), Boyle et al. (1997), Broadie and
Glasserman (1996), or in the monographs Glasserman (2003), Jaeckel (2002),
or Korn et al. (2010). Books on pricing derivative contracts under stochas-
tic volatility include Lewis (2000), Gatheral (2006), Fouque et al. (2000), or
Rebonato (2004). The Heston stochastic volatility model was introduced in
Heston (1993).

Exercises
3.1 Prove that the perspective mapping preserves convexity: If f Y (x) is a
convex function, then the function f X (x) = f Y ( x1 ) · x, x > 0 is also convex.

(a) Show the result for the twice differentiable function f Y .

Hint: Twice differentiable convex functions satisfy [f Y (x)]′′ ≥ 0. Show
that this implies [f X (x)]′′ ≥ 0.

(b) Show the result for a general convex function f Y (x).

3.2 Show that

XY (t) · φ(d+ ) − K · φ(d− ) = 0,
 Diffusion Models 135
2
x
where φ(x) = √1
2π
· e− 2 , and
  √
XY (t)
d± = √1 · log ± 21 σ T − t.
σ T −t K

3.3 Show that the price

√
uY (t, x) = N ( σ√1T −t · log x
K − 21 σ T − t)

of the Arrow–Debreu security V that pays off VT = I(XY (T ) ≥ K) · YT in the

geometric Brownian motion model satisfies the partial differential equation

uYt (t, x) + 12 σ 2 x2 uYxx (t, x) = 0,

with the terminal condition uY (T, x) = I(x ≥ K), and the boundary condition
uY (t, 0) = 0.

3.4 (a) Find the hedging portfolio for an Arrow–Debreu security V that
pays off VT = I(XY (T ) ≥ K) · YT in a geometric Brownian motion
model.
(b) Find the hedging portfolio for an Arrow–Debreu security V that pays off
UT = I(XY (T ) ≥ K) · XT in the same model.
(c) Combining the results from (a) and (b), find the hedging portfolio for a
contract W that pays off WT = UT − K · VT . Note that W is a European
call option.

3.5 Assume that XY (0) = 1, and consider a contract that pays off the more
valuable asset at time T ; i.e., the payoff is in the form

max(XT , YT ) = XI (XY (T ) ≥ 1) + Y I (XY (T ) < 1) .

The price XY (t) follows Geometric Brownian Motion:

dXY (t) = σXY (t)dW Y (t).

(a) Compute the price of this contract.

(b) Determine the hedge of the contract: ∆X (t) and ∆Y (t).

3.6 Let the price process XY (t) follow the geometric Brownian motion

dXY (t) = σXY (t)dW Y (t),

and let K be a general positive constant.

(a) What is the price of a contract that pays a unit of Y when XY (T ) ≥ K
when T → ∞? How would you hedge such a contract?
136 Stochastic Finance: A Numeraire Approach

(b) What is the price of a contract that pays a unit of X when XY (T ) ≥ K

when T → ∞? How would you hedge such a contract?
(c) Determine the price and the hedge of a European call option with the
payoff (XT − K · YT )+ when T → ∞.
Chapter 4
Interest Rate Contracts

This chapter covers financial contracts that depend on interest rates. The
market that trades interest rate contracts is known as a fixed income market.
The main message of this chapter is that the interest rate traded on the ex-
changes or on the over-the-counter markets is just a special case of the concept
of price, and thus the pricing techniques developed in the previous chapters
immediately apply to the interest rate market as well. The interest rate is dis-
tinct in the sense that the corresponding reference asset is not a currency, but
rather a bond, or a portfolio of bonds. Thus the numeraire approach to use a
bond as a natural reference asset is especially helpful in the determination of
the price of the interest rate contracts.

The most traded interest rate contracts depend on the London Interbank
Offered Rate, or LIBOR for short. Spot LIBOR L(T, T ) represents a simple
annualized interest rate that corresponds to borrowing money for a period
between times T and T + δ, where δ is typically 3 months. The market also
trades contracts that depend on the value of LIBOR in the future. The sim-
plest contract, the backset LIBOR, agrees to deliver L(T, T ) units of a dollar
at time T + δ. The price of this contract is a forward LIBOR L(t, T ), a simple
interest rate that corresponds to borrowing money between times T and T + δ
as seen at an earlier time t. The forward LIBOR is in fact a price, where the
reference asset is a bond with maturity T + δ.

Exchanges trade on the LIBORs as using futures contracts rather than for-
ward contracts. The connection between futures and forwards was discussed in
detail in Section 1.10, and we will not revisit this analysis in this chapter. The
futures market on the LIBOR has one specific convention as the settlement
price is quoted as 100 · (1 − L(T, T )) rather than simply L(T, T ). For exam-
ple, the interest rate 1.98% corresponds to the quote of 98.02 on the futures
market for LIBOR. An option on the LIBOR rate is known as either a caplet
(call option on LIBOR) or a floorlet (put option on LIBOR). Since LIBOR is
a price, the price of the caplet and the floorlet is given by the Black–Scholes
formula, but it is known as the Black caplet formula in the fixed income mar-
ket.

Another widely traded contract is a swap. A holder of a swap agrees to

receive floating interest rate payments for an exchange of fixed interest rate

137
138 Stochastic Finance: A Numeraire Approach

payments. A swap rate is the level of the fixed interest rate payments that
makes the value of the swap equal to zero. Similar to the case of the for-
ward LIBOR, forward swap rates are particular instances of a price, where
the reference asset is a portfolio of bonds. The right to enter a swap contract
is known as a swaption, and the price of the swaption is also given by the
Black–Scholes formula.

Models of the interest rate consider instantaneous rates corresponding to

a situation when the borrowed money is returned in an infinitesimal instant.
This is the case of a forward rate f (t, T ), an instantaneous rate for borrow-
ing money at time T as seen at time t. The forward rates must follow a
no-arbitrage evolution given by the Heath–Jarrow–Morton model (HJM for
short). It turns out that the spot rate models are just special cases of the
HJM model.

4.1 Forward LIBOR

The forward LIBOR is defined as a simple interest rate that corresponds
to borrowing money over the time interval between T and T + δ as seen at
time t ≤ T . We denote the forward LIBOR by L(t, T ). Suppose that $ 1 is
borrowed at time T , and assume that L(t, T ) is the simple interest rate for the
period between T and T + δ. Then the agent should return 1 + δL(t, T ) dollars
at time T + δ. Thus L(t, T ) can be defined by the following relationship:

(1 + δL(t, T )) · BtT +δ = BtT . (4.1)

The right hand side of the above relationship indicates that one dollar will be
delivered at time T . The left hand side indicates that (1 + δL(t, T )) dollars
will be returned at time T + δ. Therefore

δBtT +δ · L(t, T ) = BtT − BtT +δ , (4.2)

or in other words
 
L(t, T ) = B T − B T +δ δB T +δ (t). (4.3)

Notice that the forward LIBOR is in fact a price, where the asset X is a
portfolio [B T − B T +δ ] (long the B T bond, and short the B T +δ bond), and
the reference asset Y is δB T +δ . Note that the forward LIBOR is an example
of a price process where the reference asset is not the currency itself. The fact
that we can treat the forward LIBOR as a price means that we can apply the
general theory of pricing contingent claims.
 Interest Rate Contracts 139

It should be also noted that L(t, T ) as a price continues to evolve up to

time T . At time T , the B T bond expires, so the last meaningful observation
of the forward LIBOR is L(T, T ). Therefore after time T , the LIBOR L(T, T )
is known, and it is a constant. Typical contracts that depend on the LIBOR
L(T, T ) are settled at time T + δ, a delay of δ units of time after observing its
value at time T . The most traded contracts have δ equal to 3 months = 41 year.

Since the forward LIBOR is a price, it is a martingale under the forward

measure PT +δ associated with the reference asset δB T +δ , δ units of the bond
maturing at time T +δ. A popular model of the forward LIBOR is a geometric
Brownian motion


dL(t, T ) = d [B T − B T +δ ]δB T +δ (t)


= σ [B T − B T +δ ]δB T +δ (t) dW T +δ (t).
In this case, we have


L(T, T ) = L(t, T ) · exp σW T +δ (T − t) − 12 σ 2 (T − t) . (4.4)
For pricing European contracts on the LIBOR, it is also important to con-
1
sider dynamics of the inverse price L(t,T ) which has to be a martingale under
probability measure associated with [B − B T +δ ] as a reference asset. We will
T

call this measure PT,T +δ ; it corresponds to the measure PX that was used in
the previous sections. We have that
   T +δ 
1
d L(t,T ) = d δB (B T −B T +δ )
(t)
 T +δ 
= σ δB (B T −B T +δ )
(t)dW T,T +δ (t). (4.5)

This gives us


L(T, T ) = L(t, T ) · exp σW T,T +δ (T − t) + 12 σ 2 (T − t) . (4.6)

4.1.1 Backset LIBOR

Backset LIBOR is a contract that pays off L(T, T ) units of a dollar $ at
time T + δ
VT +δ = L(T, T ) · $T +δ .
The payoff can be also expressed in terms of a B T +δ bond as
VT +δ = L(T, T ) · BTT +δ
+δ
= L(T, T ) · $T +δ . (4.7)
When t ≤ T , the price of this contract is given by
 
T +δ
Vt = EtT +δ L(T, T ) · BB T +δ
T +δ (T + δ) · Bt = L(t, T ) · BtT +δ
 T   
= B − B T +δ δB T +δ (t) · BtT +δ = δ1 BtT − BtT +δ . (4.8)
140 Stochastic Finance: A Numeraire Approach

We have used the fact that L(t, T ) is a PT +δ martingale. The hedge of this
contract is obvious; one should long 1δ units of the bond B T , and short δ1
units of the bond B T +δ . When the LIBOR L(T, T ) is already known at times
T ≤ t ≤ T + δ, the price of the contract is given by
 
Vt = EtT +δ L(T, T ) · BB
T +δ T +δ
T +δ (T + δ) · Bt = L(T, T ) · BtT +δ . (4.9)

The hedge is to hold L(T, T ) units (now a known number) of the bond B T +δ .

4.1.2 Caplet
A caplet is a contract that pays off

VT +δ = (L(T, T ) − K)+ · $T +δ = 1δ (L(T, T ) − K)+ · δBTT +δ

+δ
(4.10)

at time T + δ. It is a European call option on LIBOR settled in dollars. The

price of caplet is given by the Black–Scholes formula
 
V EC (X, K · Y, T ) = 1δ · PX Y
t (XY (T ) ≥ K) · X − K · Pt (XY (T ) ≥ K) · Y ,
(4.11)
where X = [B T − B T +δ ], and Y = δB T +δ . In this case we can write
h
Vt = δ1 · PtT,T +δ (L(T, T ) ≥ K) · [B T − B T +δ ]
i
−K · PtT +δ (L(T, T ) ≥ K) · δB T +δ
 
= L(t, T ) · PtT,T +δ (L(T, T ) ≥ K) − K · PtT +δ (L(T, T ) ≥ K) · B T +δ

for t ≤ T. For the geometric Brownian motion model, we have


√ 
PtT,T +δ (L(T, T ) ≥ K) = N σ√1T −t · log K
1
· L(t, T ) + 12 σ T − t ,

and

√ 
PTt (L(T, T ) ≥ K) = N √1
σ T −t
· log 1
K · L(t, T ) − 21 σ T − t .

The Black–Scholes formula for the price of the caplet is also known as the
Black caplet formula.

The hedge of the European call option in a geometric Brownian motion

model is given by
• ∆X (t) = PX
t (XY (T ) ≥ K) units of an asset X, and

• ∆Y (t) = −KPYt (XY (T ) ≥ K) units of an asset Y .

For the case of the caplet, one should have
 Interest Rate Contracts 141

1 T,T +δ
• δ Pt (L(T, T ) ≥ K) units of [B T − B T +δ ], and

• −KPtT +δ (L(T, T ) ≥ K) units of B T +δ .

The payoff of the caplet is scaled by a factor of δ1 , so the hedging positions

are scaled correspondingly. This is the same as having

1 T,T +δ
• δ Pt (L(T, T ) ≥ K) units of the bond B T , and

• − δ1 PtT,T +δ (L(T, T ) ≥ K) − KPtT +δ (L(T, T ) ≥ K) units of the

bond B T +δ .

4.2 Swaps and Swaptions

A swap is a series of payments of the form

δ[L(Tk−1 , Tk−1 ) − K] · $Tk (4.12)

for k = 1, · · · , n, and Tk = T0 + δk. The payments are paid at times Tk , where

the time lag between the payments is given by Tk − Tk−1 = δ. In practice, δ is
typically 21 or 41 years. The holder of the contract receives L(Tk−1 , Tk−1 ) units
of a dollar $Tk , which represents the LIBOR rates observed at the previous
times Tk−1 . Since the LIBOR rates are random variables, the corresponding
payments are known as floating payments. On the other hand, the holder
of the contract agrees to pay K units of a dollar $Tk at time Tk , which cor-
responds to fixed payments. Note that the backset LIBOR is a special case
of a swap with one payment only, n = 1.

Let V be the swap with the payments (4.12). We can rewrite the contract
as taken at time t in a more compact form and in terms of no-arbitrage assets
that have no time value as

n
X
Vt = δ · [L(t, Tk−1 ) − K] · BtTk . (4.13)
k=1
142 Stochastic Finance: A Numeraire Approach

1
 T 
Furthermore, if we substitute for L(t, Tk−1 ) = δ B k−1 − B Tk B Tk (t), we get
n
X
Vt = δ · [L(t, Tk−1 ) − K] · BtTk
k=1
Xn
1   
=δ· δ B Tk−1 − B Tk B Tk
(t) − K · BtTk
k=1
n h
X  i
T
= Bt k−1 − BtTk − δ · K · BtTk
k=1
h i n
X
= BtT0 − BtTn − δ · K · BtTk .
k=1

Therefore the swap is equal to

h i n
X
Vt = BtT0 − BtTn − δ · K · BtTk . (4.14)
k=1

The forward swap rate y(t) is the value of K in (4.14) which makes the
value of the swap V equal to zero. Solving for Vt = 0, we get
h i n
X
BtT0 − BtTn − δ · y(t) · BtTk = 0.
k=1

Therefore  
y(t) = B T0 − B Tn Pn δ·B Tk
(t), (4.15)
k=1

so the forward swap rate is also a special instance of a price, where

P the as-
set X is equal to B T0 −B Tn , and the reference asset Y is equal to nk=1 δ·B Tk .
Pn
The price of the swap V in terms of the reference asset k=1 δ · B Tk can
be written as
 
VPnk=1 δ·B Tk (t) = B T0 − B Tn Pn δ·B Tk (t) − K = [y(t) − K] . (4.16)
k=1

A closely related contract is the right to enter the swap contract at time
T0 , known as a swaption. The payoff of the swaption is given by
!+
Xn h i n
X
+ Tk T0 Tn Tk
VT0 = (y(T0 )−K) · δ·BT0 = BT0 − BT0 − K · δ · BT0 . (4.17)
k=1 k=1

of an option with a payoff (XT0 − KYT0 )+ ,

Thus a swaption is a special case P
n
where X = B − B , and Y = k=1 δ · B Tk . The value of the swaption at
T0 Tn
 Interest Rate Contracts 143

time t ≤ T0 is given by the Black–Scholes formula (1.77) as

h i n
X
Vt = PTt 0 ,Tn (y(T0 ) ≥ K)· BtT0 − BtTn −K ·PΣ
t (y(T0 ) ≥ K)· δ·BtTk . (4.18)
k=1

T0 ,Tn
Here, the probability measure P corresponds to the reference asset
X = B T0 − B Tn ,Pand the probability measure PΣ corresponds to the ref-
n
erence asset Y = k=1 δ · B Tk .

According to the First Fundamental Theorem of Asset Pricing, the for-

ward swap rate y(t) is a PΣ martingale. A popular approach for modeling the
evolution of the forward swap rate is to assume geometric Brownian motion
dynamics
dy(t) = σy(t)dW Σ (t). (4.19)
The Black–Scholes formula then simplifies to
h i n
X
Vt = N (d+ ) · BtT0 − BtTn − K · N (d− ) · δ · BtTk ,
k=1

where   √
y(t)
d± = √1 · log ± σ T − t.
σ T −t K

4.3 Term Structure Models

Term structure models describe a no-arbitrage evolution of the prices
of bonds with different maturities. Recall that a bond is a contract to deliver
one dollar at time T . We denote this contract as B T . Given that T can be
chosen arbitrarily, we have in principle an infinite number of such contracts. A
closely related product is a money market M , an investment account that
gains interest.

The reader should note that bonds with different maturities represent dif-
ferent contracts only because the underlying asset to deliver, a dollar or a
different currency, is an arbitrage asset. Imagine that the asset to be delivered
is a no-arbitrage asset, such as a stock S. A contract to deliver a no-arbitrage
asset is the no-arbitrage asset itself. So even if one had infinitely many pos-
sible delivery dates T , the corresponding contract to deliver coincides with
the original underlying asset, and the term structure becomes trivial in such
a case.

Let us focus on term structure models where the underlying asset, say a
dollar $, is an arbitrage asset. Besides the dollar $ and the bonds B T we also
144 Stochastic Finance: A Numeraire Approach

have another asset, a money market account M . Money market accounts are
another no-arbitrage proxy asset for a dollar, where the dollars are invested
in the bond with the shortest available maturity. When this bond expires,
dollars are reinvested to the next bond with the shortest available maturity.
Since the dollar is an arbitrage asset and the money market is a no-arbitrage
asset, the price of the money market in terms of dollars appreciates (has a
nonnegative dt term):
dM$ (t) = r(t)M$ (t)dt. (4.20)
T
The First Fundamental Theorem of Asset Pricing implies that BM (t) should
M T2
be P martingales for all maturities T , and that BB T1 (t) or MB T1 (t) should
be PT1 martingales. The pricing measure PM that corresponds to the money
market is also known as a risk-neutral measure, and the pricing measure
PT that corresponds to the bond with maturity T is known as a T-forward
measure.

The key question is how one can capture the martingale evolution of the
prices when there are infinitely many available assets. The bonds differ only
in maturity T , and it is natural to expect that bonds with higher maturity
are more volatile with respect to the money market than bonds with smaller
maturity. Thus it is reasonable to assume that
T
dBM (t) = σ ∗ (t, T )BM
T
(t)dW M (t), (4.21)

where σ ∗ (t, T1 ) ≤ σ ∗ (t, T2 ) for T1 ≤ T2 . Note that Equation (4.21) describes

the price evolution of an unlimited number of assets B T that differ in maturity
time T . The monotonicity of σ ∗ (t, T ) can also be captured by expressing it as
an integral
Z T
σ ∗ (t, T ) = σ(t, u)du
t
for σ(t, u) ≥ 0. Therefore we can write
"Z #
T
T T
dBM (t) = BM (t) σ(t, u)du dW M (t). (4.22)
t

The inverse price can be also considered; it is given by

"Z #
T
dMB T (t) = MB T (t) σ(t, u)du dW T (t). (4.23)
t

The relationship between W M (t) and W T (t) is given by

"Z #
T
dW T (t) = −dW M (t) + σ(t, u)du dt, (4.24)
t
 Interest Rate Contracts 145

or equivalently by
"Z #
T
dW M (t) = −dW T (t) + σ(t, u)du dt. (4.25)
t

REMARK 4.1 Deterministic interest rates

It was noted in Section 1.4 that when the interest rate is deterministic, the
bonds and the money market are just deterministic multiples of each other.
In this case all T-forward measures PT and the risk-neutral measure PM are
the same. This corresponds to the situation when σ(t, T ) ≡ 0.

Let us determine the continuously compounded interest rate R(t, T, T + ǫ)

that corresponds to borrowing money at time T , returning it at time T + ǫ
as seen at time t. Assume that 1 $ is borrowed at time T . The amount to be
returned at time T + ǫ is equal to exp(ǫR(t, T, T + ǫ)) units of a dollar. Having
1 $ at time T is the same as having one bond B T at time t. The bond B T
will deliver one dollar at time T . Similarly, having exp(ǫR(t, T, T + ǫ)) dollars
at time T + ǫ is the same as having exp(ǫR(t, T, T + ǫ)) bonds B T +ǫ at time
t. The two assets should have the same price at time t, so we need
BtT = exp(ǫR(t, T, T + ǫ)) · BtT +ǫ . (4.26)
We can rewrite the above equation in terms of dollar prices as
B$T (t) = exp(ǫR(t, T, T + ǫ)) · B$T +ǫ (t), (4.27)
leading to the formula for R(t, T, T + ǫ)
1  
R(t, T, T + ǫ) = − · log(B$T +ǫ (t)) − log(B$T (t)) . (4.28)
ǫ
The forward rate f (t, T ) is defined as a limiting case when ǫ → 0 which
corresponds to an instantaneous interest rate at time T as seen from time t.
Formally,
∂
f (t, T ) = lim R(t, T, T + ǫ) = − log(B$T (t)). (4.29)
ǫ→0 ∂T
The goal is to determine the evolution of f (t, T ) as time t passes for a fixed T ,
or in other words the dynamics of df (t, T ). The strategy is to first determine
RT
the evolution of t f (t, u)du. Note that
Z T
 
f (t, u)du = − log(B$T (t)) − log(B$t (t)) = − log(B$T (t)). (4.30)
t

Now the evolution follows from applying Ito’s formula to − log(B$T (t)). The
price B$T (t) is equal to BM
T
(t) · M$ (t), and thus
T
− log(B$T (t)) = − log(BM T
(t) · M$ (t)) = − log(BM (t)) − log(M$ (t)).
146 Stochastic Finance: A Numeraire Approach

Hence,
"Z #
T    
d T
f (t, u)du = −d log(B$T (t)) = −d log(BM (t)) − d [log(M$ (t))]
t
1 T 1 T 2 1
=− T (t)
dBM (t) + 21  2 (dBM (t)) − dM$ (t)
BM T
BM (t) M $ (t)
"Z # "Z #2
T T
=− σ(t, u)du dW M (t) + 1
2 σ(t, u)du dt − r(t)dt. (4.31)
t t

Differentiating with respect to T leads to the following result.

THEOREM 4.1 Heath–Jarrow–Morton

The forward rate f (t, T ) has the dynamics
"Z #
T
df (t, T ) = σ(t, T ) · σ(t, u)du dt − σ(t, T )dW M (t). (4.32)
t

Note that the above relationship (4.32) represents in fact infinitely many
equations, one equation for each T . As time t passes, the forward rate f (t, T )
fluctuates according to (4.32). However, the forward rates f (t, T ) for different
times T are driven by the same Brownian motion W M (t), a single noise factor.
It is possible to generalize the Heath–Jarrow–Morton model to include more
noise factors.

REMARK 4.2 Relationship to spot rate models

The evolution of the spot rate r(t) is already determined by the HJM model.
Note that
Z t
r(t) = f (t, t) = df (u, t)du + f (0, t)
0
Z t Z t
= f (0, t) + [σ(u, t) · σ ∗ (u, t)] du + σ(u, t)dW M (u). (4.33)
0 0
A more compact formula for r(t) is possible for particular choices of the volatil-
ity function σ(t, T ). Assume for instance σ(t, T ) = σe−γ(T −t) for σ > 0, γ > 0,
which means that the shorter forward rates fluctuate more than the longer
forward rates, a model that is consistent with empirical data. Then
Z T Z T
σ
σ ∗ (t, T ) = σ(t, u)du = σe−γ(T −u) du = − · (e−γ(T −t) − 1),
t t γ
and consequently
σ 2 −γ(T −t) −γ(T −t)
df (t, T ) = − ·e (e − 1)dt + σe−γ(T −t) dW M (t),
γ
 Interest Rate Contracts 147

and
Z t 
σ
r(t) = f (0, t) − · (e−γ(T −t) − 1) du
σe−γ(T −t) ·
0 γ
Z t
+ σe−γ(T −t) dW M (u)
0
Z t
σ2 −γt 2
= f (0, t) + 2 (1 − e ) + σe−γ(T −t) dW M (u).
2γ 0
Rt
Note that the random variable 0 σe−γ(T −t) dW M (u) has a normal distribution
as a stochastic integral of a deterministic function. The stochastic integral
itself is a limit of the sum of normal increments, and the sum of normal
random variables is normal. Therefore negative values of the interest rate r(t)
are not excluded by this model. The interest rate r(t) satisfies the following
stochastic differential equation

dr(t) = (a(t) − γr(t))dt + σdW M (t), (4.34)

2
σ
where a(t) = γm(t)+m′ (t), and m(t) = f (0, t)+ 2γ 2 (1−e
−γt 2
) . The evolution
of the spot rate in (4.34) is known as Vasicek interest rate model, or Hull–
White interest rate model which generalized Vasicek’s model. The model
is mean reverting to the value a(t)
γ that makes the dt term zero in (4.34). Since
this model allows for a negative interest rate r(t), several modifications were
suggested. Another popular spot rate model is
p
dr(t) = (a(t) − b(t)r(t))dt + σ r(t)dW M (t) (4.35)

which cannot become negative. This is known as the Cox–Ingersoll–Ross

interest rate model.

References and Further Reading

Historically the mathematical treatment of the fixed income markets started
with modeling the spot rate. The Vasicek model appears first in Vasicek
(1977), and it was later generalized for time-varying coefficients by Hull and
White (1990). Another popular spot rate model is due to Cox–Ingersoll–Ross
which appeared in Cox et al. (1985). The Heath–Jarrow–Morton model in
Heath et al. (1992) established the no-arbitrage evolution of the forward rates.
Since the spot rate is a special case of the forward rate, it also determines all
possible no-arbitrage evolutions of the spot rate.
148 Stochastic Finance: A Numeraire Approach

The link between the HJM model and the no-arbitrage evolution of the
LIBOR rates is due to Brace et al. (1997). The formula for caplets appeared
earlier in Black (1976). Jamshidian (1997) extended these results to swap
rates. The reader may refer to the monographs of Brigo and Mercurio (2006),
Sadr (2009), Filipovic (2009), James and Webber (2000), Pellser (2000), or
Musiela and Rutkowski (2008) for more information on the subject.

Exercises
4.1 Consider the evolution of the forward rate given by

df (t, T ) = −σdW M (t).

Since the evolution violates the no-arbitrage condition in the HJM model,
arbitrage is possible. Find the arbitrage portfolio.

Hint: The goal is to find a portfolio P consisting of one or several bonds and
the money market such that

dPM (t) = µ(t)dt

for µ(t) > 0. Then we can start with P0 = 0 and end up with PT > 0. Finding
T
the arbitrage portfolio can be split in two steps: first determine dBM (t). It
T
turns out that in this model, dBM (t) has a positive dt term, but it also has a
noise term dW M (t), and thus the arbitrage is not locked considering just two
assets B T and M . Second, the arbitrage can be locked with another bond of a
different maturity which would cancel the noise term dW M (t) in the evolution
of the portfolio in the form

Pt = B T2 + ∆T1 (t) · B T1 + ∆M (t) · M,

where T2 > T1 . Determine ∆T1 such that

 
T2 T1
dPM (t) = d BM + ∆T1 (t)BM = µ(t)dt

for µ(t) > 0.

Chapter 5
Barrier Options

A barrier option is a contract whose payoff depends on the event that the
underlying price crosses a certain boundary. Typical variants of the barrier
option depend only on the assets X and Y , and thus they can be regarded as
contracts on two assets. Barrier options are cheaper than their corresponding
plain vanilla counterparts. They may appeal to investors who want to have a
higher exposure on the payoff of the plain vanilla option. Such investors may
buy more units of the barrier option than the plain vanilla options for the
same price, but with the risk that the barrier option may expire worthless in
contrast to the plain vanilla option. Barrier options typically appear in foreign
exchange markets.

The first section of this chapter describes different types of barrier options.
Barrier options come in two flavors: knock-out and knock-in. Knock-out
options expire worthless if the barrier is hit during the life of the option. On
the other hand, knock-in options convert to a plain vanilla option at the time
of the first hit of the barrier. The barrier is usually a constant in terms of
dollar prices. However, in terms of no-arbitrage assets, such as the bond B T ,
the barrier takes an exponential form because of the time value of money.
This is an important observation for pricing such contracts since the price of
the option has to be computed with respect to a no-arbitrage asset.

The second section of this chapter shows that a barrier contract has a
corresponding plain vanilla “sibling” contract that has the same price up to
the first time of hitting the barrier. This result is based on the exchangeability
of the assets X and Y , meaning that the returns of the prices with respect
to the reference asset Y and the reference asset X have the same distribution
in the corresponding martingale measures. We can express this assumption
mathematically as
   
XY (T ) YX (T )
LYt = LXt .
XY (t) YX (t)

As a consequence, the holder of the barrier contract is ambivalent between

having a contract that pays off when the price first hits a lower barrier and
then goes up, and a contract that pays off when the price first hits a lower
barrier and ends up even lower. The second contract is a plain vanilla type

149
150 Stochastic Finance: A Numeraire Approach

option since the barrier must be crossed in order to collect the payoff. As a
side result of this approach, we can easily derive the distribution of the hitting
time of the boundary. Determination of the distribution of the hitting time
is an interesting problem on its own since it may have consequences on the
value of a portfolio even if it does not include barrier options. For instance a
very low boundary may be hit during a market crash.

When the interest rate r is positive, the barrier takes an exponential form,
and the corresponding “sibling” contract depends on a power option Rα . A
specific choice of α converts the exponential barrier problem for the assets X
and Y to a constant barrier problem for the assets Rα and Y . For computing
the price of the barrier option, we decompose its payoff in two Arrow–Debreu
securities, and determine their “sibling” plain vanilla contracts in terms of the
power options Rα . The third section illustrates how to compute the price of
the down-and-in call option by combining the two Arrow–Debreu securities
in order to express the barrier option as a plain vanilla European call option
on two power options. This part of the book is mathematically demanding,
in particular the transformation of the exponential boundary to the constant
boundary for the asset Rα . The reader who is interested in the conceptual
ideas rather than in the computational details may just follow the text which
covers the case when r = 0, which is relatively simple.

5.1 Types of Barrier Options

Let X and Y be two arbitrary assets (both arbitrage or no-arbitrage). A
barrier knock-out option pays off f Y (XY (T )) units of an asset Y , subject to
f¯Y (t, XY (t)) ≥ 0 at all times t ∈ [0, T ] for some function f¯Y (t, x). The last
condition says that the price XY (t) must stay in a certain region at all times
in order to collect the option payoff. Similarly, the barrier knock-out option
can be defined as a contract that pays off f X (YX (T )) units of an asset X,
subject to f¯X (t, YX (t)) ≥ 0 at all times t ∈ [0, T ] for some function f¯X (t, x).
When the two definitions describe the same contract, the functions f¯Y (t, x)
and f¯X (t, x) are related by



f¯Y (t, x) = f¯X t, x1 · x, or f¯X (t, x) = f¯Y t, x1 · x. (5.1)
This is analogous to the relationship between the payoff functions f Y and f X .

Example 5.1
A down-and-out call option pays off

(XT − K · YT )+ if min XY (t) ≥ L. (5.2)

0≤t≤T
 Barrier Options 151

The payoff function is the same as for a European call option, where f Y (x) =
(x − K)+ , and f X (x) = (1 − K · x)+ . The barrier condition is given by
f¯Y (t, x) = x − L, or equivalently by f¯X (t, x) = 1 − L · x.

Similarly, the up-and-out call option pays off

(XT − K · YT )+ if max XY (t) ≤ U. (5.3)

0≤t≤T

The barrier condition is given by f¯Y (t, x) = U − x, or equivalently by

f¯X (t, x) = U · x − 1.

A barrier knock-in option pays off f Y (XY (T )) units of an asset Y , subject

to f¯Y (t, XY (t)) < 0 for at least one time t ∈ [0, T ] for some function f¯Y (t, x).
The last condition says that the price XY (t) must enter a certain region for
at least one time in order to collect the option payoff. Similarly, the barrier
knock-in option can be defined as a contract that pays off f X (YX (T )) units
of an asset X, subject to f¯X (t, YX (t)) < 0 for at least one time t ∈ [0, T ] for
some function f¯X (t, x). When the two definitions describe the same contract,
the functions f¯Y and f¯X are related by



f¯Y (t, x) = f¯X t, x1 · x, or f¯X (t, x) = f¯Y t, x1 · x.

Example 5.2
A down-and-in call option pays off

(XT − K · YT )+ if min XY (t) < L. (5.4)

0≤t≤T

The payoff function is the same as for a standard European call option, where
f Y (x) = (x − K)+ , and f X (x) = (1 − K · x)+ . The barrier condition is given
by f¯Y (t, x) = x − L, or equivalently by f¯X (t, x) = 1 − L · x.

Similarly, the up-and-in call option pays off

(XT − K · YT )+ if max XY (t) > U. (5.5)

0≤t≤T

The barrier condition is given by f¯Y (t, x) = U − x, or equivalently by

f¯X (t, x) = U · x − 1.

If we denote by V KO a barrier knock-out call option and by V KI a barrier

knock-in call option, we have a simple relationship with the plain vanilla
European call option V EC :

V EC = V KO + V KI . (5.6)
152 Stochastic Finance: A Numeraire Approach

This relationship is easy to see. When the barrier is crossed, the knock-in
option is activated but the corresponding knock-out option becomes worthless.
Similarly, if the barrier is not crossed, the knock-out option is alive but the
corresponding knock-in option is worthless.

REMARK 5.1 Exponential boundaries

Consider the situation when Y is an arbitrage asset $, and X is a no-arbitrage
asset S. Then the down-and-in call option pays off

(ST − K · $T )+ if min S$ (t) < L.

0≤t≤T

Assuming a constant interest rate with B$T (t) = e−r(T −t) , we have S$ (t) =
SB T (t) · B$T (t) = e−r(T −t) · SB T (t), and the above contract can be rewritten
in terms of no-arbitrage assets S and B T as

(ST − K · BTT )+ if min e−r(T −t) SB T (t) < L.

0≤t≤T

In this case, the barrier condition is given by

f¯T (t, x) = x − L · er(T −t)

when we express it in terms of the bond B T . Therefore we will consider this

boundary type in the next section.

5.2 Barrier Option Pricing via Power Options

The basic idea of pricing barrier options is that the barrier option has a
related plain vanilla option that has the same price. The pricing problem of the
barrier option then reduces to the pricing problem of the corresponding plain
vanilla option. Let us illustrate this technique in the simplest case when the
interest rate r is zero, and the barrier function is in the form f¯Y (t, x) = x − L.

5.2.1 Constant Barrier

Consider a down-and-in call option that pays off

(XT − K · YT )+ if min XY (t) ≤ L.

0≤t≤T

The payoff of a European call option can be expressed as a combination of

two Arrow–Debreu securities, one that pays off a unit of the asset X on the
event
A = { min XY (t) ≤ L; XY (T ) ≥ K}, (5.7)
0≤t≤T
 Barrier Options 153

and one that pays off −K units of the asset Y on the same event. Thus it is
sufficient to find the price of the corresponding Arrow–Debreu securities. Let
us denote by V the Arrow–Debreu security that pays off
VT = IA (ω) · YT .
Note that when the price of XY (t) is on the boundary L (meaning XY (t) = L),
the price of the Arrow–Debreu security is simply
Vt = PYt (XY (T ) ≥ K) · Yt
since the barrier is already hit. Furthermore, we can write
Vt = PYt (XY (T ) ≥ K) · Yt


= PYt XY (t) · exp(σW Y (T − t) − 21 σ 2 (T − t)) ≥ K · Yt


= PX
t XY (t) · exp(σW X (T − t) − 12 σ 2 (T − t)) ≥ K · L1 · Xt
 
YX (T )
= PX
t X Y (t) · ≥ K · L1 · Xt
YX (t)
 
XY (T )
= PX
t
1
K ≥ · L1 · Xt
[XY (t)]2
 2 
= PX
t
L 1
K ≥ XY (T ) · L · Xt . (5.8)

The equality between the second and the third line follows from the fact that
Xt = L · Yt on the boundary, and that the distribution of W Y under PY is
the same as the distribution of W X under PX . In mathematical notation we
can write
LY (W Y (T )) = LX (W X (T )),
meaning that the probability laws of the corresponding Brownian motions
agree. In a geometric Brownian motion model, this statement is equivalent to
   
XY (T ) YX (T )
LYt = LX
t , (5.9)
XY (t) YX (t)
implying that the assets X and Y are exchangeable. We can also rewrite this
relationship as  
Y X [XY (t)]2
Lt (XY (T )) = Lt .
XY (T )
In conclusion, we have proved that on the boundary XY (t) = L, the Arrow–
Debreu security is equal to
 2 
Vt = PYt (XY (T ) ≥ K) · Yt = PX
t
L
K ≥ X Y (T ) · L1 · Xt .
 2 
L 1 1
But PXt K ≥ XY (T ) · L · Xt corresponds to L units of the Arrow–Debreu
security U with a payoff
UT = IB (ω) · XT ,
154 Stochastic Finance: A Numeraire Approach

where B is the event

2
B = { LK ≥ XY (T )}.
Let τL be the first hitting time of the boundary

τL = inf{t ≥ 0 : XY (t) ≤ L}. (5.10)

We have already shown that VτL = L1 UτL . Moreover, if τL > T , both U and
V are worthless. Therefore we have also the relationship
1
VτL ∧T = L · UτL ∧T , (5.11)

where x ∧ y = min(x, y). Thus for all times t ≤ τL , we also have

VY (t) = EYt [VY (τL ∧ T )] = 1 Y

L Et [UY (τL ∧ T )] = 1
L UY (t).

Therefore we have Vt = L1 Ut for all times t ≤ τL . The price of the Arrow–

Debreu security with a barrier feature V can be computed from the price of
the Arrow–Debreu security U that is a plain vanilla contract.

Figure 5.1 illustrates this situation. Before the price XY (t) hits the barrier
τL for the first time, the price of the Arrow–Debreu security V agrees with
the price of L1 U . The price of the plain vanilla Arrow–Debreu security that
pays one unit of an asset Y at time T if XY (T ) ≥ K dominates the price of V
which has the barrier feature up to the first hitting time τL . Once the barrier
is hit, the security V agrees with its plain vanilla counterpart, and the price
of L1 U is no longer relevant.

Similar to an Arrow–Debreu security V that pays off a unit of an asset Y

e
on the event A defined in (5.7), we can consider an Arrow–Debreu security U
that pays off a unit of an asset X on the same event A

eT = IA (ω) · XT .
U

Following the same arguments as in (5.8), we can show that when XY (t) = L,
we have
et = PX (XY (T ) ≥ K) · Xt = PY ( L2 ≥ XY (T )) · LYt .
U t t K

e has the same price as L units of an Arrow–Debreu

Therefore the security U
2
e
security V that pays off an asset Y when { LK ≥ XY (T )} up to the first time
τL when the barrier is hit.

In conclusion, the barrier option

(XT − K · YT )+ if min XY (t) ≤ L

0≤t≤T
 Barrier Options 155

2.5

1.5
Price

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Time

FIGURE 5.1: The evolution of the price XY (t) (top graph). Consider an
Arrow–Debreu security V that pays a unit of an asset Y at time T if the
price of XY (T ) exceeds K, subject to the condition that XY (t) ≤ L for some
t ∈ [0, T ]. The parameters are XY (0) = 1.5, K = 1.5 (top solid line), L = 1
2
(middle solid line), LK ≈ 0.666 (bottom solid line), σ = 0.2, T = 10. The price
of V agrees up to time τL with the price of L1 units of a plain vanilla contract
2
that pays a unit of X if XY (T ) ≤ LK (the bottom graph starting at 0.082 and
ending at 0). The first hit happens at time τL = 7.175 (vertical dash line).
Compare the price of V with the price of the Arrow–Debreu security that
pays a unit of an asset Y when XY (T ) ≥ K (plain vanilla type, the graph
starting at 0.376 and ending at 1); it dominates the price of V up to time τL .
When the price XY (t) hits the barrier L = 1, the prices of V and L1 U agree:
V = L1 U . Once the barrier is hit, the price of V agrees with its plain vanilla
counterpart. The contract U becomes irrelevant once the barrier is hit.
156 Stochastic Finance: A Numeraire Approach

has the same price as a plain vanilla option with a payoff

K +
(LYT − L XT )

up to the first time τL when the barrier is hit.

REMARK 5.2 Distribution of the first hitting time of a geometric

Brownian motion.
It is interesting to note that we can determine the distribution of the hitting
time
τL = inf{t ≥ 0 : XY (t) ≤ L}
from the prices of two Arrow–Debreu securities. Consider a financial contract
V that pays off a unit of Y at time τL

VτL = YτL .

We can also rewrite this payoff as

VT = I(τL ≤ T ) · YT .

Define two Arrow–Debreu securities V 1 and V 2 by

VT1 = I( min XY (t) ≤ L, XY (T ) > L) · YT ,

t∈[0,T ]

and
VT2 = I( min XY (t) ≤ L, XY (T ) ≤ L) · YT .
t∈[0,T ]

Note that
Vt = Vt1 + Vt2 .
As for the price of V 2 , the condition XY (T ) ≤ L already implies
mint∈[0,T ] XY (t) ≤ L, and thus it is a plain vanilla security with price

Vt2 = PY (XY (T ) ≤ L) · Y.

The security V 1 is a knock-in Arrow–Debreu security with the barrier L and

strike K = L, and thus its price is equal to L1 units of a plain vanilla Arrow–
Debreu security with the payoff I(XY (T ) ≤ L) · L1 · X as explained in the
previous text. Thus the price of V 1 is given by

Vt1 = PX (XY (T ) ≤ L) · 1
L ·X

up to time τL . Thus we have

V = PY (τL ≤ T ) · Y
= PX (XY (T ) ≤ L) · 1
L · X + PY (XY (T ) ≤ L) · Y.
 Barrier Options 157

Therefore

PY (τL ≤ T ) = PX (XY (T ) ≤ L) · XYL(0) + PY (XY (T ) ≤ L)

   √ 
= N σ√1 T log XYL(0) − 12 σ T · XYL(0)
   √ 
+N σ√1 T log XYL(0) + 21 σ T .

Thus we have determined the cumulative distribution function of the stopping

time τL . Its density is given by
      √ 2 
XY (0) 1 L 1
∂ Y log L 
√
σ T
log XY (0) + 2 σ T 
P (τL ≤ T ) = √ exp − .
∂T σ 2πT 3 2T

5.2.2 Exponential Barrier

When the interest rate r is greater than zero, the barrier takes an exponen-
tial form. We will show here that when XY (t) hits an exponential boundary,
there is a corresponding power option Rα whose price is hitting a constant
boundary. Thus we are able to transform the problem of an exponential bound-
ary to a constant boundary, although this requires a proxy asset Rα .

Recall that a power option Rα pays off [XY (T )]α units of an asset Y at
time T . Note that R0 = Y , and R1 = X, so the assets X and Y are just
special cases of power options. The price of the power option is given by

RYα (t) = exp( 12 α(α − 1)σ 2 (T − t)) · [XY (t)]α . (5.12)

The evolution of the power option price is given by

dRYα (t) = ασRYα (t)dW Y (t) (5.13)

with a closed form solution

RYα (T ) = RYα (t) · exp ασW Y (T − t) − 21 α2 σ 2 (T − t) . (5.14)

The volatility of the power option Rα is α times the volatility σ of the price
XY . Since Rα itself is a no-arbitrage asset with a positive price, it can be used
as a numeraire, leading to a pricing measure P(α) that is associated with Rα .
From the exchangeability of the assets Y and Rα , the dynamics of the inverse
price YRα (t) are given by

dYRα (t) = ασYRα (t)dW (α) (t). (5.15)

158 Stochastic Finance: A Numeraire Approach

Now consider an Arrow–Debreu security V that pays off a unit of an asset

Y at time T on the event
A = { min (e−r(T −t) XY (t)) ≤ L; XY (T ) ≥ K}.
0≤t≤T

The first time of hitting the boundary is given by

τL = inf{t ≥ 0 : XY (t) ≤ Ler(T −t) }.
Let us show that for all t ≤ τL ,


1 α (α)
Vt = L · Vt ,

where V (α) is an Arrow–Debreu security that pays off a unit of the power
option Rα at time T on the event
L2
B = {XY (T ) ≤ K }
2r
for α = 1 − σ2 .

THEOREM 5.1
The Arrow–Debreu security with a barrier feature V that pays off
VT = IA (ω) · YT
for
A = { min (e−r(T −t) XY (t)) ≤ L; XY (T ) ≥ K}
0≤t≤T

has the same price up to the first hitting time of the barrier
τL = inf{t ≥ 0 : XY (t) ≤ Ler(T −t) }


1 α
as L units of a plain vanilla Arrow–Debreu security V (α) that pays off
(α)
VT = IB (ω) · RTα
for
L2
B = {XY (T ) ≤ K },
2r
with α = 1 − σ2 .

α
PROOF Let us first show that V and L1 · V (α) have the same price
when the price XY (t) is on the barrier at time t, meaning XY (t) = Ler(T −t) .
Note that in this case, the price of the power option RYα (t) is given by
RYα (t) = exp( 21 α(α − 1)σ 2 (T − t)) · [XY (t)]α
= exp( 21 α(α − 1)σ 2 (T − t)) · eαr(T −t) · Lα
= exp([r + 21 (α − 1)σ 2 ] · α(T − t)) · Lα .
 Barrier Options 159
2r
In particular, when α = 1 − 2r
σ2 , the above price simplifies to L1− σ2 , so we
have
2r 2r
(1− σ2 )
RY (t) = L1− σ2
2r
when XY (t) is on the barrier Ler(T −t) . As for the value of V , for α = 1 − σ2
we get (assuming α > 0)

Vt = PYt (XY (T ) ≥ K) · Yt
= PYt ([XY (T )]α ≥ K α ) · Yt = PYt (RYα (T ) ≥ K α ) · Yt



= PYt RYα (t) · exp ασW Y (T − t) − 21 α2 σ 2 (T − t) ≥ K α · Yt
   

(α) 1 α
= Pt RYα (t) · exp ασW (α) (T − t) − 12 α2 σ 2 (T − t) ≥ K α · L · Rtα
 
(α) YRα (T )
α
= Pt RYα (t) · ≥ K α · L1 · Rtα
YRα (t)
 α 2 α 
α
(α) [RY (t)] α
= Pt K ≥ R Y (T ) · L1 · Rtα
!
2r
(1− σ2 )
 2 (1− 2r2 ) 2r
(1− σ2 )
2r 2r
(1− σ2 )
L σ 1 (1− σ2 )
= Pt K ≥ R Y (T ) · L · Rt

2r
(1− σ2 )
 
2r 2r
(1− σ2 )
L2 1 (1− σ2 )
= Pt K ≥ XY (T ) · L · Rt .

2r
(1− σ2 )
When α ≤ 0, we obtain the same result. We have used the fact that Rt =
2r
1− σ2 r(T −t)
L · Yt when XY (t) = Le , and that the distribution of the Brownian
motion W Y under the probability measure PY is the same as the distribution
of the Brownian motion W (α) under the probability measure P(α) . The latter
statement is another way of saying that the assets Y and Rα are exchangeable,
meaning that  α   
RY (T ) (α) YRα (T )
LYt = Lt .
RYα (t) YRα (t)
We can also rewrite this relationship as
 
(α) [RYα (t)]2
LYt (RYα (T )) = Lt .
RYα (T )

We have shown that

2r
(1− σ2 )
 
2r 2r
(1− σ2 )
L2 1 (1− σ2 )
Vt = PYt (XY (T ) ≥ K) · Yt = Pt K ≥ XY (T ) · L · Rt .

Thus when XY (t) is on the barrier Ler(T −t) , the price of the Arrow–Debreu
2r  
2r 2r
(1− 2 ) L2 1 (1− σ2 ) (1− σ2 )
security is also equal to Pt σ K ≥ X Y (T ) · L ·Rt . But this
2r
representation corresponds to ( L1 )(1− σ2 ) units of an Arrow–Debreu security
160 Stochastic Finance: A Numeraire Approach

V (α) that pays off

2r
(α) (1− σ2 )
VT = IB (ω) · RT ,
L2
where B = {XY (T ) ≤ K }. Since

2r
1 (1− σ2 ) (α)
VτL ∧T = L · VτL ∧T ,

we also have

2r
1 (1− σ2 ) (α)
Vt = L · Vt

for all t ≤ τL .

Figure 5.2 illustrates the price evolution of the Arrow–Debreu security V

in more detail together with the corresponding

α prices of the asset X, power
option Rα , the plain vanilla security L1 V (α) and a plain vanilla Arrow–
Debreu security that pays off a unit of the asset Y at time T if XY (T
) ≥ K.
α
Up to the first hitting time τL of the boundary, the prices of V and L1 V (α)
agree, and they are dominated by the price of the plain vanilla Arrow–Debreu
security that pays off a unit of the asset Y at time T if XY (T ) ≥ K. Once
the barrier is hit, the security V agrees with its plain vanilla counterpart, and
the security V (α) is no longer relevant.

Similarly, the Arrow–Debreu security U that pays off a unit of an asset X

on the event

A = { min (e−r(T −t) XY (t)) ≤ L; XY (T ) ≥ K}

0≤t≤T

has the same price up to time τL as an Arrow–Debreu security U (α) that pays
2r 2r
off L1+ σ2 units of a power option R(− σ2 ) on the event
L2
B = {XY (T ) ≤ K }.

See Exercise 5.2 for more details.

5.3 Price of a Down-and-In Call Option

Let us illustrate how to compute a price of a down-and-in call option using
power options. The prices of other types of barrier options can be determined
in an analogous way. The down-and-in barrier option with a payoff

(XT − K · YT )+ if min (e−r(T −t) XY (t)) ≤ L

0≤t≤T
 Barrier Options 161

2.5

1.5
Price

0.5

0
0 1 2 3 4 5 6 7 8 9 10
Time

FIGURE 5.2: The evolution of the prices XY (t) (top graph) and RYα (t)
(immediately below it) with XY (0) = 1.5, K = 1.5, L = 1, r =
 0.01,
 σ = 0.2,
2 α
T = 10. This gives the values of α = 1− σ2r2 = 0.5, K α ≈ 1.225, LK ≈ 0.816.
Consider an Arrow–Debreu security V that pays off one unit of an asset Y
at time T when XY (T ) ≥ K = 1.5 (upper solid line) if the price XY (t) gets
below the barrier Ler(T −t) (middle solid curve). The security V is equivalent
to the one that pays off a unit of Y when RYα (T ) ≥ K α ≈ 1.225 (top dash
line), subject to RYα (t) ≤ Lα = 1 (middle dash line) in terms of the price of
the power option Rα . Note that each time the price XY (t) hits the barrier
Ler(T −t) , the price of RYα (t) is at Lα = 1. The first hitting time of the barrier
happens at time τL = 5.445 (vertical dash line). Up to time τL , the security V
has the same price as ( L1 )α units of a security V (α) that pays off one unit of Rα
 2 α
when RYα (T ) ≤ LK ≈ 0.816 (bottom dash line). This event is equivalent
2 2
to XY (T ) ≤ LK ≈ 0.666 (bottom solid line). Since LK < Ler(T −t), the price of
XY (t) must cross the barrier so that V (α) does not expire worthless. Therefore
V (α) is a plain vanilla contract. The price of ( L1 )α · V (α) is the bottom graph
up to time τL ; it agrees with the price of V . The graph immediately above
corresponds to the price of a plain vanilla Arrow–Debreu security that pays
off a unit of an asset Y if XY (T ) ≥ K. It is more expensive since it does not
have a barrier feature in contrast to the security V . Once the price XY (t) hits
the barrier Ler(T −t) at time τL , the security V is knocked in, and from that
moment on it agrees with its plain vanilla counterpart.
α Also at time τL , the
prices of V (now a plain vanilla contract) and L1 V (α) agree. The price of
V expires at 1 since XY (T ) ends up above K
αin this scenario. Compare the
price of V from time τL with the price of L1 V (α) that expires worthless.
162 Stochastic Finance: A Numeraire Approach

is equivalent up to time τL to a plain vanilla European call option V with a

payoff

 2r 2r +
2r (− 2 ) 2r (1− 2 )
VT = L σ2 +1 RT σ − KL σ2 −1 RT σ
 2r 2r +
2r (− 2 ) K (1− σ2 )
= L σ2 · LRT σ − L RT . (5.16)

This is clear from the representation of an option payoff as two Arrow–Debreu

securities. The security that pays off a unit of an asset X has a corresponding
2r
plain vanilla counterpart that is settled in units of R(− σ2 ) ; the security that
pays off a unit of an asset Y has a corresponding plain vanilla counterpart
2r
that is settled in units of R(1− σ2 ) .

The payoff that corresponds to power options can also be expressed in terms
of the assets X and Y as
 2r 2r +
2r (− 2 ) K (1− σ2 )
L σ2 · LRT σ − R
L T
 +
2r 2r 2r
= L σ2 L[XY (T )]− σ2 · YT − K
L [XY (T )] 1− σ2
· YT
  2r2
L σ
K

+
= · L · YT − L · XT .
XY (T )

Let us determine the price of the option V . From the Black–Scholes formula,
 2r   2r
(− 2 ) −2r/σ2 2r − 2
Vt = LPt σ RR1−2r/σ2 (T ) ≥ K L · L σ2 × Rt σ

 2r   2r
(1− σ2 ) −2r/σ2 2r 1− 2
+ −K P
L t R R 1−2r/σ 2 (T ) ≥ K
L · L σ2 × Rt σ .

Since
α α α
RR 1+α (T ) = RY (T ) · YRα+1 (T ) = [XY (T )] · [YX (T )]α+1 = YX (T ),

we can rewrite the above formula as

 2r  2r
(− σ2 ) 2r − 2
L2
Vt = LPt (XY (T ) ≤ K ) · L σ2 · Rt σ
 2r  2r
(1− σ2 ) 2r 1− σ2
L2
+ −K L P t (XY (T ) ≤ K ) · L σ2 · Rt . (5.17)
 Barrier Options 163
(α) 2
Let us determine Pt (XY (T ) ≤ LK ) for α > 0:
2
  2 α 
(α) (α)
Pt (XY (T ) ≤ LK ) = Pt [XY (T )]α ≤ LK
  2 α  

(α) (α) K α
= Pt RYα (T ) ≤ LK = Pt L 2 ≤ Y R α (T )



(α) K α
= Pt L2 ≤ YRα (t) · exp(ασW (T − t) − 21 α2 σ 2 (T − t))
(α)



(α) K α α (α) 1 2
= Pt L 2 ≤ [YX (t)] · exp(ασW (T − t) − α(α − 2 )σ (T − t))
 (α)
 2
 √ 
(α) W 1 L 1
= Pt − √T −t ≤ σ√T −t log XY (t)·K − (α − 2 )σ T − t
  2
 √ 
= N σ√1T −t log XY L(t)·K − (α − 21 )σ T − t .

We get the same result for α ≤ 0. We can also express the power option Rα
in terms of the asset Y as
Rtα = exp( 12 α(α − 1)σ 2 (T − t))[XY (t)]α · Yt .
Therefore
 2r  2r
(− 2 ) 2r − σ2
L2
Vt = LPt σ (XY (T ) ≤ K) · L σ2 · Rt
 2r  2r
(1− σ2 ) 2r 1− σ2
L2
+ −K P
L t (X Y (T ) ≤ K ) · L σ 2
· Rt
h   2
 √ i
= N σ√1T −t log XY L(t)·K + ( σ2r2 + 12 )σ T − t (5.18)
h r(T −t) i 2r2 +1
σ
× LeXY (t) · Xt
h   2
 √ i
−K N σ√1T −t log XY L(t)·K + ( σ2r2 − 12 )σ T − t
h i 2r2 −1
Ler(T −t) σ
× XY (t) · Yt .

This is a relationship of the down-and-in option V before its knock in with

the no-arbitrage assets X and Y . If we define
uY (t, x) = EY [VY (T )|XY (t) = x],
we can also write the price of the down-and-in barrier option as
h   2  √ i
uY (t, x) = N σ√1T −t log x·K L
+ ( σ2r2 + 21 )σ T − t (5.19)
h i 2r2 +1
Ler(T −t) σ
× x ·x
h    √ i
√1 L2
−K N σ T −t
log x·K + ( σ2r2 − 21 )σ T − t
h i 2r2 −1
Ler(T −t) σ
× x .
164 Stochastic Finance: A Numeraire Approach

When the option is written on a stock S and on a dollar $, the corresponding

no-arbitrage assets are X = S, and Y = B T . In order to get the price of V in
terms of a dollar $, we can use the change of numeraire formula. Note that

S$ (t) = SB T (t) · B$T (t) = e−r(T −t) · SB T (t) = e−r(T −t) XY (t).

Thus the dollar price V$ (t) of the contract takes the following form

   √  h i 2r2 +1
L2 σ
V$ (t) = N √1 log + ( σr2 + 21 )σ T − t · S$L(t) · S$ (t)
σ T −t S$ (t)·K
   √  h i 2r2 −1
L2 σ
− Ke−r(T −t)N √1
σ T −t
log S$ (t)·K + ( σr2 − 21 )σ T − t · S$L(t) .

If we denote by v $ (t, S$ (t)) = V$ (t) the dollar price function, we can also write
  √ 
2r +1
L2
v $ (t, x) = N √1 + ( σr2 + 21 )σ T − t · Lx σ2 · x
σ T −t
log x·K
  2 √ 
2r −1
− Ke−r(T −t) N σ√1T −t log x·K
L
+ ( σr2 − 21 )σ T − t · Lx σ2 . (5.20)

The functions v $ (t, x) and uY (t, x) are linked by

v $ (t, x) = e−r(T −t) · uY (t, er(T −t) x), (5.21)

which is the same as equation (3.59). When comparing Equation (5.18) with
Equation (5.20), the term inside the normal cumulative distribution function
changes from σ2r2 to σr2 due to the change of numeraire from the bond prices
to the dollar prices.

Let us find the hedging portfolio for the down-and-in barrier call option.
Since the Black–Scholes formula in (5.17) is written in a form of a self-financing
portfolio, it is also a hedging portfolio. Thus one should have
2r 2r
(− σ2 ) L2
Pt (XY (T ) ≤ · L σ2 +1
K )
  2
 √  2r
= N σ√1T −t log XY L(t)·K + ( σ2r2 + 21 )σ T − t · L σ2 +1

2r
units of a power option R− σ2 , and
2r 2r
(1− σ2 ) 2
− KPt (XY (T ) ≤ LK ) · L σ2 −1
  2
 √  2r
= −KN σ√1T −t log XY L(t)·K + ( σ2r2 − 21 )σ T − t · L σ2 −1

2r 2r 2r
units of a power option R1− σ2 at time t. The power options R− σ2 and R1− σ2
can be replicated by trading in the underlying assets X and Y according to
 Barrier Options 165

the following self-financing representations

" # " #
2r
− σ2
h r(T −t) i 2r2 +1 h i 2r2
2r e σ 2r r(T −t) er(T −t) σ
Rt = −( σ2 ) XY (t) · Xt + (1 + σ2 )e XY (t) · Yt ,

(5.22)
and
" #
2r
(1− 2 )
h i 2r2 −1
2r er(T −t) σ 1
Rt σ = (1 − σ2 ) XY (t) · XY (t) · Xt
" #
2r

h er(T −t) i σ2r2 −1
+ σ2 XY (t) · Yt . (5.23)

See Exercise 5.1 for more details. We conclude that the hedging positions are
given up to time τL by
  2
 √ 
∆X (t) = N σ√1T −t log XY L(t)·K + ( σ2r2 + 12 )σ T − t (5.24)
" #
h r(T −t) i 2r2 +1
σ
× −( σ2r2 ) LeXY (t)
   √ 
√1 L2
−KN σ T −t
log XY (t)·K + ( σ2r2 − 21 )σ T − t
" #
h i 2r2 −1
2r Ler(T −t) σ 1
× (1 − σ2 ) XY (t) · XY (t) ,

and
   √ 
L2
∆Y (t) = N √1
σ T −t
log XY (t)·K + ( σ2r2 + 21 )σ T − t (5.25)
" #
h r(T −t) i 2r2
σ
× (1 + σ2r2 )Ler(T −t) LeXY (t)
   √ 
√1 L2
−KN σ T −t
log XY (t)·K + ( σ2r2 − 21 )σ T − t
" #
2r

h Ler(T −t) i σ2r2 −1
× σ2 XY (t) .

5.4 Connections with the Partial Differential Equations

The price function uY in Equation (5.19) satisfies the Black–Scholes partial
differential equation (3.52)

uYt (t, x) + 21 σ 2 x2 uYxx (t, x) = 0

166 Stochastic Finance: A Numeraire Approach

for times before the knock-in and before the expiration time T . This is clear
from the fact that the price process uY (t, XY (t)) must be a PY martingale,
and thus the corresponding dt term must be equal to zero. The form of the
partial differential equation agrees with its plain vanilla counterpart, and it is
the same for all types of the barrier options that depend on the price process
XY (t).

The only difference is in the boundary and the terminal conditions. The
knock-in barrier option has the terminal condition equal to

uY (T, x) = 0 (5.26)

for times before the knock-in and before the expiration time T . This corre-
sponds to the situation that the barrier is not hit by the expiration, and thus
the option expires worthless. On the other hand, when the barrier is hit, the
price of the barrier option agrees with the price of its European option coun-
terpart. The barrier is given by the relationship x = Ler(T −t) , and thus the
boundary condition is

uY (t, Ler(T −t) ) = EY [(XY (T ) − K)+ |XY (t) = Ler(T −t) ] (5.27)
r(T −t)
= N (d+ )Le − N (d− ),

where
 √ 
XY (t)
d± = √1 log ± 12 σ T − t (5.28)
σ T −t K

√
L
= σ√1T −t log K + ( σr2 ± 21 )σ T − t.

One can easily check that the price of the barrier option in (5.19) satisfies
these boundary conditions. A straightforward computation also confirms that
the price satisfies the Black–Scholes partial differential equation.

Similarly, the price function v $ from Equation (5.20) satisfies the same
Black–Scholes partial differential equation as in (3.61):

−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx

$
(t, x) = 0.

The terminal condition for v $ agrees with the terminal condition for uY :

v $ (T, x) = 0, (5.29)

but the exponential boundary Ler(T −t) for the function uY transforms into a
constant boundary L for the function v $ :

v $ (t, L) = N (d+ )L − N (d− ), (5.30)

where
√
d± = √1 log L
+ ( σr2 ± 21 )σ T − t. (5.31)
σ T −t K
 Barrier Options 167

The price functions uX (t, x) = uY (t, x1 ) · x and v S (t, x) = v $ (t, x1 ) · x satisfy

the same partial differential equations as their European option counterparts
(Equations (3.55), and (3.69)). The terminal and the boundary conditions can
be determined from the perspective mapping.

References and Further Reading

The price of the barrier option in the geometric Brownian motion model was
computed in Rubinstein and Reiner (1991). Kunitomo and Ikeda (1992) de-
termined prices of barrier options with exponential boundaries. Our approach
is related to the work of Carr and Chou (1997) who found several connections
between barrier and plain vanilla options based on the symmetry argument.
Broadie et al. (1997) studied the difference between the prices of discretely
and continuously monitored barrier options. Other papers related to pricing
barrier options include Brown et al. (2001), Zvan et al. (2000), Figlewski and
Gao (1999), or Carr et al. (1998).

Exercises
5.1 Consider a contract that pays off

RTα = [XY (T )]α · YT .

(a) Compute the price of this contract in a geometric Brownian motion model
from the stochastic representation

uY (t, x) = EYt [[XY (T )]α |XY (t) = x].

(b) Compute the price of this contract in the same model by solving the
partial differential equation

uYt (t, x) + 12 σ 2 x2 uYxx (t, x) = 0

with the terminal condition uY (T, x) = xα .

Hint: Find the solution of the form uY (t, x) = h(t) · xα .

(c) Determine the hedge of the contract: ∆X (t) and ∆Y (t).

168 Stochastic Finance: A Numeraire Approach

(d) Use Ito’s formula to show

dRYα (t) = ασRYα (t)dW Y (t).

5.2 Show that the price of an Arrow–Debreu security U that pays off a unit
of an asset X on the event
A = { min (e−r(T −t) XY (t)) ≤ L; XY (T ) ≥ K}
0≤t≤T

has the same price up to time τL as an Arrow–Debreu security U (α) that pays
2r 2r
off L1+ σ2 units of a power option R(− σ2 ) on the event
L2
B = {XY (T ) ≤ K }.

5.3 Determine
P(α) (XY (T ) ≥ K),
where P(α) is a martingale measure associated with a power option Rα .
Check that your result agrees with the special cases P(0) (XY (T ) ≥ K) =
PY (XY (T ) ≥ K) = N (d− ), and P(1) (XY (T ) ≥ K) = PX (XY (T ) ≥ K) =
N (d+ ).

5.4 Find the price and the hedge of the contract V with the payoff
VT = (XY (T ) − K)+ · XT .
Hint: This is like a standard European call option, only settled in a “wrong”
asset X instead of Y . Note that XY (T )·XT = [XY (T )]2 ·YT , which represents
a power option R2 . Thus
VT = (RT2 − K · XT )+ .
You can apply Black–Scholes formula using the assets R2 and X.

5.5 Consider a perpetual barrier option that pays off a unit of Y when XY
hits a level L > XY (0) in a geometric Brownian motion model. Let
τ = inf{t ≥ 0 : XY (t) = L}
be the first time that the price hits the barrier. Obviously, the price of the
perpetual barrier option is given by
EY [I(τ < ∞)] = PY (τ < ∞).
Determine PY (τ < ∞).
Hint: Define an auxiliary stopping time τǫ as
τǫ = inf{t ≥ 0 : XY (t) = L or XY (t) = ǫ}
for 0 < ǫ < XY (0).
 Barrier Options 169

(a) Use the Optional Sampling Theorem (Theorem A.1: EY [XY (τǫ )] =
XY (0)) to determine PY (XY (τǫ ) = L).
(b) Compute
PY (τ < ∞) = lim PY (XY (τǫ ) = L).
ǫ→0

You may easily generalize the formula to determine PYt (τ < ∞) when
XY (t) < L.
(c) Determine the hedging portfolio of this contract. Show that your hedging
portfolio delivers a unit of Y when the barrier is hit.
Chapter 6
Lookback Options

This chapter studies lookback options. The payoff of a lookback option de-
pends on either the maximum price of XY (t) or the minimum price of XY (t).
We describe the contract on the maximum price. In order to collect the pay-
off that depends on the maximum, all the intermediate price levels between
zero and the maximum have to be reached during the lifetime of the option,
and thus one can think of the lookback option as a combination of knock-in
option contracts. Once a particular level K is reached, the price may end up
either above or below this level. The first case corresponds to a plain vanilla
European option; the second case corresponds to a knock-in barrier option.
Using the results from the previous sections, the second contract can also be
expressed as a plain vanilla European option. Thus we obtain a representa-
tion of the lookback option price in terms of two plain vanilla European call
options. We also give partial differential equations that correspond to the pric-
ing of the lookback option together with the characterization of the hedging
portfolio. The last section introduces the maximum drawdown, a widely used
portfolio performance measure.

6.1 Connections of Lookbacks with Barrier Options

A lookback option is a contract whose payoff depends on the maximal
or the minimal price of XY (t). Let us illustrate how the maximal asset M ∗
defined as
 
∗
Mt = max XY (s) · Yt
0≤s≤t

is related to barrier contracts. The maximal asset is also known as a high

watermark in hedge fund management. Note that M ∗ is an arbitrage asset,
but a contract V that agrees to deliver this asset at time T is a no-arbitrage
asset. The most typical contracts monitor the maximum of the price in terms
of a dollar, so let us choose Y to be a dollar $, and X to be a stock S.

171
172 Stochastic Finance: A Numeraire Approach

Note that the contract V that pays off

   
∗
VT = MT = max S$ (t) · $T = max S$ (t) · BTT
0≤t≤T 0≤t≤T

can be also expressed as

Z ∞ 
VT = I(τL ≤ T )dL · BTT . (6.1)
0

The hitting time τL is defined as the first time the price of the stock S$ reaches
level L
τL = inf{t ≥ 0 : S$ (t) ≥ L}.

The price level L is reached from below rather than from above in contrast to
the situations studied in the previous sections, but all the analysis remains the
same. The insight of the relationship (6.1) is the following. When the maximal
price is equal to m, i.e., M$∗ (T ) = m, the price process S$ (t) must cross all
levels L ≤ m by time T , while the higher levels are not reached by time T .

A contract that pays off I(τL ≤ T ) units of B T can be split in the following
way

I(τL ≤ T ) · BTT = I(τL ≤ T, S$ (T ) ≥ L) · BTT

+ I(τL ≤ T, S$ (T ) < L) · BTT . (6.2)

Since the event S$ (T ) ≥ L already implies that τL ≤ T , we have

I(τL ≤ T, S$ (T ) ≥ L) · BTT = I(S$ (T ) ≥ L) · BTT .

The second contract with a payoff I(τL ≤ T, S$ (T ) < L) · BTT is a knock-in

Arrow–Debreu security with a barrier equal to L in terms of a dollar price,
and a payoff on the event S$ (T ) < L, so the strike is equal to the barrier.
We have seen that this security has its plain vanilla counterpart that has the
same price up to time τL . Let us determine the corresponding plain vanilla
contract. First, we need to express the barrier in terms of a no-arbitrage asset
B T in the following way

S$ (t) ≥ L ⇐⇒ SB T (t) ≥ Ler(T −t) ,

assuming B$T (t) = e−r(T −t) . Thus the problem corresponds to hitting an ex-
ponential barrier

Ler(T −t) . The knock-in contract on the barrier has the same
1 α
price as L units of a plain vanilla Arrow–Debreu security that pays off a
power option Rα at time T when S$ (T ) ≥ L for α = 1 − σ2r2 .
 Lookback Options 173

Let M$∗ (t) = m be the current running maximum of the stock price S$ (t).
Based on the previous arguments, the payoff of VT is equivalent to
R ∞ 
VT = m · BTT + m
I(S$ (T ) ≥ L)dL · BTT
R ∞
α 
+ m L1 I(S$ (T ) ≥ L)dL · RTα . (6.3)

Since
R ∞  hR i
S (T )∨m
m
I(S$ (T ) ≥ L)dL · BTT = m$ dL · BTT
   
= (S$ (T ) − m)+ · BTT = (SB T − m)+ · BTT = (ST − mBTT )+ , (6.4)
R ∞ 
the contract with a payoff m I(S$ (T ) ≥ L)dL · BTT is a plain vanilla call
option with strike m
(ST − m · BTT )+ .
use the
notation x ∨ y = max(x, y). The form of the contract that pays
We R
∞ α
off m L1 I(S$ (T ) ≥ L)dL · ST depends on the value of α.

6.1.1 Case α = 1
When α = 1, or equivalently when r = 0, we get
R ∞  hR i
1 S (T )∨m 1
m L
· I(S$ (T ) ≥ L)dL · ST = m$ L dL · ST
  +     
S$ (T ) SB T (T ) +
= log m · ST = log m · ST . (6.5)

Therefore the the lookback option V has the same value as a combination of
two plain vanilla options.
  + 
SB T (T )
VT = m · BTT + (ST − m · BTT )+ + log m · ST . (6.6)

Let
uT (t, x, y) = ET [VB T (T )|SB T (t) = x, MB∗ T (t) = y]
be the price of the lookback option V with respect to the bond B T . In the
case when r = 0, it also agrees with the dollar price. Furthermore, MB∗ T (t) =
M$∗ (t) = m, so we also have y = m. The price of the first contract with the
payoff (ST − y · BTT )+ is simply given by the Black–Scholes formula; the price
of the second contract follows from a straightforward computation, and it is
a subject of Exercise 6.1. Thus we get
√
uT (t, x, y) = y + xN (d+ ) − yN (d− ) + xσ T − t [d+ N (d+ ) + φ(d+ )] ,
(6.7)
174 Stochastic Finance: A Numeraire Approach

where √
d± = √1 · log( xy ) ± 21 σ T − t,
σ T −t
x2
and where φ(x) = √12π · e− 2 is a density of a standard normal random
variable. The hedging portfolio is given by
√
∆S (t) = uTx (t, x, y) = 2N (d+ ) + σ T − t [d+ N (d+ ) + φ(d+ )] , (6.8)
∆T (t) = uTy (t, x, y) = yN (−d− ) − xN (d+ ). (6.9)

6.1.2 Case α < 1

2r
When α = 1 − σ2 < 1, which is the case when r > 0, we have
hR
i hR i
∞ α S (T )∨m 1
α
m
1
L · I(S$ (T ) ≥ L)dL · RTα = m$ L dL · RTα
 
1  1−α

1−α +
= · [S$ (T )] −m · RTα
1−α
 
1  1−α

1−α +
= · [SB T (T )] −m · RTα
1−α
1  +
= · ST − m1−α RTα
1−α
 +
σ2 2r (1− 2r )
= · ST − m σ2 RT σ2 . (6.10)
2r
The second-to-the last relationship follows from
[SB T (T )]1−α · RTα = [SB T (T )]1−α · [SB T (T )]α · BTT = SB T (T ) · BTT = ST .
Thus the lookback option V has the same price as a combination of two plain
vanilla options, one expressed in terms of the assets S and B T that pays off
(ST − m · BTT )+ ,
and one expressed in terms of the assets S and Rα that pays off
 +
1 1−α α + σ2 2r (1− σ2r2 )
1−α [ST − m RT ] = 2r · ST − m σ RT .
2

Thus we can write

 +
σ2 2r (1− 2r )
VT = m · BTT + (ST − m · BTT )+ + 2r · ST − m σ2 RT σ2 . (6.11)

In order to express the price of the lookback option V in terms of the prices
with respect to the bond B T , we must substitute MB∗ T (t) for M$∗ (t) = m that
appears in the above formula. Let MB∗ T (t) = y. Then from

MB∗ T (t) = M$∗ (t) · $B T (t) = mer(T −t) = y,

 Lookback Options 175

we have that m = ye−r(T −t) , and thus we can also write

VT = ye−r(T −t) · BTT + (ST − ye−r(T −t) · BTT )+

   2r (1− 2r ) +
σ2 −r(T −t) σ2
+ 2r · ST − ye RT σ2 . (6.12)

Let uT (t, x, y) be the price of V in terms of B T defined as

uT (t, x, y) = ET [VB T (T )|SB T (t) = x, MB∗ T (t) = y].

We can easily obtain the closed form solution using the fact that the price of
V agrees with the sum of the prices of two plain vanilla options. Recall that
the price of a plain vanilla option with a payoff (XT − K · YT )+ is given by
the Black–Scholes formula

PX Y
t (XY (T ) ≥ K) · Xt − KPt (XY (T ) ≥ K) · Yt ,

which also represents the hedging portfolio for this option. Thus the price of
the first option with a payoff (ST − ye−r(T −t) · BTT )+ is simply
   
PSt SB T (T ) ≥ ye−r(T −t) · SB T (t) − yer(T −t) · PTt SB T (T ) ≥ ye−r(T −t)
= xN (d+ ) − ye−r(T −t) N (d− ) (6.13)

according to the Black–Scholes formula, where

  √
x
d± = σ√1T −t log ye−r(T −t) ± 12 σ T − t.
 

2r (1− 2r ) +
The price of the second option with a payoff ST − ye−r(T −t) σ2 RT σ2
is also given by the Black–Scholes formula
   2r2 
PSt SRα (T ) ≥ ye−r(T −t)
σ
· SB T (t)
  2r2    2r2  (1− 2r )
(α)
− ye−r(T −t) SRα (T ) ≥ ye−r(T −t)
σ σ
· Pt · RB T σ2 (t).

But since

T
SRα (T ) = SB T (T ) · BR α (T )

2r
= SB T (T ) · [SB T (T )]−α = [SB T (T )]1−α = [SB T (T )] σ2 ,

2r
the event SRα (T ) ≥ ye−r(T −t) σ2 is equivalent to the event SB T (T ) ≥
ye−r(T −t) . Furthermore, the price of the power option Rα in terms of the
176 Stochastic Finance: A Numeraire Approach

bond B T is given by
α 1 2 α
RB T (t) = exp( α(α − 1)σ (T − t)) · [SB T (t)]
2
 α  1− 2r2
σ
= xe−r(T −t) = xe−r(T −t) .

Thus the price of the second option can be expressed as

h i
PSt SB T (T ) ≥ ye−r(T −t) · SB T (t)
  2r2 h i (1− 2r )
(α)
− ye−r(T −t) SB T (T ) ≥ ye−r(T −t) · RB T σ2 (t)
σ
· Pt
  2r2  1− 2r2  
σ
= xN (d+ ) − ye−r(T −t) · xe−r(T −t)
σ
· N d 2r  , (6.14)
1− σ2

where h i
(α)
Pt SB T (T ) ≥ ye−r(T −t) = N (dα )
for   √
dα = √1 log x
+ (α − 21 )σ T − t.
σ T −t ye−r(T −t)

See Exercise 5.3 for more details on how to determine P(α) (XY (T ) ≥ K).
Combining (6.13) and (6.14), we conclude that
uT (t, x, y) = ye−r(T −t) + xN (d+ ) − ye−r(T −t) N (d− )
  1−α  α 
1
+ · xN (d+ ) − ye−r(T −t) · xe−r(T −t) · N (dα )
1−α
2
= ye−r(T −t) + xN (d+ ) − ye−r(T −t) N (d− ) + σ2r · xN (d+ )
  2r2  1− 2r2  
2 σ σ
− σ2r · ye−r(T −t) · xe−r(T −t) · N d 2r 
1− σ2

−r(T −t) σ2
= ye N (−d− ) + (1 + 2r )xN (d+ )
 
2
2r  −r(T −t) 
y σ2
− σ2r · x · xe 
· N d 1− 2r  . (6.15)
σ2

We have shown the following result.

THEOREM 6.1
The price of a contract V that delivers the maximal asset M ∗ at time T defined
as
VB T (t) = uT (t, x, y) = ET [MB∗ T (T )|SB T (t) = x, MB∗ T (t) = y]
is given by
2
uT (t, x, y) = ye−r(T −t) N (−d− ) + (1 + σ2r )xN (d+ )
 
σ2

2r  −r(T −t) 
y σ2
− 2r · x · xe 
· N d 1− 2r  . (6.16)
σ2
 Lookback Options 177

This corresponds to the terminal condition

f T (x, y) = y.

REMARK 6.1 The payoff of the lookback option V written as

VT = ye−r(T −t) · BTT + (ST − ye−r(T −t) · BTT )+

   2r2 (1− 2r ) +
2
+ σ2r · ST − ye−r(T −t)
σ
RT σ2

can be expressed in a more compact form by observing that both options

expire in the money on the same event SB T (T ) ≥ ye−r(T −t) , and thus the two
option payoffs can be combined into one
VT = ye−r(T −t) · BTT (6.17)
   2r2 (1− 2r ) +
2
σ2
+ (1 + σ2r )ST − ye−r(T −t) · BTT − ye−r(T −t)
σ
2r RT σ2 .

Furthermore, since R(0) = B T , R(1) = S, we can also write

(0)
VT = ye−r(T −t) · RT (6.18)
   2r2 +
2 (1) (0) σ2 (1− 2r )
+ (1 + σ2r )RT − ye−r(T −t) · RT − ye−r(T −t)
σ
2r RT σ2 .

REMARK 6.2 We can also express the price of the contract V that
delivers the maximal asset M ∗ at time T in terms of a dollar $ from the
change of numeraire formula:
V$ (t) · $t = VB T (t) · B T .
If we denote v $ (t, S$ (t), M$∗ (t)) = V$ (t), we can obtain the relationship to the
price function uT (t, SB T (t), MB∗ T (t)) = VB T (t), which is given by

v $ (t, x, y) = e−r(T −t) · uT (t, er(T −t) x, er(T −t) y). (6.19)
This formula generalizes Equation (3.59). Note that for the particular payoff
we have been considering, f T (x, y) = y, the function uT (t, x, y) is homoge-
neous in the variables x and y, meaning that
uT (t, ax, ay) = a · uT (t, x, y). (6.20)
If the prices of SB T (t) and MB∗ T (t) are multiplied by a factor of a, the whole
contract becomes a times more expensive. This is also seen directly from
Equation (6.16). Therefore we have
v $ (t, x, y) = e−r(T −t) · uT (t, er(T −t) x, er(T −t) y) = uT (t, x, y)
178 Stochastic Finance: A Numeraire Approach

for the case of the contract that delivers the maximal asset at time T . This is
true only for special payoff functions f T (x, y), not in general.

6.1.3 Hedging
We have already determined the hedging portfolio for a contract that agrees
to deliver the maximal asset M ∗ in the case when r = 0, which corresponds
to the case of α = 1. Let us show the hedging portfolio for the case of α < 1.
Recall that the contract that delivers the maximal asset corresponds to a
combination of two European options

VT = ye−r(T −t) · BTT + (ST − ye−r(T −t) · BTT )+

   2r2 (1− 2r ) +
2
+ σ2r · ST − ye−r(T −t)
σ
RT σ2 .

The hedging portfolio for the option with the payoff (ST − ye−r(T −t) · BTT )+
is given by N (d+ ) units of the stock S and −ye−r(T −t)N (d− ) units
of the bond B T . The hedging portfolio for the option with the payoff
 
σ2

2r (1− σ2r2 ) + 2

2r ST − ye
−r(T −t) σ2
RT is given by σ2r N (d+ ) units of the stock
 
σ2

2r 2r
S and − 2r ye −r(T −t) σ2
N d 
2r
 units of the power option R(1− σ2 ) .
1− σ2
(1− σ2r2 )
The power option R itself has the hedging portfolio given in Equa-
h r(T −t) i 2r2 −1
σ
tion (5.23), which corresponds to the (1 − σ2r2 ) eS T (t) 1
SB T (t) units of
B

h r(T −t) i σ2r2 −1

the stock S and σ2r2 eS T (t) units of the bond B T . Combining these
B
expressions, we get the following result.

THEOREM 6.2
The hedging portfolio for the contract V that delivers the maximal asset M ∗
at time T is given by
 
S σ2 σ2

2r −r(T −t)
y σ2
∆ (t) = (1 + 2r ) · N (d+ ) + (1 − 2r ) · x ·e ·N d
2r
 , (6.21)
1− σ2

 
T −r(T −t) y

2r2  −r(T −t) 
∆ (t) = ye N (−d− ) − σ · xe 
· N d 1− 2r  , (6.22)
x
σ2

where x = SB T (t) = er(T −t) S$ (t), and y = MB∗ T (t) = er(T −t) M$∗ (t).

REMARK 6.3 Hedging positions and the hitting times

The price of the contract V that delivers a maximal asset M ∗ at time T can
 Lookback Options 179

be written as

uT (t, x, y) = ET [MB∗ T (T )|SB T (t) = x, MB∗ T (t) = y].

We have seen that the function uT (t, x, y) is homogeneous in the variables x

and y, meaning that

uT (t, ax, ay) = a · uT (t, x, y).

Homogeneous functions satisfy the following relationship

uT (t, x, y) = xuTx (t, x, y) + yuTy (t, x, y),

which follows from differentiating (6.20) with respect to a. Substituting the

prices of SB T (t) and MB∗ T (t) for the variables x and y, we get

uT (t, SB T (t), MB∗ T (t)) = SB T (t)uTx (t, SB T (t), MB∗ T (t))

+ MB∗ T (t)uTy (t, SB T (t), MB∗ T (t)).

Note that we also have

uT (t, SB T (t), MB∗ T (t)) = ∆S (t)SB T (t) + ∆T (t)

from the hedging representation of the contract V . Therefore we must have

∆T (t) = MB∗ T (t) · uTy (t, SB T (t), MB∗ T (t)).

Let us determine uTy (t, SB T (t), MB∗ T (t)). Recall that we can write
Z ∞
∗ ∗
M$ (T ) = M$ (t) + I(τL ≤ T )dL,
M$∗ (t)

so the future dollar price of the maximal asset consists of two components:
the present dollar price of the maximal asset plus all the price levels that will
be crossed between time t and T . We can also rewrite the above relationship
in the terms of the bond prices as
Z ∞
∗ −r(T −t) ∗
MB T (T ) = e MB T (t) + I(τL ≤ T )dL.
e−r(T −t) M ∗ T (t)
B

Thus we have

uT (t, SB T (t), MB∗ T (t)) = ETt [MB∗ T (T )]

" Z #
∞
= ETt e −r(T −t)
MB∗ T (t) + I(τL ≤ T )dL
e−r(T −t) M ∗ T (t)
B
Z ∞
= e−r(T −t) MB∗ T (t) + PTt (τL ≤ T )dL.
e−r(T −t) M ∗ T (t)
B
180 Stochastic Finance: A Numeraire Approach

Taking the derivative with respect to the variable y = MB∗ T (t), we get
 
uTy (t, SB T (t), MB∗ T (t)) = e−r(T −t) · 1 − PTt (τM$∗ (t) ≤ T )
= e−r(T −t) · PTt (τM$∗ (t) > T ) = e−r(T −t) · PTt (M$∗ (T ) = M$∗ (t)).

Therefore

∆T (t) = e−r(T −t) MB∗ T (t) · PTt (M$∗ (T ) = M$∗ (t)). (6.23)

Thus the hedging position in the asset Y is between zero and the current
level of M$∗ (t). The event M$∗ (T ) = M$∗ (t) means that the current level of
M$∗ (t) will remain the same until the expiration of the contract. In particular,
when S$ (t) = M$∗ (t), which means that the price of S$ (t) is at the historical
maximum, PTt (M$∗ (T ) = M$∗ (t)) = 0 as the price of S$ (t) will fluctuate to
a higher level for sure. Let us determine the hedging position ∆S (t). The
hedging portfolio can be expressed as

Vt = ∆S (t) · St + ∆T (t) · BtT ,

and thus we have

∆S (t) = VS (t) − ∆T (t) · BST (t)

= VS (t) − e−r(T −t) MB∗ T (t) · PTt (M$∗ (T ) = M$∗ (t)) · BST (t).

We conclude that

∆S (t) = VS (t) − e−r(T −t) MS∗ (t) · PTt (M$∗ (T ) = M$∗ (t)). (6.24)

6.2 Partial Differential Equation Approach for Look-

backs
The previous section showed the relationship of the maximal asset M ∗
defined as  
Mt∗ = max S$ (s) · $t (6.25)
0≤s≤t

to European options. However, the payoff that depends on the asset M ∗ can be
more complicated. In general, a lookback option is a contract that depends
on underlying assets S and $ and upon the maximum of the price process S$ (t).
The asset M ∗ itself is an arbitrage asset. It can still be used as a numeraire,
 Lookback Options 181

but there is no probability measure that would have M ∗ as a reference asset.

More formally, we can define:

DEFINITION 6.1 A lookback option is a contract that pays off one of

the following:


• f T SB T (T ), MB∗ T (T ) units of an asset B T ,


• f S BST (T ), MS∗ (T ) units of an asset S,


T
• f ∗ SM ∗ (T ), BM ∗ (T ) units of an asset M ∗ .

We can substitute a dollar $ with the corresponding bond B T in the payoff

of the lookback option. Should the different settlements represent the same
contract, we must have

f T (x, y) = f S ( x1 , xy ) · x, f S (x, y) = f T ( x1 , xy ) · x,

f T (x, y) = f ∗ ( xy , y1 ) · y, f ∗ (x, y) = f T ( xy , y1 ) · y,

f S (x, y) = f ∗ ( 1y , xy ) · y, f ∗ (x, y) = f S ( xy , x1 ) · x.
This is a generalization of the perspective mapping for three assets. Let us
show for instance the relationship of the two payoff functions f S and f ∗ . In
order for them to represent the same contract, we must have



T
f S BST (T ), MS∗ (T ) · S = f ∗ SM ∗ (T ), BM ∗
∗ (T ) · M ,

or in other words



T
f S BST (T ), MS∗ (T ) = f ∗ SM ∗ (T ), BM ∗
∗ (T ) · MS (T ).

Therefore the x variable in f S stands for BST (T ), and the y variable stands for
MS∗ (T ). It is easy to see that SM ∗ (T ) is now y1 , and BM
T T
∗ (T ) = BS (T )·SM ∗ (T )
x
is represented by y .

Example 6.1
Consider a lookback option contract with a payoff

MT∗ − ST . (6.26)

This is known as a drawdown. One can think of the drawdown as a difference

of two assets, indicating how far the asset X is from the maximal asset M ∗ .
Expressing the contract in terms of the payoff functions, we get

f T (x, y) = y − x, f S (x, y) = y − 1, f ∗ (x, y) = 1 − x.

Note that MT∗ − ST ≥ 0.

182 Stochastic Finance: A Numeraire Approach

Let V be the lookback option. We can write

V = VB T (t) · B T = VS (t) · S = VM∗ (t) · M ∗ .

In the Markovian case, we can also write

V = uT (t, SB T (t), MB∗ T (t)) · B T = uS t, BST (t), MS∗ (t) · S


T
= u∗ t, SM ∗ (t), BM ∗
∗ (t) · M ,

giving us the following relationships between uT , uS , and u∗ in terms of a

perspective mapping:

uT (t, x, y) = uS (t, x1 , xy ) · x, uS (t, x, y) = uT (t, x1 , xy ) · x, (6.27)

uT (t, x, y) = u∗ (t, xy , y1 ) · y, u∗ (t, x, y) = uT (t, xy , y1 ) · y, (6.28)

uS (t, x, y) = u∗ (t, y1 , xy ) · y, u∗ (t, x, y) = uS (t, xy , x1 ) · x. (6.29)

In this situation, we have only two no-arbitrage assets: S and B T . The asset
M ∗ is an arbitrage asset. The hedging of the lookback option must be done
in no-arbitrage assets S and B T only, and no position can be taken in the
asset M ∗ . The First Fundamental Theorem of Asset Pricing gives just two
possible martingale measures that can be used for pricing: PT and PS . The
corresponding price functions uT and uS take the following forms:

uT (t, x, y) = ET [VB T (T )|SB T (t) = x, MB∗ T (t) = y]

 
= ET f T (SB T (T ), MB∗ T (T )) |SB T (T ) = x, MB∗ T (t) = y , (6.30)

 
uS (t, x, y) = ES VS (T )|BST (t) = x, MS∗ (t) = y


= ES f S BST (T ), MS∗ (T ) |BST (t) = x, MS∗ (t) = y . (6.31)

The price function u∗ does not have a stochastic representation with respect
to P∗ since no such measure exists. But we can still compute the price of
the contract with respect to M ∗ chosen as a numeraire using the perspective
mapping:
u∗ (t, x, y) = uT (t, xy , y1 ) · y = uS (t, xy , x1 ) · x. (6.32)

Let us again assume a geometric Brownian motion model with

dSB T (t) = σSB T (t)dW T (t),

and
dBST (t) = σBST (t)dW S (t).
 Lookback Options 183

The evolution of the price of the maximum asset M ∗ is given by

dMB∗ T (t) = d M$∗ (t) · $B T (t)
= $B T (t) · dM$∗ (t) + M$∗ (t) · d$B T (t)
 
= $B T (t) · d max S$ (t) − rM$∗ (t)$B T dt
0≤s≤t
 
= $B T (t) · d max S$ (t) − rMB∗ T (t)dt,
0≤s≤t

or by
 
dMS∗ (t) = d MB∗ T (t) · BST (t)
= MB∗ T (t) · dBST (t) + BST (t) · dMB∗ T (t)
= MB∗ T (t) · σBST (t)dW S (t)
   
T ∗
+BS (t) · $B T (t) · d max S$ (t) − rMB T (t)dt
0≤s≤t
 
∗
= σMS (t)dW (t) + $S (t) · d max S$ (t) − rMS∗ (t)dt.
S
0≤s≤t

THEOREM 6.3
The price function
 
uT (t, x, y) = ET f T (SB T (T ), MB∗ T (T )) |SB T (t) = x, MB∗ T (t) = y ,

satisfies the partial differential equation

uTt (t, x, y) − ryuTy (t, x, y) + 21 σ 2 x2 uTxx (t, x, y) = 0, (6.33)

with the boundary condition

uTy (t, x, x) = 0, (6.34)

and the terminal condition

uT (T, x, y) = f T (x, y). (6.35)

The price function



uS (t, x, y) = ES f S BST (T ), MS∗ (T ) |BST (t) = x, MS∗ (t) = y ,

satisfies the partial differential equation

uSt (t, x, y) − ryuSy (t, x, y)

+ 21 σ 2 x2 uSxx (t, x, y) + 2xyuSxy (t, x, y) + y 2 uSyy (t, x, y) = 0, (6.36)
184 Stochastic Finance: A Numeraire Approach

for y ≥ 1 with the boundary condition

uSy (t, x, 1) = 0, (6.37)

and the terminal condition

uS (T, x, y) = f S (x, y). (6.38)

The price function

u∗ (t, x, y) = uT (t, xy , y1 ) · y = uS (t, xy , x1 ) · x.

satisfies the partial differential equation

− ru∗ (t, x, y) + u∗t (t, x, y) + rxu∗x (t, x, y)

+ ryu∗y (t, x, y) + 21 σ 2 x2 u∗xx (t, x, y) = 0, (6.39)

for x ≤ 1, with the boundary condition

u∗ (t, 1, y) − u∗x (t, 1, y) − x1 u∗y (t, 1, y) = 0, (6.40)

and the terminal condition

u∗ (T, x, y) = f ∗ (x, y). (6.41)

PROOF The process uT (t, SB T (t), MB∗ T (t)) is a PT martingale. From Ito’s
formula, we have

duT (t, SB T (t), MB∗ T (t)) = uTt (t, SB T (t), MB∗ T (t))dt
+ uTx (t, SB T (t), MB∗ T (t))dSB T (t) + uTy (t, SB T (t), MB∗ T (t))dMB∗ T (t)
h
+ 21 uTxx (t, SB T (t), MB∗ T (t))(dSB T (t))2 = uTt (t, SB T (t), MB∗ T (t))
i
− rMB∗ T (t)uTy (t, SB T (t), MB∗ T (t)) + 12 σ 2 (SB T (t))2 uTxx (t, SB T (t), MB∗ T (t)) dt
+ uTx (t, SB T (t), MB∗ T (t))dSB T (t)
 
+ $B T (t)uTy (t, SB T (t), MB∗ T (t)) · d max S$ (t) .
0≤s≤t

The martingale part in the above evolution corresponds to

uTx (t, SB T (t), MB∗ T (t))dSB T (t),

which also gives us the hedging position in the stock S

∆S (t) = uTx (t, SB T (t), MB∗ T (t)).

 Lookback Options 185

The dt term must also vanish, giving us the partial differential equation

uTt (t, x, y) − ryuTy (t, x, y) + 12 σ 2 x2 uTxx (t, x, y) = 0.

The part that corresponds to the term

 
$B T (t)uTy (t, SB T (t), MB∗ T (t)) · d max S$ (t)
0≤s≤t

must also disappear, as it does not correspond to a martingale, and it also

does not correspond to a dt term. The d [max0≤s≤t S$ (t)] term is zero with
probability one as the probability that the stock price is at its maximum is
zero, PT (S$ (t) = M$∗ (t)) = 0. However, d [max0≤s≤t S$ (t)] is not zero when
the stock price is at its maximum. In order to have
 
T ∗
$B T (t)uy (t, SB T (t), MB T (t)) · d max S$ (t) = 0
0≤s≤t

at all times, uTy (t, SB T (t), MB∗ T (t)) must be zero when S$ (t) = M$∗ (t), or
equivalently, when SB T (t) = MB∗ T (t). This implies the boundary condition

uTy (t, x, x) = 0.

The partial differential equation for uS follows from the relationship

uT (t, x, y) = uS (t, x1 , xy ) · x,

and similarly, the partial differential equation for u∗ follows from the relation-
ship
uT (t, x, y) = u∗ (t, xy , y1 ) · y.

THEOREM 6.4
The hedging portfolio Pt of the lookback option satisfies each of the following
equivalent relationships:
h i
Pt = uTx (t, SB T (t), MB∗ T (t)) · S (6.42)
h i
+ uT (t, SB T (t), MB∗ T (t)) − SB T (t) · uTx (t, SB T (t), MB∗ T (t)) · B T ,

h


Pt = uS t, BST (t), MS∗ (t) − BST (t) · uSx t, BST (t), MS∗ (t)

i
−MS∗ (t) · uSy t, BST (t), MS∗ (t) · S (6.43)
h

i
+ uSx t, BST (t), MS∗ (t) + MB∗ T (t) · uSy t, BST (t), MS∗ (t) · B T ,
186 Stochastic Finance: A Numeraire Approach

h
i
T
Pt = u∗x t, SM ∗ (t), BM ∗ (t) ·S
h

+ MB∗ T (t) · u∗ t, SM ∗ (t), BMT
∗ (t) (6.44)

i
−SB T (t) · u∗x T
t, SM ∗ (t), BM ∗ (t) · BT .

PROOF The hedging position ∆S (t) in terms of the price function uT

was already determined in the proof of Theorem 6.3. The hedging position
∆T (t) satisfies

Pt = uT (t, SB T (t), MB∗ T (t)) · B T = ∆S (t) · S + ∆T (t) · B T ,

and thus
∆T (t) = uT (t, SB T (t), MB∗ T (t)) − ∆S (t) · SB T (t).
The hedging position in terms of the function uS follows from the relationship

uT (t, x, y) = uS (t, x1 , xy ) · x,

and the hedging position in terms of the function u∗ follows from the rela-
tionship
uT (t, x, y) = u∗ (t, xy , y1 ) · y.

Note that hedging is done only in the assets S and B T , and no position is
taken in the arbitrage asset M ∗ .

REMARK 6.4 Reduction of the pricing equations

When the payoff of the lookback option depends only on the assets M ∗ and
S, it is possible to obtain simpler pricing equations. This is, for instance, the
case of a drawdown with a payoff MT∗ − ST . We have seen that it corresponds
to the payoff functions

f T (x, y) = y − x, f S (x, y) = y − 1, f ∗ (x, y) = 1 − x.

The payoff is a function of two variables when B T is taken as a reference

asset, but only one variable when S or M ∗ is taken as a reference asset. This
makes sense, when we use B T as a reference asset; we must consider both
prices SB T and MB∗ T . But the asset B T does not enter the contract directly,
and thus both S and M ∗ are more natural numeraires for this problem. When
the payoff does not depend on B T , the pricing problem does not depend on
the price SB T , and thus the partial differential equations for uS and u∗ reduce
 Lookback Options 187

by one dimension that represents this price. The price function uS does not
depend on the variable x, and thus we get

uSt (t, y) − ryuSy (t, y) + 12 σ 2 y 2 uSyy (t, y) = 0, (6.45)

for y ≥ 1 with the boundary condition

uSy (t, 1) = 0, (6.46)

and the terminal condition

uS (T, y) = f S (y). (6.47)

The representation of the hedging portfolio simplifies to

h i
Pt = uS (t, MS∗ (t)) − MS∗ (t) · uSy (t, MS∗ (t)) · S
h i
+ MB∗ T (t) · uSy (t, MS∗ (t)) · B T . (6.48)

The price function u∗ does not depend on the variable y, and thus we get

−ru∗ (t, x) + u∗t (t, x) + rxu∗x (t, x) + 12 σ 2 x2 u∗xx (t, y) = 0, (6.49)

for x ≤ 1, with the boundary condition

u∗ (t, 1) − u∗x (t, 1) = 0, (6.50)

and the terminal condition

u∗ (T, x) = f ∗ (x). (6.51)

The hedging portfolio is of the form

h i
Pt = u∗x (t, SM ∗ (t)) · S
h i
+ MB∗ T (t) · u∗ (t, SM ∗ (t)) − SB T (t) · u∗x (t, SM ∗ (t)) · B T . (6.52)

6.3 Maximum Drawdown

One of the popular portfolio performance measures is the maximum draw-
down. It is related to the drawdown asset defined as the difference between
the maximal asset M ∗ and the stock S. A large value of the price of the draw-
down M$∗ (t) − S$ (t) indicates that the price of the asset is far from its running
188 Stochastic Finance: A Numeraire Approach

maximum, which is negatively regarded if the stock is a part of an investment

portfolio. Drawdown also plays a role in the compensation of hedge fund port-
folio managers which is based in part on the performance of the price of the
portfolio. A typical hedge fund charges 20% of the returns above the so-called
high watermark, which is another name for the price of the maximal asset.
Let us illustrate this compensation scheme on the following example. If the
initial price of the fund is 100 million, and at the end of the first year the price
of the fund is 110 million, the return is 10 million, and the compensation of
the hedge fund manager is 2 million. If the price of the fund drops back to
100 million at the end of the year 2, the return will be negative, and the per-
formance compensation will be zero for that year. If the price of the fund will
increase to 115 million at the end of the third year, the return for the year will
be 15 million, but the compensation is computed only from the return above
the price of the maximal asset, which is only 5 million, the difference between
115 million and the previous maximum at 110 million. Thus the performance
fee of a portfolio manager is conditional on the situation that the drawdown
is zero, and the value of the portfolio reaches a new maximum.

For both investors and hedge fund managers it is important to have the
value of the drawdown as small as possible. Since this is not possible to achieve
at all times, one can keep the track of the historical value of the largest
drawdown. More formally, consider two basic assets: Y which plays the role
of the reference asset in the economy, and X which represents the hedge fund
portfolio. In particular, we can consider X to be a single primary asset S
representing a stock or a stock index. The maximal asset M ∗ is defined as
 
Mt∗ = max XY (s) · Yt .
0≤s≤t

The drawdown is defined as an asset

MT∗ − XT ,

not as a price. For the price, we have three natural numeraires to consider:
Y , X, and M ∗ . The maximal drawdown can be defined in three ways:
 
DTY = max (MY∗ (t) − XY (t)) · YT , (6.53)
0≤t≤T

which is known as an absolute maximum drawdown, or

 
X ∗
DT = max (MX (t) − 1) · XT , (6.54)
0≤t≤T

or  
∗
DTM = max (1 − XM ∗ (t)) · MT∗ , (6.55)
0≤t≤T
 Lookback Options 189

which is known as a relative maximum drawdown. For hedge fund managers,

the maximum drawdown indicates the largest distance they have been from
being able to collect the performance fee. In fact, when the price of the max-
imum drawdown is large, it may lead to a liquidation of the fund itself.

∗
Since DY , DX and DM can be regarded as assets, one can find prices
of the contracts that depend on them from the First Fundamental Theorem
of Asset Pricing. For instance, we can consider a contract V to deliver the
maximum drawdown DY at time T . The price of this contract is given by

VY (t) = EY [DYY (T )]. (6.56)

In general, we can define a contract that depends on the maximum drawdown

as a contract that pays off f Y (XY (T ), MY∗ (T ), DYY (T )) units of the asset Y , or
equivalently using the remaining three assets X, M ∗ DY as alternative refer-
ence assets. The price of such a contract now depends on three price processes,
XY (T ), MY∗ (T ), DYY (T )), and thus the most general form of the correspond-
ing partial differential equation has three spatial variables. For special payoff
functions, such as for the case of a contract to deliver the maximum drawdown
f Y (x, y, z) = z, it is possible to reduce the dimensionality of the problem to
two spatial dimensions, and solve the pricing problem numerically.

References and Further Reading

The price of the lookback option was first given in Goldman et al. (1979),
and later studied, for instance, by Hobson (1998), Buchen and Konstandatos
(2005) or Eberlein and Papapantoleon (2005). The connection between the
hedging positions of lookback options and hitting times was presented in
Pospisil and Vecer (2010). Maximum drawdown serves as an important per-
formance measure for hedge or investment funds. Magdon-Ismail and Atiya
(2004) and Magdon-Ismail et al. (2004) found the distribution of the maximum
drawdown of the Brownian motion. The maximum drawdown as an asset was
introduced in Vecer (2006) and the corresponding partial differential equation
was solved numerically in Pospisil and Vecer (2008).
190 Stochastic Finance: A Numeraire Approach

Exercises
6.1 Find the price and the hedge of the contract V with the payoff
"  + #
XY (T )
VT = log · XT .
K

6.2 Consider a contract V to deliver the maximal asset M ∗ at time T in the

situation when r = 0. We have seen that its uY price is given by Equation
(6.7).
(a) Determine the uS price of the contract V .
Hint: Use perspective mapping from (6.27). Note that the price function
uS depends only on one spatial variable y.
(b) Show that the uS function satisfies the partial differential equation (6.45)
in the reduced form (r = 0). Verify the boundary conditions, and deter-
mine the hedge using the function uS . Check that the hedging portfolio
agrees with the previously obtained representation in (6.8) and (6.9).
(c) Do the same analysis for the price u∗ of the contract V .

6.3 Determine the probability

PTt (τM$∗ (t) > T ) = PTt (M$∗ (T ) = M$∗ (t))

that appears in Equation (6.23) and verify that it leads to the same represen-
tation of the hedge ∆T (t) as in (6.22).
Chapter 7
American Options

American options have the same payoff as their European option counterparts
but an American option can be settled at any time τ before the maturity
time T . American options pay off either f Y (XY (τ )) units of an asset Y , or
f X (YX (τ )) units of an asset X at the exercise time τ ∈ [0, T ]. The exercise
time is chosen by the holder of the option. When these two payoffs correspond
to the same contract, they are related by the perspective mapping



f Y (x) = f X x1 · x, or f X (x) = f Y x1 · x. (7.1)

We will distinguish two cases: when the underlying assets are no-arbitrage
assets, and when one of the underlying assets is an arbitrage asset.

The holder of the American option has at each moment two choices: either
exercise the option immediately and obtain f Y (XY (τ )) units of an asset Y (or
f X (YX (τ )) units of an asset X), or keep the option and exercise it at some
later time. The holder has to compare the value of the immediate exercise
(known as the intrinsic value of the option) with the continuation value. The
option should be exercised the first time its value coincides with its intrinsic
value. This is a situation when the continuation value of the option is not
larger than the intrinsic value.

The first section shows that when an American option with a convex payoff
function is written on two no-arbitrage assets, the optimal exercise strategy is
to wait until the maturity time T , and never exercise the option early. If the
option holder wants to liquidate the contract earlier, he should sell it rather
than exercise it earlier. However, typical American option contracts have at
least one arbitrage asset as an underlying asset, so we focus our attention on
this case.

When the contract is a call or a put, we can get relatively tight bounds for
its price in terms of the price of its European counterpart. The payoff of the
American call or put option cannot be expressed as two Arrow–Debreu secu-
rities and priced separately since the optimal exercise time also needs to be
considered. The problem is that the resulting Arrow–Debreu securities must
be exercised at exactly the same moment, and thus we would not gain any
computational advantage by splitting the payoff. When we compare the prices

191
192 Stochastic Finance: A Numeraire Approach

of the American put option, it is within the price of its European put option
counterpart plus a term that corresponds to the time value of the strike price.
The holder of the American put option has to consider two competing effects:
holding it for a longer time and improving the option value because of the
convexity of the payoff, but the option holder is on the receiving side of the
arbitrage asset (currency) that is deteriorating in time. When the loss of the
value of money dominates the convexity factor of the option, it is better to
exercise the contract. On the other hand, the two factors (convexity and time
value of money) play both in favor to the holder of the American call option,
and thus it is never optimal to exercise this option early. The price is the same
as its European call option counterpart.

Therefore the most interesting case is the American put option written on
the stock and the currency. In general, there is no known analytical solution,
and the pricing problem has to be solved numerically. The reason why there
is not too much hope for analytical methods is that American put option is
technically a contract on three assets: the stock, the currency, and the money
market which appears implicitly in hedging the option. Moreover, the Amer-
ican put option has a barrier feature; the option should be exercised as soon
as the dollar price of the stock reaches a certain region. One can think of
the American put option as a barrier option on three assets (stock, currency,
money market), where the barrier is determined by the optimal actions of the
option holder.

A closed form solution exists in the case of the perpetual American put
option. The reason is that the barrier which determines the exercise region in
terms of the dollar price should not depend on time, and thus it must be a
constant. This greatly simplifies the pricing problem. We apply power assets
to compute the perpetual option price and its hedging portfolio. When the
option has a finite maturity time T , we derive the partial differential equation
that corresponds to the pricing problem.

7.1 American Options on No-Arbitrage Assets

Let us first consider the situation when both assets X and Y are no-
arbitrage assets. It turns out that when the payoff functions f Y and f X
are convex, the option should never be exercised early (τ ∗ = T ), as the
following result suggests. Convexity of the payoff function f Y implies that
the payoff function f X is convex, and vice versa. The perspective mapping
f Y (x) = f X ( x1 ) · x preserves convexity.
 American Options 193

THEOREM 7.1
The optimal exercise time τ ∗ of the American option written on two no-
arbitrage assets X and Y that pays off either f Y (XY (τ )) units of an asset Y ,
or f X (YX (τ )) units of an asset X at the exercise time τ ∈ [0, T ], where f Y
and f X are convex functions, is given by

τ ∗ = T.

PROOF This is a result that follows from Jensen’s inequality that states

E[f (X)] ≥ f (E[X])

for a random variable X and a convex function f . For any time t, the value
of the European option V with the same payoff function as its American
counterpart can be written as

VY (t) = EYt [f (XY (T ))] .

But according to Jensen’s inequality, we also have

EYt [f Y (XY (T ))] ≥ f Y EYt [XY (T )] = f Y (XY (t))

Thus we conclude that at any time t ≤ T , the value of the European option
Vt dominates the intrinsic value of the American option f Y (XY (t)), and thus
it is never optimal to exercise the American option early for convex payoff
functions f Y and f X .

REMARK 7.1 Convexity of the payoff functions f Y and f X is necessary

for Theorem 7.1 to hold. Consider for instance the payoff function f Y (x) =
I(x ≥ K), so that the holder of the option can receive an asset Y if the price
XY (t) is greater or equal to K. The indicator function is not convex, and
clearly the option should be exercised the first time the price reaches K, even
when both underlying assets are no-arbitrage assets. The corresponding payoff
function f X is given by f X (x) = f Y ( x1 ) · x = K
1 1
· I(x = K ).

Theorem 7.1 implies that American puts and calls should never be exercised
early when both underlying assets are no-arbitrage assets. Call options have
a payoff function f (x) = (x − K)+ , put options have a payoff function f (x) =
(K −x)+ , and both payoff functions are convex. In particular, an American put
option with a payoff (K ·Yτ − Xτ )+ should never be exercised before maturity.
This may be slightly surprising because an American put option written on
a stock S and the dollar $ with a payoff (K · $τ − Sτ )+ can be optimally
exercised before the maturity of the contract, but that does not contradict
our result since currencies are arbitrage assets. However, if the payoff of the
American put option written on the stock and the dollar is slightly modified,
194 Stochastic Finance: A Numeraire Approach

when the dollar is replaced with a no-arbitrage asset B T so that the payoff
becomes (K · BτT − Sτ )+ , such an option should not be exercised early. We will
use this fact to obtain tight bounds on the price of American options written
on one arbitrage asset.

7.2 American Call and Puts on Arbitrage Assets

The most traded types of American options involve typically one arbitrage
asset represented by the dollar $. For instance, an American put option writ-
ten on a stock S and the dollar $ pays off (K · $τ − Sτ )+ at the exercise time
τ. Since the hedging of the contract must be done in no-arbitrage assets, we
also need to include a no-arbitrage proxy asset for the dollar $, such as a bond
B T that matures at the expiration time of the American option T .

The following theorem shows that an American call option written on a

stock S and on a dollar $ should be exercised at maturity, and thus the price
of the American call option coincides with the corresponding European call
option. The price of the American put option written on a stock S and on $
is constrained by relatively tight bounds. We denote by V AC , V AP , V EC , and
V EP the American call option, American put option, European call option,
and European put option, respectively. The first two arguments represent
the assets that settle the contract; the third argument is the maturity of the
contract. Note that V EP (S, K · B T , T ) = V EP (S, K · $, T ) since the arbitrage
asset $ can be replaced by a corresponding contract to deliver B T .

THEOREM 7.2
The price of an American put option written on a stock S and on the dollar
$ is bounded by

VtEP (S, K · B T , T ) ≤ VtAP (S, K · $, T ) ≤ VtEP (S, K · B T , T ) + ($t − BtT ) · K.

(7.2)
The price of an American call option written on a stock S and on the dollar
$ is identical to its European call option counterpart

VtAC (S, K · $, T ) = VtEC (S, K · B T , T ). (7.3)

PROOF The left hand side of the inequality for the American put option
is trivial. Let us show the right hand side of the inequality. Since

(K · $t − St )+ ≤ (K · BtT − St )+ + ($t − BtT ) · K,

 American Options 195

we also have that

VtAP (S, K · $, T ) ≤ VtAP (S, K · B T , T ) + ($t − BtT ) · K

= VtEP (S, K · B T , T ) + ($t − BtT ) · K.

Similarly for the American call option we have

VtEC (S, K · B T , T ) ≤ VtAC (S, K · $, T ),

so the price of the American call option dominates the price of the European
call option. On the other hand we have

(St − K · $t )+ ≤ (St − K · BtT )+ + (BtT − $t )+ · K = (St − K · BtT )+ .

Therefore
VtAC (S, K · $, T ) ≤ VtEC (S, K · B T , T ),
and thus the price of the American call option written on the stock S and on
the dollar $ coincides with the price of its European call option counterpart.

COROLLARY 7.1
When BtT ≡ 1 · $t (or in other words when r = 0), we have that

VtAP (S, K · $, T ) = VtEP (S, K · B T , T ),

and thus it is not optimal to exercise the American put option early in the
situation when the interest rate is zero.

7.3 Perpetual American Put

Consider a contract V with a payoff

Vτ = (K · $τ − Sτ )+ = (K − S$ (τ ))+ · $τ , (7.4)

where τ is a stopping time with no upper bound. This contract is known as

a perpetual American put since it has no expiration. Obviously, the opti-
mal exercise strategy should be independent of time; the holder should decide
when to exercise based only on the price level of S$ . Time is not an issue;
there is always an infinite amount of time left.

Since the perpetual American option has an infinite time horizon, the exer-
cise strategy is independent of time. The only strategy that is independent of
196 Stochastic Finance: A Numeraire Approach

American and European put option

0.5

0.45

0.4

0.35

0.3
Option price

0.25

0.2

0.15

0.1

0.05

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

FIGURE 7.1: The price V$EP (0) of a European put option with a payoff
($T − K · ST )+ (bottom curve), the price V$AP (0) of the corresponding Amer-
ican put option counterpart (middle curve), and the price of the European
put option V$EP (0) plus (1 − B$T (0)) · K (top curve), together with the option
payoff (1 − K · S$ )+ as a function of the stock price S$ . The parameters are
r = 0.02, K = 21 , σ = 0.2, T = 10.
 American Options 197

time is a stationary strategy. Let us first compute the price of the perpetual
American put that is associated with the exercise strategy

τL = inf{t ≥ 0 : S$ (t) ≤ L} (7.5)

for a given value of L < K. Next we will find the value of L∗ that maximizes
the price of the perpetual American option among different choices of the
exercise levels L.

In order to compute the price of the perpetual American put using the First
Fundamental Theorem of Asset Pricing, we need to express the the contract
in terms of no arbitrage assets. Since the contract has an infinite time horizon,
there is no bond with an infinite maturity that can be used as a no-arbitrage
proxy to a dollar $, and thus we should use a money market M instead. The
relationship between dollars and the money market is

Mt = ert · $t .

In particular, the stopping time τL can be written as

τL = inf{t ≥ 0 : SM (t) ≤ L · e−rt },

and the payoff of the option can be expressed as

(K − S$ (τL ))+ · $τ = (K − L)+ · e−rτL · Mτ .

The price can be computed from the Optional Sampling Theorem (Theorem
A.1) as

VM (0) = EM [(K − L) · e−rτL ] = (K − L) · EM [e−rτL ].

The Optional Sampling Theorem states that under certain conditions, the
martingale keeps its expected value even when it is stopped at a stopping
time τ . However, computation of EM [e−rτL ] requires one to determine the
distribution of the hitting time τL , which is a nontrivial task. A more elegant
solution uses properties of a power asset. The power asset is analogous to
the power option that was studied in the previous chapters. The fact that the
perpetual option has an infinite horizon prevents us from defining the power
option as a contract with a particular payoff at some finite time T . Instead,
one should define the initial value of a power asset as
h iα
α SM (0)
RM (0) = L , (7.6)

and assume that

α α
dRM (t) = ασRM (t)dW M (t), (7.7)
198 Stochastic Finance: A Numeraire Approach

which is the same evolution as (5.13). The price of the perpetual American
put can be computed by finding α such that the assets Rα and L1 units of S
agree on the exercise boundary, in which case

α SM (τL )
RM (τL ) = = e−rτL ,
L
and the value of the perpetual American put would follow from
h iα
EM [e−rτL ] = EM [RM
α α
(τL )] = RM (0) = SML(0) .

α
Let us find the corresponding α. Since RM (t) has a geometric Brownian
motion evolution, it can be written as
α α
RM (t) = RM (0) · exp(ασW M (t) − 12 α2 σ 2 t)
h iα
= SML(0) · exp(ασW M (t) − 12 α2 σ 2 t)
h iα
= SML(t) · exp(− 21 α(α − 1)σ 2 t). (7.8)

The last equality follows from

h iα h iα
SM (t)
L = SML(0) · exp(ασW M (t) − 21 ασ 2 t).

α SM (τL )
Should RM (τL ) = L , we must have
h iα
α SM (τL ) SM (τL )
RM (τL ) = L · exp(− 12 α(α − 1)σ 2 τL ) = L . (7.9)

SM (τL )
Since L = e−rτL , the above equality is satisfied when

exp((α − 1)(−r − 21 ασ 2 )τL ) = 1, (7.10)

which is true either when α = 1, or when α = − σ2r2 . Only the second solution
represents a suitable power asset with the price given by
2r
− 2
h i− 2r2
SM (t) σ 2r
RMσ (t) = L · exp(−r(1 + σ2 )t). (7.11)

Let SM (t) ≥ Le−rt . The value of the perpetual American put with the exercise
boundary L at time t is given by
 2r 
M −rτL M − 2
VM (t) = (K − L) · E [e ] = (K − L) · E RMσ (τL )|SM (t)
h i− 2r2
SM (t) σ 2r
= (K − L) · L · exp(−r(1 + σ2 )t), (7.12)
 American Options 199

which corresponds to the price function uM (t, x). Thus

2r  − 2r2
x − σ2 ert x σ
uM (t, x) = (K − L) · L · exp(−r(1 + σ2r2 )t) = (K − L) · e−rt · . L
(7.13)
This is valid for x ≥ Le−rt . When x ≤ Le−rt, the perpetual American option
should be exercised immediately, collecting K units of a dollar $t , and shorting
a unit of a stock St . Thus Vt = Ke−rt · Mt − St , or equivalently,

VM (t) = e−rt K − SM (t).

In terms of the price function uM , we have

uM (t, x) = e−rt K − x. (7.14)

The price of the perpetual American put in dollar terms is given by

h i− 2r2
S$ (t) σ
V$ (t) = VM (t) · M$ (t) = (K − L) · L . (7.15)

If we define the price function v $ as v $ (t, S$ (t)) = V$ (t), the prices of v $ and
uM are related by
v $ (t, x) = ert · uM (t, e−rt x). (7.16)
Thus

2r
x − σ2
v $ (t, x) = (K − L) · L (7.17)
when x ≥ L. When x ≤ L, we have

v $ (t, x) = K − x. (7.18)

The optimal choice of L∗ maximizes the function

2r
x − σ2
h(L) = uM (0, x) = v $ (t, x) = (K − L) · L .

It is straightforward to show that the function h(L) is maximized for the value
2r
L∗ = · K, (7.19)
2r + σ 2
which can be seen from h′ (L∗ ) = 0, and h′′ (L∗ ) < 0.

7.4 Partial Differential Equation Approach

This section studies the American stock option in a geometric Brownian
motion model that pays off either f $ (S$ (τ )) units of the dollar $, or f S ($S (τ ))
200 Stochastic Finance: A Numeraire Approach

units of a stock S at the exercise time τ ∈ [0, T ]. Let us assume that the
dynamics of the bond price with respect to the dollar are given by

dB$T (t) = rB$T (t)dt, (7.20)

implying the dynamics of the stock price with respect to the dollar are

dS$ (t) = d[SB T (t) · B$T (t)]

= SB T (t) · dB$T (t) + B$T (t) · dSB T (t)
= rS$ (t)dt + σS$ (t)dW T (t).

We can price the American stock option contract with respect to the dollar
$, or with respect to the stock S. When the dollar $ is chosen as a reference
asset, we can define the dollar price of the American option contract as:
h i
v $ (t, x) = max ETt e−r(τ −t)V$ (τ )|S$ (t) = x .
τ

Then e−rt v $ (t, S$ (t)) is a PT martingale. Using Ito’s formula, we get

 
d e−rt v $ (t, S$ (t)) = −re−rt v $ (t, S$ (t)) dt + e−rt dv $ (t, S$ (t))
h
= −re−rt v $ (t, S$ (t)) dt + e−rt vt$ (t, S$ (t)) dt
i
$
+vx$ (t, S$ (t)) dS$ (t) + 12 vxx (t, S$ (t)) d2 S$ (t)
h
= e−rt −rv $ (t, S$ (t)) + vt$ (t, S$ (t))
i
+rS$ (t)vx$ (t, S$ (t)) + 21 σ 2 S$ (t)2 v$xx (t, S$ (t)) dt
+σS$ (t)vx$ (t, S$ (t)) dW T (t).

Since the dt term must be zero, we get the following partial differential equa-
tion:
−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx
$
(t, x) = 0.
This partial differential equation is valid as long as the option should not
be exercised, which is true when the option value exceeds its intrinsic value
v $ (t, x) > f $ (x). When the option should be exercised, we have that the
option value coincides with its intrinsic value:

u$ (t, x) = f $ (x).

In case that the option is not exercised when it should be, the contract starts
to lose the value and the dt term becomes negative. In this situation, the
holder of the American option is creating an arbitrage opportunity at his own
expense for the benefit of the seller of the option. The First Fundamental
Theorem of Asset Pricing states that when there is no arbitrage, the prices
 American Options 201

Exercise boundary
0.55

0.5

0.45
Holding region
0.4
S Price
$

0.35

0.3

0.25
Exercise region
0.2
0 10 20 30 40 50 60 70 80 90 100
Time

FIGURE 7.2: Optimal exercise boundary for the American put option with
a payoff (K · $τ − Sτ )+ with parameters r = 0.02, K = 21 , σ = 0.2, T = 100
as a function of time.

are martingales under the probability measure that is associated with the
reference asset. However, the holder of the American option may fail to take
an optimal action that leads to the martingale price of the contract. In this
case, the price of the contract becomes a supermartingale, which is a process
that is nonincreasing in expectation:

Es [S(t)] ≤ S(s). (7.21)

Thus we have

−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx

$
(t, x) < 0

in the region where the option should be exercised. We conclude that the
American option is characterized by linear complementarity conditions:

−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 21 σ 2 x2 vxx

$
(t, x) ≤ 0, (7.22)

and
v $ (t, x) ≥ f $ (x). (7.23)
The above inequalities do not lead to a closed form solution, and the price of
the contract has to be computed numerically.

Figure 7.2 shows the exercise boundary for the American put option with
parameters r = 0.02, K = 21 , σ = 0.2. The price of the option above
202 Stochastic Finance: A Numeraire Approach

FIGURE 7.3: The dollar price of the American put option with a payoff
(K · $τ − Sτ )+ with parameters r = 0.02, K = 12 , σ = 0.2 as a function of the
stock price S$ and time to maturity T − t.

the boundary satisfies the Black-Scholes partial differential equation for the
function v $ , and the value of the option dominates its intrinsic value. The
option should be exercised if the stock price is below the exercise bound-
ary, where the option price agrees with its intrinsic value, and the term
−rv $ (t, x) + vt$ (t, x) + rxvx$ (t, x) + 21 σ 2 x2 vxx
$
(t, x) is smaller than zero. Note
that the exercise boundary flattens for large time to maturity, and the limit-
2r 0.04 1
ing level corresponds to L∗ = 2r+σ 2 · K = 0.04+0.04 · 2 = 0.25, the exercise

boundary for the perpetual American put.

Figure 7.3 shows the dollar price of the American put option as a func-
tion of the price and time to maturity. For large times to maturity, the
price of the contract approaches the price of the perpetual American put
option, which is given by v $ (t, x) = K − x for x ≤ L∗ = 0.25, and

−1
v $ (t, x) = (K − L∗ ) Lx∗ 1
= 16x for x ≥ 0.25.

Figures 7.4 and 7.5 show the hedging positions in the assets S and M . The
hedging positions are similar to the hedging positions of its European option
counterpart with the exception that the hedging position in the stock should
be set to −1 in the exercise region. When the contract should be exercised,
K · $t − St = K · Mt − St should be collected. Thus the hedging becomes trivial
(short the stock, and long K units of the money market). The exercise region
is clearly visible in both graphs.
 American Options 203

FIGURE 7.4: The hedging position in the stock S for the American put
option contract with a payoff (K ·$τ − Sτ )+ with parameters r = 0.02, K = 21 ,
σ = 0.2 as a function of the stock price S$ and time to maturity T − t.

FIGURE 7.5: The hedging position in the money market M for the Ameri-
can put option contract with a payoff (K ·$τ −Sτ )+ with parameters r = 0.02,
K = 21 , σ = 0.2 as a function of the stock price S$ and time to maturity T − t.
204 Stochastic Finance: A Numeraire Approach

REMARK 7.2 Stock as a reference asset

When the stock S is chosen as a reference asset, we can define the price of
the American option with respect to the stock as
v S (t, x) = max ESt [VS (τ )|$S (t) = x] .
τ

S S
We have that v (t, $S (t)) is a P martingale. From Ito’s formula we have
dv S (t, $S (t)) = vtS (t, $S (t)) dt + +vxS (t, $S (t)) d$S (t)
S
+ 1 vxx (t, $S (t)) d2 $S (t)
h 2
= vtS (t, S$ (t)) − r$S (t)vxS (t, $S (t))
i
+ 12 σ 2 $S (t)2 vxx
S
(t, $S (t)) dt
+σ$S (t)vxS (t, $S (t)) dW S (t).
We have used the dynamics of the price of the dollar with respect to the stock
d$S (t) = d[BST (t) · $B T (t)]
= BST (t) · d$B T (t) + $B T (t) · dBST (t)


= $S (t) −rdt + σ · dW S (t) .
Since the price of the contract with respect to the stock is a martingale, the
dt term must be zero and thus we obtain
vtS (t, x) − rxvxS (t, x) + 12 σ 2 x2 vxx
S
(t, x) = 0.
This is true when the option value exceeds its intrinsic value v S (t, x) > f S (x)
and the option should not be exercised. This partial differential equation has
a slight numerical advantage over the classical Black-Scholes since it contains
only 3 terms. In the region where the option should be exercised, the option
value coincides with its intrinsic value
v S (t, x) = f S (x),
but keeping the American contract would lead to a loss, and the corresponding
dt term would become negative. Thus we have
vtS (t, x) − rxvxS (t, x) + 12 σ 2 x2 vxx
S
(t, x) ≤ 0
when the option should be exercised. We conclude that the option price with
respect to the reference asset S satisfies the linear complementarity conditions

vtS (t, x) − rxvxS (t, x) + 12 σ 2 x2 vxx

S
(t, x) ≥ 0, (7.24)

and
v S (t, x) ≥ f S (x). (7.25)
 American Options 205

References and Further Reading

The price of the perpetual American put was first computed by McKean
(1965), who also gave an analytic characterization of the American put price
with a finite expiration. Merton (1973) observed that the price of the Ameri-
can call option agrees with its European option counterpart for a no-dividend-
paying stock, which in our notation means that the asset X is a no-arbitrage
asset. The bounds for the American put option price appeared in Carr et al.
(1992). The probabilistic approach for pricing the American put option is
given in Bensoussan (1984) and in Karatzas (1988). Longstaff and Schwartz
(2001) and Tsitsiklis and Van Roy (2001) suggested the method of the least
squares to estimate the continuation value and the optimal exercise boundary
for the American put option. See also Broadie and Glasserman (1997) and
Rogers (2002). For other references on pricing American options, see for in-
stance, Barone-Adesi and Whaley (1987), Geske and Johnson (1984), Boyle
et al. (1997), Jaillet et al. (1990), Haugh and Kogan (2004) Broadie and De-
temple (1996), Zhu (2006), Carr (1998), Zvan et al. (1998), Clement et al.
(2002), Boyarchenko and Levendorskii (2002), Levendorskii (2004), Pham
(1997), and Zhang (1997).

Exercises
7.1 Show that the price uM (t, x) in (7.13) and (7.14) satisfies the Black–
Scholes partial differential equation
uM 1 2 2 M
t (t, x) + 2 σ x uxx (t, x) = 0

when x ≥ Le−rt . When x ≤ Le−rt , the price uM (t, x) satisfies the partial
differential inequality
uM 1 2 2 M
t (t, x) + 2 σ x uxx (t, x) ≤ 0.

7.2 Determine the hedging portfolio for the perpetual American put option.
Hint: ∆S (t) = uM
x (t, x).

7.3 Show that the price v $ (t, x) in (7.17) and (7.18) satisfies the partial
differential equation
−rv $ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx
$
(t, x) = 0
when x ≥ L. When x ≤ L, the price v $ (t, x) satisfies the partial differential
inequality
−rv $ (t, x) + rxvx$ (t, x) + 12 σ 2 x2 vxx
$
(t, x) ≤ 0.
Chapter 8
Contracts on Three or More Assets:
Quantos, Rainbows and “Friends”

Exotic options are contracts that depend on three or more underlying assets.
We have already studied lookback options that depend on a stock S, a bond
B T (or equivalently a dollar $), and on the maximal asset M ∗ . The maximal
asset itself depends on the stock and the bond, so it is not a free asset that
exists on its own. This chapter studies contracts on three “full” underlying
assets; let us call them X, Y , and Z. There are several variants of these con-
tracts. A quanto is a contract that pays off some function of the price of
XY (T ) units of the third asset Z. The underlying assets for quanto contracts
are usually currencies, but the definition is not limited to them. For example,
a quanto forward pays off XY (T )−K units of Z, and a quanto call option pays
off (XY (T ) − K)+ units of Z. The quanto contracts have only one underlying
price process, but the option is settled in a “wrong” asset. A contract that
depends on three or more assets and two or more price processes is known
as a rainbow option. The price of a rainbow option thus depends on the
joint distribution of two or more price processes, and they are sensitive to the
correlation structure of the prices. Some rainbow option contracts are known
under a specific name, for instance a call option on the maximum of two as-
sets, etc. We call them “the friends of quantos and rainbows.”

An option on three underlying assets can be formally defined as:

DEFINITION 8.1 An option on three underlying assets is a contract

that pays off one of the following:

• f Y (XY (T ), ZY (T )) units of an asset Y ,

• f X (YX (T ), ZX (T )) units of an asset X,

• f Z (XZ (T ), YZ (T )) units of an asset Z.

Example 8.1
We have seen in Example 3.1 that the contract that pays off the best of two
assets is completely symmetric (and dual to itself). The contract on the best

207
208 Stochastic Finance: A Numeraire Approach

of two assets is directly linked to European options via

(XT − K · YT )+ = max(XT , K · YT ) − K · YT ,

and
(K · YT − XT )+ = max(XT , K · YT ) − XT ,
and thus most of the traded contracts are related to it. Therefore an analogous
contract that pays off the best of three assets

max(XT , K1 YT , K2 ZT )

is a natural candidate whose variants would generalize the concept of most

traded European options. Indeed, when we consider, for instance, the differ-
ence of the best of three assets and one of the assets itself, we get

max(XT , K1 YT , K2 ZT ) − K2 ZT = (max(XT , K1 YT ) − K2 ZT )+ , (8.1)

which represents the payoff of the call option on the maximum of two assets.

As we saw earlier for the case of the general European contract, it is possible
to settle the payoff in each of the underlying assets. Should the different
settlements represent the same contract, the payoff functions must be linked
by the perspective mapping

f Y (x, y) = f X ( x1 , xy ) · x, f X (x, y) = f Y ( x1 , xy ) · x,

f Y (x, y) = f Z ( xy , y1 ) · y, f Z (x, y) = f Y ( xy , y1 ) · y,
f X (x, y) = f Z ( 1y , xy ) · y, f Z (x, y) = f X ( xy , x1 ) · x.

Example 8.2
Consider a quanto call option that pays off (XY (T ) − K)+ · Z. In terms of the
reference asset Y , the payoff takes the following form

(XY (T ) − K)+ · ZY (T ) · Y,

representing (XY (T ) − K)+ · ZY (T ) units of the asset Y . Thus the payoff

function f Y is given by

f Y (x, y) = (x − K)+ · y.

Similarly, when the reference asset is chosen to be X, the payoff is of the form

(XY (T ) − K)+ · ZX (T ) · X

which is represented by the payoff function f X given by

f X (x, y) = ( x1 − K)+ · y.
Contracts on Three or More Assets: Quantos, Rainbows and “Friends” 209

Finally, when the reference asset is Z the payoff is simply

(XY (T ) − K)+ · Z

which is represented by the payoff function f Z given by

f Z (x, y) = ( xy − K)+ .

8.1 Pricing in the Geometric Brownian Motion Model

Let V denote a contract on three assets. We can use the change of numeraire
formula to compute the price of V , this time using all three available assets:

V = VY (t) · Y = VX (t) · X = VZ (t) · Z.

In a Markovian pricing model, we can represent the prices VY , VX , and VZ

by functions uY , uX , and uZ :

V = uY (t, XY (t), ZY (t)) · Y

= uX (t, YX (t), ZX (t)) · X = uZ (t, XZ (t), YZ (t)) · Z,

giving us the following relationships known as a perspective mapping between

price functions uY , uX , and uZ :

uY (t, x, y) = uX (t, x1 , xy ) · x, uX (t, x, y) = uY (t, x1 , xy ) · x, (8.2)

uY (t, x, y) = uZ (t, xy , y1 ) · y, uZ (t, x, y) = uY (t, xy , y1 ) · y, (8.3)

uX (t, x, y) = uZ (t, y1 , xy ) · y, uZ (t, x, y) = uX (t, xy , x1 ) · x. (8.4)

When X, Y , and Z are no-arbitrage assets, we also have a stochastic rep-
resentation of the prices:

uY (t, x, y) = EY [VY (T )|XY (t) = x, ZY (t) = y]

 
= EY f Y (XY (T ), ZY (T )) |XY (t) = x, ZY (t) = y , (8.5)

uX (t, x, y) = EX [VX (T )|YX (t) = x, ZX (t) = y]

 
= EX f X (YX (T ), ZX (T )) |YX (t) = x, ZX (t) = y , (8.6)
210 Stochastic Finance: A Numeraire Approach

uZ (t, x, y) = EZ [VZ (T )|XZ (t) = x, YZ (t) = y]

 
= EZ f Z (XZ (T ), YZ (T )) |XZ (t) = x, YZ (t) = y . (8.7)

In order to get a more specific representation of the price of the contract on

three assets, let us assume a diffusion model of the prices of the underlying
assets. There are six possible prices to consider:

dXY (t) = σxy XY (t)dW 1,Y (t), dZY (t) = σyz ZY (t)dW 2,Y (t), (8.8)

dYX (t) = σxy YX (t)dW 1,X (t), dZX (t) = σxz ZX (t)dW 2,X (t), (8.9)

dXZ (t) = σxz XZ (t)dW 1,Z (t), dYZ (t) = σyz YZ (t)dW 2,Z (t). (8.10)
Since there are two price processes for each reference asset, they are driven
by two Brownian motions W 1 and W 2 that can be correlated. Let us assume
for instance
dW 1,Y (t) · dW 2,Y (t) = ρdt.
The value of ρ represents the correlation. The formula for dW 1,Y (t) and
dW 2,Y (t) determines the correlation of W 1,X (t) and W 2,X (t), and the corre-
lation of W 1,Z (t) and W 2,Z (t). For instance, from Ito’s formula we get

dZX (t) = d (ZY (t) · YX (t))

= ZY (t) · dYX (t) + YX (t) · dZY (t) + dZY (t) · dYX (t) (8.11)
 1,X 2,Y
 2
= ZX (t) σxy dW (t) + σyz dW (t) + ZX (t) σxy σyz [−ρ] dt.

Therefore
dYX (t) dZX (t)
dW 1,X (t) · dW 2,X (t) = ·
σxy YX (t) σxz ZX (t)
2
σxy − ρσxy σyz σxy − ρσyz
= = .
σxy · σxz σxz

Similarly we obtain
σyz − ρσxy
dW 1,Z (t) · dW 2,Z (t) = .
σxz

From (8.11) we also get

d2 ZX (t) 2 2
= (σxy − 2ρσxy σyz + σyz )dt,
ZX (t)2

and therefore the volatility σxz of ZX (t) is constrained to satisfy

2 2 2
σxz = σxy − 2ρσxy σyz + σyz .
Contracts on Three or More Assets: Quantos, Rainbows and “Friends” 211

THEOREM 8.1
The price function
h i
uY (t, x, y) = EY f (XY (T ), ZY (T )) |XY (t) = x, ZY (t) = y ,

satisfies partial differential equation

2
uYt (t, x, y) + 21 σxy x2 · uYxx (t, x, y)
2 2
+ ρσxy σyz xy · uYxy (t, x, y) + 12 σyz y · uYyy (t, x, y) = 0, (8.12)

with the terminal condition

uY (T, x, y) = f Y (x, y), (8.13)

the price function

h i
uX (t, x, y) = EX f X (YX (T ), ZX (T )) |YX (t) = x, ZX (t) = y ,

satisfies partial differential equation

uX 1 2 2 X X
t (t, x, y) + 2 σxy x · uxx (t, x, y) + σxy (σxy − ρσyz )xy · uxy (t, x, y)
2 2
+ 21 (σxy − 2ρσxy σyz + σyz )y 2 · uX
yy (t, x, y) = 0, (8.14)

with the terminal condition

uX (T, x, y) = f X (x, y), (8.15)

and the price function

h i
uZ (t, x, y) = EZ f Z (XZ (T ), YZ (T )) |XZ (t) = x, YZ (t) = y

satisfies partial differential equation

uZ 1 2 2 2 Z
t (t, x, y) + 2 (σxy − 2ρσxy σyz + σyz )x · uxx (t, x, y)

+ σyz (σyz − ρσxy )xy · uZ 1 2 2 Z

xy (t, x, y) + 2 σyz y · uyy (t, x, y) = 0, (8.16)

with the terminal condition

uZ (T, x, y) = f Z (x, y). (8.17)

PROOF Let us derive the partial differential equation for uX ; the other
partial differential equations can be proved in a similar fashion, or using the
212 Stochastic Finance: A Numeraire Approach

perspective mapping relationship between functions uY , uX and uZ . The pro-

cess uX (t, YX (T ), ZX (T )) = EX [VX (T )|YX (t), ZX (t)] is a martingale. Using
Ito’s formula, we get

duX (t, YX (t), ZX (t)) = uX X X

t dt + ux dYX (t) + uy dZX (t)

+ 21 uX 2 X 1 X 2
xx d YX (t) + uxy dYX (t)dZX (t) + 2 uyy d ZX (t)
"
= uX 1 2 2 X X
t + 2 σxy x · uxx + σxy (σxy − ρσyz )xy · uxy

#
2 2
+ 21 (σxy − 2ρσxy σyz + σyz )y 2 · uX
yy dt

+uX X
x dYX (t) + uy dZX (t).

The dt term of a martingale has to be zero, giving us the partial differential

equation for uX .

Example 8.3
Consider the quanto forward with a payoff

(XY (T ) − K) · Z

that corresponds to payoff functions f Y (x, y) = (x − K) · y, f X (x, y) = ( x1 −

K) · y, and f Z (x, y) = ( xy − K). It is not difficult to show that the price of the
quanto forward is given by
 
uY (t, x, y) = x · eρσxy σyz (T −t) − K · y

when Y is used as a reference asset. We can immediately obtain the price of

the quanto forward with respect to the reference asset X
 
uX (t, x, y) = uY (t, x1 , xy ) · x = x1 · eρσxy σyz (T −t) − K · y,

and the price of the quanto forward with respect to the reference asset Z
 
uZ (t, x, y) = uY (t, xy , y1 ) · y = xy · eρσxy σyz (T −t) − K .

One can check that the functions uY , uX and uZ satisfy the partial differential
equations in Theorem 8.1 (Exercise 8.1).

Example 8.4
Consider the quanto call option with a payoff
+
(XY (T ) − K) · Z
Contracts on Three or More Assets: Quantos, Rainbows and “Friends” 213

that corresponds to payoff functions f Y (x, y) = (x − K)+ · y, f X (x, y) = ( x1 −

K)+ · y, and f Z (x, y) = ( xy − K)+ . A slightly more complicated computation
shows that

uY (t, x, y) =
 h   i
x·exp(ρσxy σyz (T −t))
= x · eρσxy σyz (T −t) · y · N σ √1T −t log K
2
+ 21 σxy (T − t)
xy
 h   i
1 x·exp(ρσxy σyz (T −t)) 1 2
− K · y · N σ T −t log √
K − 2 σxy (T − t) (8.18)
xy

when Y is used as a reference asset. We can immediately obtain the price of

the quanto call option with respect to the reference asset X

uX (t, x, y) = uY (t, x1 , xy ) · x =
 h   i
exp(ρσxy σyz (T −t))
= x1 · eρσxy σyz (T −t) · y · N σ √1T −t log K·x + 1 2
σ
2 xy (T − t)
xy
 h   i
exp(ρσ σ (T −t)) 1 2
− K · y · N σ √1T −t log xy yz
K·x − 2 σxy (T − t) (8.19)
xy

and the price of the quanto call option with respect to the reference asset Z

uZ (t, x, y) = uY (t, xy , y1 ) · y =
 h   i
x·exp(ρσxy σyz (T −t))
= xy · eρσxy σyz (T −t) · N σ √1T −t log K·y
2
+ 12 σxy (T − t)
xy
 h   i
1 x·exp(ρσxy σyz (T −t)) 1 2
− K · N σ √T −t log K·y − σ
2 xy (T − t) . (8.20)
xy

Functions uY , uX and uZ satisfy the partial differential equations in Theorem

8.1.

8.2 Hedging
Contracts on three assets admit a perfect hedge in the geometric Brownian
motion model driven by two Brownian motions. The hedging is done in all
three underlying assets, and has the form summarized in the next theorem.

THEOREM 8.2
The hedging portfolio P of the contract on three assets is given by each of the
214 Stochastic Finance: A Numeraire Approach

following equivalent relationships:

h i
Pt = uYx (t, XY (t), ZY (t)) · X
h
+ uY (t, XY (t), ZY (t)) − XY (t) · uYx (t, XY (t), ZY (t))
i
−ZY (t) · uYy (t, XY (t), ZY (t)) · Y (8.21)
h i
+ uYy (t, XY (t), ZY (t)) · Z

h
Pt = uX (t, YX (t), ZX (t)) − YX (t) · uX x (t, YX (t), ZX (t))
i
−ZX (t) · uX y (t, YX (t), Z X (t)) ·X
h i
+ uXx (t, YX (t), ZX (t)) · Y (8.22)
h i
+ uXy (t, YX (t), Z X (t)) ·Z

h i
Pt = uZx (t, XZ (t), YZ (t)) · X
h i
+ uZ y (t, XZ (t), YZ (t)) · Y (8.23)
h
+ uZ (t, XZ (t), YZ (t)) − XZ (t) · uZ x (t, XZ (t), YZ (t))
i
−YZ (t) · uZy (t, XZ (t), YZ (t)) · Z.

PROOF The hedging portfolio is given by positions ∆X (t), ∆Y (t) and

∆Z (t) in the underlying assets X, Y and Z. Thus it can be written as

Pt = ∆X (t) · X + ∆Y (t) · Y + ∆Z (t) · Z.

Since the hedging portfolio is self-financing, it also has to satisfy

dPY (t) = ∆X (t) · dXY (t) + ∆Z (t) · dZY (t).

But we also have

dVY (t) = duY (t, XY (t), ZY (t))

= uYx (t, XY (t), ZY (t)) dXY (t) + uy (t, XY (t), ZY (t)) dZY (t)
Contracts on Three or More Assets: Quantos, Rainbows and “Friends” 215

from Ito’s formula after realizing that the dt term is zero. In order to have
Pt = Vt at all times, we must have

∆X (t) = uYx (t, XY (t), ZY (t)) ,

and
∆Z (t) = uYy (t, XY (t), ZY (t)) .
The position in the remaining asset Y is determined from

∆Y (t) = PY (t) − ∆X (t) · XY (t) − ∆Z (t) · ZY (t).

This proves the representation of the hedge in terms of the price function
uY . The remaining expressions are equivalent, which is easy to see from the
relationships

uX (t, x, y) = uY (t, x1 , xy ) · x, uZ (t, x, y) = uY (t, xy , y1 ) · y.

This implies

uYx (t, x, y) = uX (t, x1 , xy ) − 1

x · uX 1 y
x (t, x , x ) −
y
x · uX 1 y Z x 1
y (t, x , x ) = ux (t, y , y ),

1 y
uY (t, x, y) − x · uYx (t, x, y) − y · uYy (t, x, y) = uX Z x 1
x (t, x , x ) = uy (t, y , y ),
1 y
uYy (t, x, y) = uX Z x 1
y (t, x , x ) = u (t, y , y ) −
x
y · uZ x 1
x (t, y , y ) −
1
y · uZ x 1
y (t, y , y ).

Example 8.5
Let us determine the hedge of the quanto forward with payoff

(XY (T ) − K) · Z.

We have seen that the price of this contract with respect to the reference asset
Y is given by  
uY (t, x, y) = x · eρσxy σyz (T −t) − K · y,
and thus
uYx (t, x, y) = eρσxy σyz (T −t) · y,
and
uYy (t, x, y) = x · eρσxy σyz (T −t) − K.
The hedging portfolio is therefore given by

Pt = ∆X (t) · X + ∆Y (t) · Y + ∆Z (t) · Z

h i
= ZY (t) · eρσxy σyz (T −t) · X
h i h i
− ZY (t) · XY (t) · eρσxy σyz (T −t) · Y + XY (t) · eρσxy σyz (T −t) − K · Z.
216 Stochastic Finance: A Numeraire Approach

Example 8.6
Similarly we can determine the hedge of the quanto call option with a payoff

(XY (T ) − K)+ · Z.

The price uY (t, x, y) of this contract is given in Equation (8.18). Thus we get

uYx (t, x, y) =
 h   i
x·exp(ρσxy σyz (T −t))
= eρσxy σyz (T −t) ·y·N σxy
1
√
T −t
log K
2
+ 21 σxy (T − t) ,

and

uYy (t, x, y) =
 h   i
x·exp(ρσxy σyz (T −t))
= x · eρσxy σyz (T −t) · N σ √1T −t log K + 1 2
σ
2 xy (T − t)
xy
 h   i
x·exp(ρσ σ (T −t)) 2
− K · N σ √1T −t log xy
K
yz
− 21 σxy (T − t) .
xy

The hedging portfolio is therefore given by

Pt = ∆X (t) · X + ∆Y (t) · Y + ∆Z (t) · Z


= ZY (t) · eρσxy σyz (T −t) ×
 h   i
1 x·exp(ρσxy σyz (T −t)) 1 2
×N √
σxy T −t
log K + 2 σxy (T − t) ·X

− ZY (t) · XY (t) · eρσxy σyz (T −t) ×
 h   i
1 x·exp(ρσxy σyz (T −t)) 1 2
×N √
σxy T −t
log K + 2 σxy (T − t) ·Y

+ XY (t) · eρσxy σyz (T −t) ×
 h   i
x·exp(ρσxy σyz (T −t)) 1 2
× N σ √1T −t log K + σ
2 xy (T − t)
xy
 h   i
x·exp(ρσxy σyz (T −t)) 1 2
−K · N σ √1T −t log K − σ
2 xy (T − t) · Z.
xy
Contracts on Three or More Assets: Quantos, Rainbows and “Friends” 217

References and Further Reading

Options on multiple assets were first priced in Stulz (1982) and later studied
by Johnson (1987) and by Boyle et al. (1989). These works generalized the
results on the exchange option obtained earlier by Margrabe (1978). Boyle
(1988) explored lattice methods for options on multiple assets. Monte Carlo
methods were employed by Broadie and Detemple (1997) for American op-
tions written on multiple assets. Gerber and Shiu (1996) studied perpetual
American options written on two assets. Cherubini et al. (2004) is a mono-
graph on copula methods in finance, which also covers techniques of pricing
exotic options. An extensive list of existing option pricing formulas that also
covers exotic options appears in Haug (1997).

Exercises
8.1 Show that the price functions in Example 8.3 satisfy the partial differ-
ential equations from Theorem 8.1.

8.2 Find the price and the hedging portfolio for a quanto put option with
payoff (K − XY (T ))+ · Z, where the prices of the underlying assets follow
geometric Brownian motion models given by Equations (8.8), (8.9), and (8.10).

8.3 (a) Find the price and the hedging portfolio of a contract V that pays
off the best asset out of X, K1 Y , and K2 Z, or in other words
VT = max(XT , K1 YT , K2 ZT )
where the prices of the underlying assets follow geometric Brownian mo-
tion models given by Equations (8.8), (8.9), and (8.10). As an immediate
consequence, find the price and the hedge for the call option on the best
of two assets with the payoff
(max(XT , K1 YT ) − K2 ZT )+ .

(b) Find the price and the hedging portfolio of a contract that pays off the
worst asset
min(XT , K1 YT , K2 ZT ).
(c) Find the price and the hedging portfolio of a contract that pays off the
“middle” asset, which can be defined as
XT + K1 YT + K2ZT − max(XT , K1 YT , K2 ZT ) − min(XT , K1 YT , K2 ZT ).
Chapter 9
Asian Options

Asian options are contracts that depend on underlying assets X and Y and
upon the average of the price process XY (t). The average price process is
captured by a no-arbitrage contract A called the average asset. The payoff
of the average asset is defined as
"Z #
T
AT = XY (t)µ(dt) · YT . (9.1)
0

The average asset is a contract that pays off a number of units of an asset Y ,
where the number of units is the weighted average price of an asset X with
respect to the asset Y . The weights are determined by the weighting measure
µ which can represent both continuous or discrete averaging. Our definition
of the average asset guarantees that its price is always positive, and thus the
average asset can be used as a numeraire. The average asset is analogous to
the maximal asset M ∗ that appears in pricing of lookback options. The im-
portant difference is that the average asset A turns out to be a no-arbitrage
asset in contrast to the maximal asset M ∗ .

The average asset is typically not traded, but we can still use it as a nu-
meraire in order to derive the pricing equations for Asian options. The pricing
techniques for Asian options do not require the existence of the average asset
as a traded contract. We will express all hedging positions in terms of assets
X and Y only. Moreover as we will show in the following text, the Asian for-
ward can be perfectly replicated by trading in the underlying assets X and Y ,
and the hedge is model independent. Therefore A itself is a no-arbitrage asset.

We can apply the First Fundamental Theorem of Asset Pricing as long as

the assets X and Y are no-arbitrage assets. This is not the case when X is a
stock S and Y is dollars $, when the average asset contract becomes
"Z #
T
AT = S$ (t)µ(dt) · $T .
0

However, we can still rewrite this contract in terms of no-arbitrage assets

when the bond price follows a deterministic term structure BtT = e−r(T −t) $t

219
220 Stochastic Finance: A Numeraire Approach

as "Z #
T
−r(T −t)
AT = SB T (t)e µ(dt) · BTT ,
0

which is of the form of (9.1), with the underlying two no-arbitrage assets S
and B T . Note that hedging must be done in no-arbitrage assets exclusively as
opposed to arbitrage assets such as currencies. A typical Asian option contract
uses equal weights. A continuously sampled average asset pays off
"Z #
T
1
AT = T S$ (t)dt · $T ,
0

which corresponds to an averaging of the form

1 −r(T −t)
µ(dt) = Te dt,

when expressed in terms of S and B T . A discretely sampled average asset

pays off
n
X
AT = n1 S$ ( nk T ),
k=1

which corresponds to an averaging of the form

n
X
µ(dt) = 1
n δ( k T ) (t)e−r(T −t) dt
n
k=1

when expressed in terms of S and B T .

Let us define the most general form of an Asian option.

DEFINITION 9.1 An Asian option is a contract that pays off one of the
following:
• f Y (XY (T ), AY (T )) units of an asset Y ,
• f X (YX (T ), AX (T )) units of an asset X,
• f A (XA (T ), YA (T )) units of an asset A.

When the payoff functions are linked by the perspective mapping f Y (x, y) =
f X ( x1 , xy )
· x = f A ( xy , y1 ) · y, the three payoffs represent the same contract.

Example 9.1
The Asian call option with a fixed strike pays off

(AT − K · YT )+ . (9.2)
 Asian Options 221

This corresponds to the payoff functions f Y (x, y) = (y − K)+ , f X (x, y) =

(y − K · x)+ , or f A (x, y) = (1 − K · y)+ in the above definition of the Asian
option. This means that the payoff can be settled in three equivalent ways:
+ + +
(AY (T ) − K) · Y = (AX (T ) − K · YX (T )) · X = (1 − K · YA (T )) · A.

The Asian call option with a floating strike pays off

(AT − K · XT )+ , (9.3)

which corresponds to the payoff functions f Y (x, y) = (y − K · x)+ , f X (x, y) =

(y − K)+ , or f A (x, y) = (1 − K ·x)+. The payoff can be settled in the following
three ways:

(AY (T ) − KXY (T ))+ · Y = (AX (T ) − K)+ · X = (1 − K · XA (T ))+ · A.

Asian options with the fixed or the floating strike are the two most typical
Asian option contracts.

It is interesting to note that the prices of the Asian fixed strike and the
Asian floating strike options can be written as a Black–Scholes formula. The
price of the fixed strike option is simply

PA Y
t (AY (T ) ≥ K) − K · Pt (AY (T ) ≥ K), (9.4)

and the price of the floating strike option is

PA X
t (AX (T ) ≥ K) − K · Pt (AX (T ) ≥ K). (9.5)

This follows from the fact that the Asian option can be written as a combina-
tion of two Arrow–Debreu securities whose price is given by the above expres-
sions. However, the hard part is that the prices AY (T ) and AX (T ) do not have
a simple analytical distribution as opposed to the case of XY (T ) which has a
known density, and thus determination of the corresponding probabilities is a
nontrivial task. Semianalytical representations of these probabilities exist for
continuous averaging, but they still require significant computational effort
to obtain any numerical result. In our text we present the partial differential
equations that correspond to the Asian option pricing problem which applies
to both discrete and continuous averaging. These partial differential equations
can be solved numerically in a straightforward way.

The foreign exchange market also trades contracts written on the harmonic
average of the price. The harmonic average is defined as the reciprocal of the
arithmetic average of the reciprocals:
1 1
RT = RT .
1
0 XY (t) µ(dt) 0
YX (t)µ(dt)
222 Stochastic Finance: A Numeraire Approach

If we denote by "Z #
T
ÃT = YX (t)µ(dt) · XT
0

the average asset where the roles of the assets X and Y are flipped, we can
define the harmonic average asset as
" #
1 1
HT = R T · YT = · YT .
0 YX (t)µ(dt)
ÃX (T )

Natural contracts to consider are the harmonic Asian option with a fixed
strike with payoff
 +
1
(HT − K · YT )+ = · YT − K · YT
ÃX (T )
and the harmonic Asian option with a floating strike with payoff
 +
1
(HT − K · XT )+ = · YT − K · XT .
ÃX (T )
We can also write the payoffs in terms of the original average asset AT if we
flip the roles of the assets Y and X (it is just a matter of naming the assets).
In this case the harmonic Asian option with a fixed strike has payoff
 +
1
AY (T ) · XT − K · XT , (9.6)

which corresponds to the payoff functions f Y (x, y) = ( xy − K · x)+ , f X (x, y) =

( xy − K)+ , and f A (x, y) = (x · y − K · x)+ . The harmonic Asian option with
a floating strike has payoff
 +
1
AY (T ) · X T − K · YT , (9.7)

which corresponds to the payoff functions f Y (x, y) = ( xy − K)+ , f X (x, y) =

( xy − K · x)+ , and f A (x, y) = (x · y − K · y)+ .

We can also consider more exotic payoffs, such as Asian powers f X (x, y) =
α
y . The advantage of this contract is that it admits a closed form solution for
integer valued α, and thus it can be used for calibrating numerical schemes.
This payoff corresponds to f Y (x, y) = y α · x1−α , or equivalently to f A (x, y) =
x1−α . We can write the payoff as


A α


A α


X 1−α

1−α
X T ·X = Y T · Y T ·Y = X A T · A.
 Asian Options 223

Let V denote an Asian option contract. The price of this contract can be
expressed in the following ways:

V = VY (t) · Y = VX (t) · X = VA (t) · A.

In the Markovian model, we can also write

Vt = uY (t, XY (t), AY (t)) · Y = uX (t, YX (T ), AX (T )) · X

= uA (t, XA (T ), YA (T )) · A,

giving us the following relationships between uY , uX , and uA via the perspec-

tive mapping:

uY (t, x, y) = uX (t, x1 , xy ) · x, uX (t, x, y) = uY (t, x1 , xy ) · x, (9.8)

uY (t, x, y) = uA (t, xy , y1 ) · y, uA (t, x, y) = uY (t, xy , y1 ) · y, (9.9)

X A 1 x A X y 1
u (t, x, y) = u (t, y, y) · y, u (t, x, y) = u (t, x, x) · x. (9.10)
When X and Y are no-arbitrage assets, then A is a no-arbitrage asset
(shown below), and from the First Fundamental Theorem of the Asset Pricing
we have the following stochastic representations:

uY (t, x, y) = EY [VY (T )|XY (t) = x, AY (t) = y]

 
= EY f Y (XY (T ), AY (T )) |XY (t) = x, AY (t) = y , (9.11)

uX (t, x, y) = EX [VX (T )|YX (t) = x, AX (t) = y]

 
= EX f X (YX (T ), AX (T )) |YX (t) = x, AX (t) = y , (9.12)

uA (t, x, y) = EA [VA (T )|XA (t) = x, YA (t) = y]

 
= EA f A (XA (T ), YA (T )) |XA (t) = x, YA (t) = y . (9.13)

Let us show that the average asset A is indeed a no-arbitrage asset.

THEOREM 9.1
Let X and Y be two no-arbitrage assets. Then the replicating portfolio for the
average asset contract that pays off
"Z #
T
AT = XY (y)µ(dt) · YT (9.14)
0

is given by "Z #
T Z t 
At = µ(ds) · X + XY (s)µ(ds) · Y. (9.15)
t 0
224 Stochastic Finance: A Numeraire Approach

This result does not depend on the dynamics of the price XY (t). In particular,
"Z #
T
dAY (t) = µ(ds) dXY (t). (9.16)
t

PROOF Let At = ∆ ¯ X (t)Xt + ∆

¯ Y (t)Yt be the replicating portfolio of the
average asset. Then
dAY (t) = ∆¯ X (t)dXY (t).

Using the product rule, this can be rewritten as

dAY (t) = ∆ ¯ X (t) · XY (t) − XY (t)d∆
¯ X (t)dXY (t) = d ∆ ¯ X (t).

Integrating this equation, we get

Z T
¯ X (T ) · XY (T ) − ∆
AY (T ) = AY (0) + ∆ ¯ X (0) · XY (0) − ¯ X (t).
XY (t)d∆
0

Since the terminal position of the average asset is completely invested in the
asset Y , and has a zero position in the asset X, we have ∆ ¯ X (T ) = 0. We thus
have the following identity:
Z T ! Z T
XY (t)µ(dt) = AY (0) − ∆ ¯ X (0) · XY (0) − XY (t)d∆ ¯ X (t).
0 0

The only way to match the payoff is when

¯ X (0) · XY (0),
0 = AY (0) − ∆

which is equivalent to
¯ X (0)X0 ,
A0 = ∆
and Z Z
T T
XY (t)µ(dt) = − ¯ X (t).
XY (t)d∆
0 0
This implies
¯ X (t) = µ(dt),
−d∆
which is the same as
Z T Z T
¯ X (t) = −
∆ ¯ X (s) =
d∆ µ(ds).
t t

¯ Y (t) in the asset Y follows from the identity

The hedging position ∆
Z t
AY (t) = ∆¯ X (t)XY (t) + XY (s)µ(ds)
0
 Asian Options 225

which concludes the proof.

REMARK 9.1 ¯ X (t) in the asset X is

Note that the hedging position ∆
deterministic:
Z T
¯ X (t) =
∆ µ(ds). (9.17)
t

1 −r(T −t)
For instance, when µ(dt) = Te dt, we get
Z T Z T  
¯ X (t) =
∆ µ(ds) = 1 −r(T −s)
= 1
1 − e−r(T −t) .
Te ds rT
t t

1
In the case of uniform weighting µ(dt) = T dt, the hedge of the average asset
simplifies to
Z T Z T

¯ X
∆ (t) = µ(ds) = 1
= 1− t
T ds T .
t t
1
Pn −r(T −t)
For discrete averaging when µ(dt) = n k=1 δ( n
k
T ) (t)e dt, we get

Z T Z T n
X
¯ X (t) =
∆ µ(ds) = 1
δ( k T ) (s)e−r(T −s) ds
n n
t t k=1
n
X
1


= n exp −r( n−k
n )T ,
h i
nt
k= T +1

where [·] denotes the integer part function. This simplifies to

 
¯ X (t) = 1 −
∆ 1
n Tt (9.18)
n

1 Pn
when the averaging is uniform, i.e. when µ(dt) = n k=1 δ( n
k
T ) (t)dt.

The insight of this result is the following: the trader who is replicating the
RT
average asset contract starts with a hedging portfolio of 0 µ(dt) units of X
and no units of Y :
Z T !
¯
∆(0) = (∆ ¯ ,∆
X ¯ )=
Y
µ(dt), 0 .
0 0
0

RT
The amount of 0 µ(dt) units of the asset X is used for replicating the average
of the price. The trader then gradually liquidates his position in the asset X,
RT
keeping just t µ(dt) fraction of it at time t, and the rest of the portfolio
226 Stochastic Finance: A Numeraire Approach

is invested in theR asset Y . The position in the asset Y corresponds to the

t
running average 0 XY (s)µ(ds). At the final time T , the hedge becomes
Z T !
¯ ) = (∆
∆(T ¯ (T ), ∆
X ¯ (T )) = 0,
Y
XY (t)µ(dt) ,
0

so the asset X is completely unloaded, and the position in the asset Y is the
number that corresponds to the average price.

9.1 Pricing in the Geometric Brownian Motion Model

The prices of assets should be martingales under their corresponding nu-
meraire measures. Since we have three underlying assets X, Y and A, we have
six price processes to consider: XY (t), AY (t), YX (t), AX (t), XA (t), and YA (t).
The price processes XY (t) and AY (t) are PY martingales, the price processes
YX (t) and AX (t) are PX martingales, and the price processes XA (t) and YA (t)
are PA martingales.

In the geometric Brownian motion model we assume the following price

dynamics:
dXY (t) = σXY (t)dW Y (t), (9.19)
and a similar evolution for the inverse price

dYX (t) = σYX (t)dW X (t). (9.20)

The evolution of AY (t) follows from the hedging formula for the average
asset:
¯ X (t)dXY (t) = σ ∆
dAY (t) = ∆ ¯ X (t)XY (t)dW Y (t). (9.21)
Note that this evolution is not Markovian in AY (t) since it depends on an-
other process XY (t), but it is Markovian in the pair (XY (t), AY (t)). Thus
even when the Asian option contract payoff depends only on AY (t), the cor-
responding pricing partial differential equation would depend on both prices.

The evolution of the average asset price under the reference asset X can be
expressed as
¯ Y (t)dYX (t)
dAX (t) = ∆
 
= AY (t) − ∆¯ X (t) · XY (t) dYX (t)
 
= AY (t) − ∆¯ X (t) · XY (t) σYX (t)dW X (t)
 
= σ AX (t) − ∆ ¯ X (t) dW X (t).
 Asian Options 227

The second equality ∆ ¯ Y (t) = AY (t) − ∆

¯ X (t) · XY (t) follows from the relation-
¯ X ¯ Y
ship At = ∆ (t) · X + ∆ (t) · Y . The reason to write the evolution of AX (t)
in terms of ∆ ¯ X (t) rather than in terms of ∆ ¯ Y (t) is that ∆
¯ X (t) is determin-
¯ Y
istic, while ∆ (t) is stochastic. This means that unlike the price evolution
of AY (t), the price evolution of AX (t) is Markovian in just one variable, and
thus contracts whose payoff depends only on AX (T ) admit a simpler partial
differential equation with one spatial variable. Thus
 
¯ X (t) dW X (t).
dAX (t) = σ AX (t) − ∆ (9.22)

Let us determine the evolution of the remaining prices: YA (t), and XA (t).
From Ito’s formula we have

dYA (t) = dAY (t)−1 = −AY (t)−2 dAY (t) + AY (t)−3 d2 AY (t)
¯ X (t)XY (t)dW Y (t)
= −YA (t)2 σ ∆
+YA (t)3 σ 2 ∆¯ X (t)2 XY (t)2 dt
h i
¯ X (t)YA (t)XA (t) − dW Y (t) + σ ∆
= σ∆ ¯ X (t)XA (t)dt .

According to the First Fundamental Theorem of Asset Pricing, the evolution

of YA (t) has to be a martingale under the corresponding PA measure. Thus
we have
¯ X (t)YA (t)XA (t)dW A (t),
dYA (t) = σ ∆ (9.23)

where W A (t) is a Brownian motion under PA measure. Similarly,

dXA (t) = dAX (t)−1 = −AX (t)−2 dAX (t) + AX (t)−3 d2 AX (t)
 
¯ X (t) dW X (t)
= −XA (t)2 σ AX (t) − ∆
 
+XA (t)3 σ 2 AX (t) − ∆¯ X (t) 2 dt
 X  h   i
¯ (t)XA (t) − 1 · dW X (t) − σ 1 − ∆
= σXA (t) · ∆ ¯ X (t)XA (t) dt .

Therefore
 X 
¯ (t)XA (t) − 1 dW A (t),
dXA (t) = σXA (t) ∆ (9.24)

which is a martingale under the PA measure.

The price of the Asian option is determined in the next theorem.

THEOREM 9.2
The price function
 
uY (t, x, y) = EY f Y (XY (T ), AY (T )) |XY (t) = x, AY (t) = y ,
228 Stochastic Finance: A Numeraire Approach

satisfies partial differential equation

h
uYt (t, x, y) + 21 σ 2 x2 uYxx (t, x, y)
i
¯ X (t)uYxy (t, x, y) + ∆
+ 2∆ ¯ X (t)2 uYyy (t, x, y) = 0 (9.25)

with the terminal condition

uY (T, x, y) = f Y (x, y). (9.26)

The price function

 
uX (t, x, y) = EX f X (YX (T ), AX (T )) |YX (t) = x, AX (t) = y ,
satisfies partial differential equation
h
uX 1 2 2 X
t (t, x, y) + 2 σ x uxx (t, x, y)
i
¯ X (t))uX
+ 2x(y − ∆ xy (t, x, y) + (y − ¯
∆ X
(t))2 X
u yy (t, x, y) = 0, (9.27)

with the terminal condition

uX (T, x, y) = f X (x, y). (9.28)

The price function

 
uA (t, x, y) = EA f A (XA (T ), YA (T )) |XA (t) = x, YA (t) = y

satisfies partial differential equation


uA 1 2 2 ¯X 2 A
t (t, x, y) + 2 σ x [x∆ (t) − 1] · uxx (t, x, y)

¯ X (t)[x∆
+ 2y ∆ ¯ X (t) − 1] · uA (t, x, y) + y 2 (∆
¯ X (t))2 · uA (t, x, y) = 0,
xy yy
(9.29)
with the terminal condition

uA (T, x, y) = f A (x, y). (9.30)

PROOF The price of the Asian option with respect to the reference asset
Y , uY (t, XY (t), AY (t)), is a PY martingale, and thus duY has a zero dt term.
Using Ito’s formula, we get

duY = uYt dt + uYx dXY (t) + uYy dAY (t)

 
+ 21 uYxx d2 XY (t) + 2uYxy dXY (t)dAY (t) + uYyy d2 AY (t)
¯ X (t)uYxy + ∆
= [uYt + 1 σ 2 x2 (uYxx + 2∆ ¯ X (t)2 uYyy )]dt
2
+uYx dXY (t) + uYy dAY (t).
 Asian Options 229

Since the dt term is zero, we obtain the following partial differential equation:
 
¯ X (t)uY (t, x, y) + ∆
uYt (t, x, y)+ 21 σ 2 x2 uYxx (t, x, y) + 2∆ ¯ X (t)2 uY (t, x, y) = 0.
xy yy

The terminal condition is given by

uY (T, x, y) = f Y (x, y).

Similarly, the price of the Asian option with respect to the reference asset X,
uX (t, YX (t), AX (t)), is a PX martingale, and thus the dt term of duX is zero.
Using the evolution of the price of the average asset under the reference asset
X, we get

duX = uX X X
t dt + ux dYX (t) + uy dAX (t)
 
+ 21 uX 2 X X 2
xx d YX (t) + 2uxy dYX (t)dAX (t) + uyy d AX (t)

= [uX 1 2 2 X ¯X X ¯X 2 X
t + σ (x uxx + 2x(y − ∆ (t))uxy + (y − ∆ (t)) uyy )]dt
2
+uX X
x dYX (t) + uy dAX (t).

Since the dt term is zero, we have the following partial differential equation
h
uX
t (t, x, y) + 1 2
2 σ x2 uX
xx (t, x, y)
i
+ 2x(y − ∆¯ X (t))uX (t, x, y) + (y − ∆
¯ X (t))2 uX (t, x, y) = 0,
xy yy

with the terminal condition

uX (T, x, y) = f X (x, y).

Finally, the price of the Asian option with respect to the reference asset A,
uA (t, XA (t), YA (t)), is a PA martingale, and thus the dt term of duA is zero.
Using the evolution of the prices of X and Y under the reference asset A, we
get

duA = uA A A
t dt + ux dXA (t) + uy dYA (t)
 
+ 12 uA 2 A A 2
xx d XA (t) + 2uxy dXA (t)dYA (t) + uyy d YA (t)
h
= uA 1 2 2 ¯X 2
t + 2 σ x [[x∆ (t) − 1] · uxx
A

i
+2y ∆ ¯ X (t)[x∆ ¯ X (t)2 · uA ] dt
¯ X (t) − 1] · uA + y 2 ∆
xy yy

+uA A
x dXA (t) + uy dYA (t).

Since the dt term is zero, we have the following partial differential equation

uA (t, x, y) + 1 2 2
σ x [x∆¯ X (t) − 1]2 · uA (t, x, y)
t 2 xx

+ 2y ∆¯ X (t)[x∆
¯ X (t) − 1] · uA 2 ¯X 2 A
xy (t, x, y) + y ∆ (t) · uyy (t, x, y) = 0,
230 Stochastic Finance: A Numeraire Approach

with the terminal condition

uA (T, x, y) = f A (x, y). (9.31)

9.2 Hedging of Asian Options

Since Asian options depend on three assets: X, Y , and the Asian forward
A, the hedge should take positions in all these assets. The hedging portfolio
should be of the form

Pt = ∆X (t) · X + ∆Y (t) · Y + ∆A (t) · A. (9.32)

However, the average asset itself can be hedged by assets X and Y :

¯ X (t) · X + ∆
At = ∆ ¯ Y (t) · Y, (9.33)

and thus the Asian option hedge can be reduced to positions in just two assets,
X and Y :

¯ X (t)] · X + [∆Y (t) + ∆A (t) · ∆

Pt = [∆X (t) + ∆A (t) · ∆ ¯ Y (t)] · Y. (9.34)

The hedging position in the underlying assets X and Y has two components:
one part (∆X (t) or ∆Y (t)) represents the usual delta sensitivity of the Asian
option price with respect to the price of the underlying asset, and the other
part represents the delta sensitivity of the Asian option price with respect to
the average asset price (∆A (t)), multiplied by the hedge of the average asset
in terms of the assets X and Y (∆ ¯ X (t), or ∆
¯ Y (t)). This feature is rather
unique among contingent claims. The exact forms of the hedging portfolio are
given in the following theorem. Recall that
Z T
¯ X (t) =
∆ µ(ds),
t

and
Z t
¯ Y (t) =
∆ XY (s)µ(ds).
0

THEOREM 9.3
The hedging portfolio Pt of the Asian option admits each of the following
 Asian Options 231

equivalent represenations:
" #
Pt = uYx ¯ X Y
(t, XY (t), AY (t)) + ∆ (t) · uy (t, XY (t), AY (t)) · X
"
+ uY (t, XY (t), AY (t)) − XY (t) · uYx (t, XY (t), AY (t))
#

Y
¯ Y
+ ∆ (t) − AY (t) · uy (t, XY (t), AY (t)) · Y, (9.35)

"
Pt = uX (t, YX (t), AX (t)) − YX (t) · uX
x (t, YX (t), AX (t))

#

X
¯ X
+ ∆ (t) − AX (t) · uy (t, YX (t), AX (t)) · X
" #
+ uX ¯Y
(t, YX (t), AX (t)) + ∆ (t) · uX (t, YX (t), AX (t)) · Y, (9.36)
x y

"
h i
uA (t, XA (t), YA (t)) − uA ¯X
Pt = y (t, XA (t), YA (t)) · YA (t) · ∆ (t)

#
h i
+ uA ¯ (t)XA (t) · X
(t, XA (t), YA (t)) · 1 − ∆X
x

"
h i
+ uA (t, XA (t), YA (t)) − uA ¯Y
x (t, XA (t), YA (t)) · XA (t) · ∆ (t)

#
h i
+ uA
y (t, XA (t), YA (t))
¯ Y
· 1 − ∆ (t)YA (t) · Y. (9.37)

PROOF Let us find a hedge for the Asian option of the form
Pt = ∆X (t) · X + ∆Y (t) · Y.
Using the fact that the process uY (t, XY (t), AY (t)) has a zero dt term, we
get
duY = uYx · dXY (t) + uYy · dAY (t)


= uY + ∆ ¯ X (t)uY · dXY (t).
x y

Thus the hedging position in the asset X is given by the formula

¯ X (t)uY (t, XY (t), AY (t)) .
∆X (t, XY (t), AY (t)) = uYx (t, XY (t), AY (t)) + ∆ y
(9.38)
232 Stochastic Finance: A Numeraire Approach

Similarly, using the evolution of uX (t, YX (t), AX (t))

duX = uX X
x · dYX (t) + uy · dAX (t)


= uX ¯Y X
x + ∆ (t) · uy · dYX (t),

we get the following representation of the hedging position in the asset Y :

∆Y (t, YX (t), AX (t)) = uX ¯Y X

x (t, YX (t), AX (t)) + ∆ (t)uy (t, YX (t), AX (t)) .
(9.39)
Therefore the hedging portfolio takes the following form
 
¯ X (t)uYy (t, XY (t), AY (t)) · X
Pt = uYx (t, XY (t), AY (t)) + ∆
 
+ uX ¯Y X
x (t, YX (t), AX (t)) + ∆ (t)uy (t, YX (t), AX (t)) · Y. (9.40)

We can also rewrite the above representation of the hedging portfolio using
the function uY of the function uX only. From

uX (t, x, y) = uY (t, x1 , yx ) · x,

we get
1 y
uX Y
x (t, x, y) = u (t, x , x ) −
1
x · uYx (t, x1 , xy ) − y
x · uYy (t, x1 , xy ),

and
1 y
uX Y
y (t, x, y) = uy (t, x , x ).

Substituting into (9.40), we get

" #
Pt = uYx ¯X
(t, XY (t), AY (t)) + ∆ (t) · uYy (t, XY (t), AY (t)) · X
"
+ uY (t, XY (t), AY (t)) − XY (t) · uYx (t, XY (t), AY (t))
#

Y
¯Y
+ ∆ (t) − AY (t) · uy (t, XY (t), AY (t)) · Y.

Similarly, from
uY (t, x, y) = uX (t, x1 , yx ) · x,
we get

uYx (t, x, y) = uX (t, x1 , xy ) − 1

x · uX 1 y
x (t, x , x ) −
y
x · uX 1 y
y (t, x , x ),

and
1 y
uYy (t, x, y) = uX
y (t, x , x ).
 Asian Options 233

Substituting to (9.40), we get

"
Pt = uX (t, YX (t), AX (t)) − YX (t) · uX
x (t, YX (t), AX (t))

#


¯X
+ ∆ X
t − AX (t) · uy (t, YX (t), AX (t)) · X

" #
+ uX ¯Y
(t, YX (t), AX (t)) + ∆ (t) · uX (t, YX (t), AX (t)) · Y.
x y

Finally, from
duA = uA A
x dXA (t) + uy dYA (t),

we get a hedging portfolio representation of the form

 A 
Pt = uA · At = uA A A A
x · Xt + uy · Yt + u − XA (t) · ux − YA (t) · uy · At .

¯ X (t) · X + ∆
Using the fact that At = ∆ ¯ Y (t) · Y , we conclude that

"
h i
Pt = uA (t, XA (t), YA (t)) − uA (t, X A (t), YA (t)) · YA (t) ¯ X (t)
·∆
y

#
h i
+ uA
x
¯ (t)XA (t) · X
(t, XA (t), YA (t)) · 1 − ∆X

"
h i
+ uA (t, XA (t), YA (t)) − uA (t, X A (t), YA (t)) · X A (t) ¯ Y (t)
·∆
x

#
h i
+ uA ¯ (t)YA (t) · Y.
(t, XA (t), YA (t)) · 1 − ∆Y
y

9.3 Reduction of the Pricing Equations

When the Asian option contract depends only on the assets A and X, such
as in the case of an Asian call option with a floating strike that has a payoff
(AT − K · XT )+ , the option pricing problem depends only on the price process
AX (t), and thus the corresponding partial differential equations depend only
on one spatial variable. In this case the pricing equation (9.27) does not depend
on the variable x that represents the price YX (t) that is irrelevant to this
234 Stochastic Finance: A Numeraire Approach

problem, and thus it reduces to the partial differential equation

uX 1 2 ¯X 2 X
t (t, y) + 2 σ (y − ∆ (t)) uyy (t, y) = 0, (9.41)

with the terminal condition

uX (T, y) = f X (y), (9.42)

where
uX (t, y) = EX [f X (AX (T ))|AX (t) = y].
We keep the notation y (as opposed to x) for the only spatial variable in
order to be consistent with the pricing problem (9.27). Similarly, when A is a
reference asset and the payoff depends only on XA (T ), the pricing equation
(9.29) does not depend on the variable y, and the partial differential equation
simplifies to

uA 1 2 2 ¯X 2 A
t (t, x) + 2 σ x [x∆ (t) − 1] · uxx (t, x) = 0, (9.43)

with the terminal condition

uA (T, x) = f A (x), (9.44)

where
uA (t, x) = EA [f A (XA (T ))|XA (t) = x].

The formulas for the hedging portfolio given in Equations (9.36) and (9.37)
also simplify to
" #


¯ X (t) − AX (t) · uX
Pt = uX (t, AX (t)) + ∆ y (t, AX (t)) · X

" #
¯Y
+ ∆ (t) · uX
y (t, AX (t)) · Y, (9.45)

and
" #
h i
¯X A
Pt = ∆ (t) · u (t, XA (t)) + uA ¯ (t)XA (t) · X
· 1−∆X
x (t, XA (t))

" #
h i
¯ Y (t) · uA (t, XA (t)) − uA
+ ∆ x (t, XA (t)) · XA (t) · Y. (9.46)

The pricing equation (9.25) does not reduce in this case, and it is strictly
suboptimal to employ it for pricing Asian options that do not depend on the
 Asian Options 235

asset Y .

When the contract depends on the assets A and Y only, such as in the case
of the Asian call option with a fixed strike that has a payoff (AT − K · YT )+ ,
the reduction of the pricing equations is possible only in special cases, not in
general. The reason is that the evolution of the price process AY (t) depends on
both prices AY (t) and XY (t) (in contrast to the evolution of the price AX (t)
that depends only on itself), and thus the partial differential equation (9.25)
cannot be reduced to only one spatial variable. However, when the payoff of
the contract is only a function of the asset F known as the Asian forward
defined as
FT = AT − K1 YT , (9.47)
a reduction of the pricing problem similar to Equation (9.41) is possible
when the asset X is taken as a numeraire. Consider a contract that pays off
f X (FX (T )) units of an asset X, where K1 in (9.47) is a constant. When the
payoff function is given by f X (x) = (x − K2 )+ , the contract that corresponds
to it is

[FX (T )−K2 ]+ ·X = (AX (T )−K1 YX (T )−K2 )+ ·X = (AT −K1 YT −K2 XT )+ ,

which covers both the floating strike option when K1 = 0, and the fixed strike
option when K2 = 0.

Let us define

uX (t, x) = EX [f X (FX (T ))|FX (t) = x].

In order to get the partial differential equation for uX , we need to determine

dFX (t). Note that

dFX (t) = d [AX (t) − K1 YX (t)]

 Y 
= ¯ (t) − K1 · dYX (t)
∆
 
= ¯ X (t) · XY (t) − K1 · dYX (t)
AY (t) − ∆
 
= ¯ X (t) · XY (t) − K1 σYX (t)dW X (t)
AY (t) − ∆
 
= σ [AX (t) − K1 YX (t)] − ∆ ¯ X (t) dW X (t)
 
= σ FX (t) − ∆ ¯ X (t) dW X (t).

Therefore  
¯ X (t) dW X (t),
dFX (t) = σ FX (t) − ∆
which is identical to an evolution of the average asset A. Therefore the pricing
partial differential equation takes the same form as (9.41):

uX 1 2 ¯X 2 X
t (t, x) + 2 σ (x − ∆ (t)) uxx (t, x) = 0 (9.48)
236 Stochastic Finance: A Numeraire Approach

with the terminal condition

uX (T, x) = f X (x). (9.49)

Thus we can also efficiently solve the Asian call option with the fixed strike
using the above partial differential equation. Note the important difference
from Equation (9.41). In the previous case, the basic price process was AX (t),
the price of the average asset A in terms of the reference asset X. The partial
differential equation (9.48) applies to the price process FX (t), the price of the
Asian forward F in terms of the reference asset X. The corresponding spatial
variables are shifted by the factor K1 YX (t) as

AX (t) − FX (t) = AX (t) − AX (t) + K1 YX (t) = K1 YX (t).

Note that while AX (t) is always positive, FX (t) can become zero or even be-
come negative, and thus the Asian forward F cannot be used as a reference
asset for the purposes of pricing. Thus in contrast to the case of the average
asset A, there is no partial differential equation where F serves as a reference
asset.

The hedging portfolio agrees with (9.45), but the value of AX (t) is replaced
by FX (t):
" #

X
X ¯ X
Pt = u (t, FX (t)) + ∆ (t) − FX (t) · ux (t, FX (t)) · X
" #
¯Y
+ ∆ (t) · uX
x (t, FX (t)) · Y. (9.50)

References and Further Reading

The approach to Asian options presented in this text extends previous works
of Vecer (2001, 2002) which used the average asset as the natural asset for
pricing. The characterization of the Asian option price with partial differential
equations was known even earlier; see for instance Rogers and Shi (1995), but
the asset that was considered for pricing was the running average which is
an arbitrage asset, and the corresponding partial differential equation had
extra terms that appear in the connection with the time value of the running
average. Use of the average asset for pricing Asian options is not limited to
the geometric Brownian motion. It is possible to generalize this approach to
other martingale models of the price as shown by Fouque and Han (2003)
for stochastic volatility or for models with jumps as shown in Vecer and Xu
(2004), and later by Bayraktar and Xing (2010). Hoogland and Neumann
 Asian Options 237

(2000) and Henderson and Wojakowski (2002) pointed out the symmetries
between the fixed and the floating strike Asian options. Other relevant papers
include Geman and Yor (1993), Curran (1994), Linetsky (2004), Dufresne
(2000), D’Halluin et al. (2005), Milevsky and Posner (1998), or Nielsen and
Sandmann (2003).

Exercises
9.1 Show that 1, x, and y are solutions of the partial differential equations
for the Asian option prices uY , uX , and uA . This is useful in calibrating
numerical schemes. What are the contracts corresponding to these solutions?

9.2 Find the price of the contract that pays off

[AY (T )]2
VT = [AX (T )]2 · X = · Y,
XY (T )

where A is the average asset.

Hint: The price function uX (t, x) = VX (t) satisfies the partial differential
equation
uX 1 2 ¯X 2 X
t (t, y) + 2 σ (y − ∆ (t)) uyy (t, y) = 0,

with the terminal condition

uX (T, y) = y 2 .

Consider a solution of the form

uX (t, y) = a2 (t)y 2 + a1 (t)y + a0 (t).

Plug it into the partial differential equation for uX (t, x), and find the functions
a2 (t), a1 (t), and a0 (t) by solving the resulting ordinary differential equations.
Chapter 10
Jump Models

This chapter studies jump models of a price evolution. A martingale in con-

tinuous time can be written as a sum of a diffusion martingale and a pure
jump martingale. Therefore a jump evolution is the second way of describing
dynamics of the price. Since real markets quote and trade only at discrete
levels and at discrete times, it may not be obvious if the true underlying price
process is continuous or if it has jumps, as long as the jump sizes are relatively
small and frequent.

The most basic process with jumps is a Poisson process N (t). The Poisson
process makes a jump in the time interval [t, t + ∆t] with probability λ∆t,
where λ is the intensity of the process. The Poisson process itself is nonde-
creasing, and thus it is not a martingale. However, the compensated Poisson
process, N (t) − λt, is a martingale, and thus can serve as the basic model
of market noise with jumps. In analogy to the geometric Brownian motion
model of the price which is driven by Brownian motion, we consider a geo-
metric Poisson process model of the price, which is driven by a compensated
Poisson process.

To price European options, we need to know the evolution of both XY (t)

and YX (t) in order to determine both martingale measures PY and PX . It
is possible to preserve the symmetry of the evolution of the prices with the
exception that the jump preserves the direction: when XY (t) jumps up, YX (t)
jumps down, and vice versa. The jump N (t) belongs to the pair of X and Y ;
it cannot be individualized to one asset in contrast to the geometric Brownian
motion model, where the noise factor W Y is associated with the asset Y , and
the noise factor W X is associated with the asset X. In the case of Poisson
evolution, it is the intensity λ of the Poisson process that is associated with
the particular asset. Under the PY measure, the process N (t) − λY t is a mar-
tingale, while under the PX measure, the process N (t) − λX t is a martingale.
We will show that the values of λY and λX are linked by the relationship
λX = eγ λY , where γ is the size of the jump of log(XY (t)).

The price of a European option can be computed via the Black–Scholes for-
mula. Equivalently, the price satisfies a difference differential equation when
the price follows a geometric Poisson process. The geometric Poisson process

239
240 Stochastic Finance: A Numeraire Approach

model of the price represents a complete market, and thus it is possible to

construct a perfect hedge for a European contract. It turns out that the hedg-
ing position in the asset X does not agree with PXt (XY (T ) ≥ K) in contrast
to the geometric Brownian motion model; it is slightly smaller when the price
process XY (t) has negative jumps. We determine the explicit form of the
hedging position, and study the difference with PXt (XY (T ) ≥ K).

A more general model of the evolution of the price with jumps considers
jumps of various sizes and intensities. When the price process is driven by
a noise with independent and identically distributed increments, we call this
noise a Lévy process. The prices of European contingent claims are character-
ized by an integro-differential equation. Models with multiple jump sizes are
incomplete, and no perfect hedging is possible in this situation.

10.1 Poisson Process

The Poisson process N (t) is a continuous time Markov process that takes
integer values and satisfies the following:
P(N (t + ∆t) = N (t)) = 1 − λ∆t + o(∆t), (10.1)
P(N (t + ∆t) = N (t) + 1) = λ∆t + o(∆t), (10.2)
P(N (t + ∆t) = N (t) + 2) = o(∆t), (10.3)
where o(∆t) represents some function that is much smaller than ∆t for small
∆t, meaning
o(∆t)
lim = 0.
∆t→0 ∆t
The parameter λ is known as the intensity of the Poisson process. Note that
the increment of the Poisson process N (t) − N (s) is independent of the infor-
mation set Fs .

One can think of a Poisson process as a limit of a binomial distribution. We

can write
n
X 


N (t) = N ni · t − N i−1 n ·t .
i=1



Each Poisson increment N ni · t − N i−1 n · t has approximately Bernoulli
distribution, meaning that it takes only two values 0 or 1 with correspond-
ing probabilities 1 − λ nt and λ nt . Thus N (t) has approximately a binomial
distribution with
   k  n−k
n λt λt
P(N (t) = k) ≈ 1− .
k n n
 Jump Models 241

As n → ∞, we have
 
n −k n(n − 1) . . . (n − k + 1) 1
lim n = lim = ,
n→∞ k n→∞ k!nk k!
and
 n−k  n  −k
λt λt λt
lim 1 − = lim 1 − lim 1 − = e−λt .
n→∞ n n→∞ n n→∞ n
Hence,
(λt)k
P(N (t) = k) = e−λt .
k!
In other words, N (t) is a Poisson random variable with parameter λt.

Let τ be the time of the first jump of the Poisson process. Note that

P(τ ≤ t) = P(N (t) ≥ 1) = 1 − P(N (t) = 0) = 1 − e−λt .

The first equality follows from the fact that if there is a jump before time t,
the value of the Poisson process must be greater than or equal to 1 at time t.
The density of the time of the first jump is given by
∂
f (t) = P(τ ≤ t) = λe−λt ,
∂t
which corresponds to an exponential random variable with parameter λ.

For modeling price processes we need a martingale market noise that would
ensure that the resulting portfolios are arbitrage free. In pricing models with
continuous paths, the role of the market noise was played by Brownian mo-
tion, where all prices were represented as stochastic integrals with respect to
Brownian motion. In the case of jump models, a Poisson process N (t) itself
is nondecreasing, and thus it is not a martingale. However, a compensated
Poisson process, N (t) − λt, is a martingale. This is easy to see from

Es [N (t) − λt] = Es [(N (t) − N (s)) + N (s) − λt]

= E[N (t) − N (s)] + N (s) − λt = λ(t − s) + N (s) − λt = Ns − λs.

The compensated Poisson process N (t) − λt is a jump analog of Brownian

motion W (t) in diffusion models. It can be regarded as model of a market
noise. Figure 10.1 shows a sample path of a compensated Poisson process.
Note that the path is discontinuous at the time of each jump; the process
jumps up by 1.

The basic idea is that all price processes that are driven by the compensated
Poisson process market noise can be represented as stochastic integrals with
242 Stochastic Finance: A Numeraire Approach

20

15
Compensated Poisson

10

−5

−10
0 1/4 1/2 3/4 1
Time

FIGURE 10.1: A sample path of a compensated Poisson process with λY =

300, T = 1.

respect to that noise. The resulting price process should be a martingale in or-
der not to have any arbitrage opportunity. As the following example suggests,
one has to exclude integrands that have prior information about the jumps,
and thus can create an opportunity for a risk-free profit. Note the following
relationship holds
Z t Z t
∆N (s)d(N (s) − λs) = dN (s) = N (t),
0 0

where ∆N (t) = N (t) − N (t−) = 1 at the time of the jump and zero other-
wise. This shows that the stochastic integrals with respect to a martingale
may not end up being a martingale. This results in an arbitrage opportunity,
which would make this model undesirable. However, taking a position ∆N (t)
is problematic at the time of the jump. This would require that the investor
set his position in the underlying market noise N (t) − λt to one at the exact
time of the jump, which is not possible since the jump happens unexpectedly.
In practice, there is a delay between first observing the jump and being able
to switch the position in the underlying asset. Thus we must incorporate this
delay in our analysis. The delay itself may be infinitesimally small as long as
the model preserves the correct order of the two events: the jump precedes
the switch of the position.

The integrands that preserve the precedence of the jump before switching
the position in the underlying asset are known as predictable processes. In
particular, processes that are continuous from the left are suitable integrands
 Jump Models 243

as the change in the trading position does not come unexpectedly in this case.
This is not the case for a Poisson process N (t) which is not left-continuous,
and the jump comes unexpectedly. A function h is called left-continuous if

h(t) = lim h(s).

s↑t

In order to use functions of the Poisson process N (t) as a legitimate position

in the underlying asset, the process must be slightly “delayed” in order to
make sure that the jump has been observed first. The corresponding delayed
process is given by
N (t−) = lim N (s), (10.4)
s↑t

which gives the pre-jump value of the Poisson process N (t).

10.2 Geometric Poisson Process

Let X (t) be a general market noise, such as a Brownian motion W (t), or a
compensated Poisson process N (t) − λt. The market noise can take negative
values, and as such, it does not serve as a good model for a possible price
evolution. One can consider the so-called stochastic exponential E(t) of the
process X (t) instead which takes positive values. The stochastic exponential
of X (t) is defined by
dE(t) = E(t−)dX (t) (10.5)
with E(0) = 1. When X (t) is a linear function X (t) = at, the stochastic
exponential reduces to an ordinary exponential E(t) = eat . The linear function
at does not correspond to a market noise, but the stochastic exponential is
still well defined, and it shows the relationship between an ordinary and a
stochastic exponential. When X (t) is a Brownian motion scaled by a factor
σ, so that X (t) = σW (t), the corresponding stochastic exponential satisfies

dE(t) = E(t)d(σ · W (t)) = σE(t)dW (t),

and its solution is a geometric Brownian motion E(t) = exp(σW (t) − 12 σ 2 t).

Let us determine the exact formula for a geometric Poisson process that
has the following dynamics

dXY (t) = (eγ − 1) · XY (t−)d(N (t) − λY t). (10.6)

This corresponds to the driving noise (eγ − 1) · (N (t) − λY t). The factor
(eγ − 1) plays a similar role as the volatility σ in the geometric Brownian
motion model. It is more convenient to use (eγ − 1) as opposed to simply
244 Stochastic Finance: A Numeraire Approach

using γ as it leads to a symmetric solution for the inverse price YX (t). Note
that using a Taylor series eγ − 1 ≈ γ.

The stochastic differential equation (10.6) is analogous to the stochastic

differential equation for geometric Brownian motion
dXY (t) = σXY (t)dW Y (t).
However, there is a major difference between the diffusion and jump models.
In diffusion models, the noise factors come with each individual reference as-
set. For instance in the geometric Brownian motion model, the asset Y has
its own noise factor W Y , and the asset X has its own noise factor W X . In
even more complex models, such as in the case of power options, each power
option Rα comes with its own noise factor W (α) . Although the driving noises
(such as W Y and W X ) are perfectly correlated in the diffusion models, one
can still assign a proprietary noise to each asset.

This is not the case for jump models. The jump N (t) belongs to the pair of
assets X and Y , and it cannot be individualized to just one asset. When the
price XY (t) jumps up, the inverse price YX (t) jumps down, and vice versa.
The individual part of the noise is the compensation factor λY ; different assets
have different compensators. The driving process, N (t) − λY t, is a martingale
under the PY measure that corresponds to the reference asset Y . This means
that the Poisson process N (t) has intensity λY , or in other words,
(λY t)k
PY (N (t) = k) = exp(−λY t) · .
k!
The individual assets may disagree on the distribution of jumps. The reader
should keep in mind that this does not indicate how likely it is that we will
get a particular number of jumps N (t), but how costly it is to deliver the
underlying asset on such an outcome.

The geometric Poisson process itself is an example of a complete model of

the price process in which the contingent claims can be perfectly replicated
by trading in the underlying assets. When γ > 0, the price process jumps up
at each jump time, but it is compensated by a deterministic drift factor in
the exponential function −(eγ − 1)λY t < 0. On the other hand when γ < 0,
the price process jumps down at each jump time, but it is deterministically
increasing between jumps with drift factor −(eγ − 1)λY t > 0.

If there were no jump, the price process would follow a simple deterministic
evolution
dXY (t) = −(eγ − 1) · XY (t−)λY dt,
that admits an exponential solution


XY (t) = XY (0) exp −(eγ − 1)λY t . (10.7)
 Jump Models 245

If there were only jumps, the price process would follow

dXY (t) = (eγ − 1) · XY (t−)dN (t),

meaning that at the time of the jump, the price process adds (eγ − 1) · XY (t−)
to its pre-jump value XY (t−) so that

XY (t) = (eγ − 1) · XY (t−) + XY (t−) = eγ · XY (t−),

and
∆XY (t) = XY (t) − XY (t−) = (eγ − 1) · XY (t−).
The price process jumps by a factor of eγ at the time of the jump. In conclu-
sion, we have
XY (t) = XY (0)eγN (t) . (10.8)
Combining the results from (10.7) and (10.8), we conclude that the geometric
Poisson process is given by


XY (t) = XY (0) exp γ · N (t) − (eγ − 1)λY t . (10.9)

REMARK 10.1 The stochastic exponential for a general process X (t) is

given by the Doléans-Dade formula

Y
E(t) = exp X (t) − 21 [X , X ]c (t) (1 + ∆X (s)) exp(−∆X (s)), (10.10)
s≤t

where [X , X ](t) is the quadratic variation of X (t) defined by

Z t
[X , X ](t) = [X (t)]2 − 2 X (s−)dX (s),
0

and where [X , X ]c (t) is the continuous part of [X , X ](t).

For instance, when X (t) = σW (t), the process X (t) has no jumps, and thus
the part in (10.10) in the product is simply equal to one. Quadratic variation
of X (t) (Brownian motion scaled by a factor σ) is given by
Z t
[σW, σW ](t) = [σW (t)]2 − 2 [σW (s)]d[σW (s)]
0
= [σW (t)]2 − ([σW (t)]2 − σ 2 t) = σ 2 t.


Thus we have E(t) = exp σW (t) − 21 σ 2 t .

Similarly, when X (t) = (eγ − 1)(N (t) − λt), the continuous part of the
quadratic variation [X , X ]c (t) is zero (only jumps contribute to the quadratic
246 Stochastic Finance: A Numeraire Approach

variation in this case). At the time of the jump, ∆X (t) = X (t) − X (t−) =
eγ − 1. Thus we have

Y
E(t) = exp X (t) − 21 [X , X ]c (t) (1 + ∆X (s)) exp(−∆X (s))
s≤t
Y
γ
= exp ((e − 1)(N (t) − λt)) (1 + (eγ − 1)) · exp(−(eγ − 1))
s≤t

= exp ((e − 1)(N (t) − λt)) · eγN (t) · exp(−(eγ − 1)N (t))
γ

= exp (γ · N (t) − (eγ − 1)λt) ,

which agrees with the formula for the geometric Poisson process.

For pricing European options written on assets X and Y , we also need

to determine the probability measure PX . We can deduce this probability
measure from the dynamics of the inverse price, YX (t). We can apply the
Ito’s formula for jump processes
Z t
f (X(t)) = f (X(0)) + f ′ (X(s−))dX(s)
0
X
+ [f (X(s)) − f (X(s−)) − f ′ (X(s−))∆X(s)] (10.11)
0≤s≤t

1
for the choice of the function f (x) = x applied to the process XY (t). Using
the fact that f ′ (x) = − x12 , we get
Z t
1 1 1
= + 2
− · (eγ − 1) · XY (s−)d(N (s) − λY s)
XY (t) XY (0) 0 X Y (s−)
X  1 1 1

+ − + ∆XY (s)
XY (s) XY (s−) XY (s−)2
0≤s≤t
Z t
= YX (0) − (eγ − 1)YX (s−)d(N (s) − λY s)
0
Z t
1 γ

+ eγ − 1 + (e − 1) YX (s−)dN (s)
0
Z t Z t
 
= YX (0) + e−γ − 1 YX (s−)dN (s) + (1 − eγ )YX (s−)d(−λY s).
0 0

Rewriting these dynamics in the differential form, we get

   
dYX (t) = e−γ − 1 YX (t−)dN (t) − eγ · e−γ − 1 YX (t−)λY dt
 
= e−γ − 1 YX (t−)d(N (t) − eγ λY t).

The inverse price YX (t) should have a martingale evolution under the proba-
bility measure that corresponds to the reference asset X. This is possible only
 Jump Models 247

when the driving process N (t) − eγ λY t itself is a martingale under the PX

measure. This means that the Poisson process N (t) has intensity λX = eγ λY
under the PX measure:
(λX t)k
PX (N (t) = k) = exp(−λX t) · ,
k!
where
λX = eγ λY . (10.12)
We conclude that the dynamics of the inverse price YX (t) satisfy
 
dYX (t) = e−γ − 1 YX (t−)d(N (t) − λX t), (10.13)
which admits the following solution:


YX (t) = YX (0) exp −γN (t) − (e−γ − 1)λX t . (10.14)

Note that at the time of the jump we have

YX (t) = e−γ − 1 · YX (t−) + YX (t−) = e−γ · YX (t−),
and

∆YX (t) = YX (t) − YX (t−) = e−γ − 1 · YX (t−).
The roles of X and Y in (10.6) and (10.13) are exchangeable when we sub-
stitute γ for −γ, and λY for λX = eγ λY .

Using the above arguments, we have shown the following mathematical

result:

THEOREM 10.1 Poisson Change of Measure

Let N (t) be a Poisson process with intensity λY under the PY measure. Then
N (t) is a Poisson process with intensity λX = eγ λY under the PX measure,
where
dPX XY (t)

Z(t) = = = exp γ · N (t) − λY (eγ − 1) t .
dPY XY (0)

Example 10.1
Let A be the event {N (T ) = k}. Using the Radon–Nikodým derivative, we
can write
Z
X X
P (N (T ) = k) = P (A) = Z(T, ω)dPY (ω)
A

(λY T )k
= exp γ · k − λY (eγ − 1) T · exp(−λY T ) ·
k!
γ Y k
(e λ T )
= exp(−eγ λY T ) ·
k!
X k
(λ T )
= exp(−λX T ) · ,
k!
248 Stochastic Finance: A Numeraire Approach

which is a Poisson distribution with parameter λX T .

10.3 Pricing Equations

Let V be a contract that pays off f Y (XY (T )) units of an asset Y at time
T . The price of the contract V with respect to the reference asset Y is given
by  Y 
VY (t) = EYt f (XY (T )) . (10.15)
The conditional expectation on the right hand side of the above equation
which gives the price of the contract with respect to Y , is a PY martingale,
and its value depends only on the price of XY (t). Thus we can write
 
uY (t, x) = EY f Y (XY (T )) |XY (t) = x . (10.16)
We can also compute the price of this contract with respect to a reference
asset X as  X 
VX (t) = EXt f (YX (T )) , (10.17)
where f X is a payoff function in terms of the asset X. The functions f Y
and f X are related by a perspective mapping f X (x) = f Y ( x1 ) · x. The price
function uX (t, x) is defined
 
uX (t, x) = EX f X (YX (T )) |YX (t) = x . (10.18)
As in the geometric Brownian motion model, the price functions uY (t, x) and
uX (t, x) in the jump model are solutions to an equation, in this case of a
difference-differential type.

THEOREM 10.2  
The price function uY (t, x) = EY f Y (XY (T )) |XY (t) = x satisfies
 
uYt (t, x) + λY uY (t, eγ x) − uY (t, x) − (eγ − 1)xuYx (t, x) = 0. (10.19)

with the terminal condition

uY (T, x) = f Y (x). (10.20)
The hedging portfolio is given by
 
uY (t, eγ XY (t)) − uY (t, XY (t))
Pt = ·X
(eγ − 1)XY (t)
 γ Y 
e · u (t, XY (t)) − uY (t, eγ XY (t))
+ · Y. (10.21)
eγ − 1
 Jump Models 249
 
The price function uX (t, x) = EX f X (YX (T )) |YX (t) = x satisfies
 X


uX
t (t, x) + λ
X
u t, e−γ x − uX (t, x) − e−γ − 1 xuX
x (t, x) = 0.
(10.22)
with the terminal condition
uX (T, x) = f X (x). (10.23)
The hedging portfolio is given by
 
e−γ · uX (t, YX (t)) − uX (t, e−γ YX (t))
Pt = ·X
e−γ − 1
 X 
u (t, e−γ YX (t)) − uX (t, YX (t))
+ · Y. (10.24)
(e−γ − 1) YX (t)

PROOF According to Ito’s formula for jump processes, we have

Z t
uY (t, XY (t)) = uY (0, XY (0)) + uYt (s, XY (s−)) ds
0
Z t
Y
+ ux (s, XY (s−)) dXY (s)
0
X h
+ uY (s, XY (s)) − uY (s, XY (s−))
0≤s≤t
i
−uYx (s, XY (s−)) ∆XY (s)
Z t
Y
= u (0, XY (0)) + uYt (s, XY (s−)) ds
0
Z t
+ uYx (s, XY (s−)) (eγ − 1)XY (s−)d(N (s) − λY s)
0
Z t
 Y
+ u (s, eγ XY (s−)) − uY (s, XY (s−))
0

−uYx (s, XY (s−)) (eγ − 1)XY (s−) dN (s)
= uY (0, XY (0))
Z t
 Y
+ ut (s, XY (s−)) − λY (eγ − 1)XY (s−)uYx (s, XY (s−))
0


+ λY uY (s, eγ XY (s−)) − uY (s, XY (s−)) ds
Z t


+ uY (s, eγ XY (s−)) − uY (s, XY (s−)) d(N (s) − λY s).
0

We have subtracted the

 Y 
u (s, eγ XY (s−)) − uY (s, XY (s−)) d(λY s)
250 Stochastic Finance: A Numeraire Approach

term from dN (s) in order to obtain the martingale part of the equality and
added it back to the corresponding ds term. Since uY (t, XY (t)) is a PY -
martingale, the corresponding ds term that is not a part of the martingale
N (s) − λY s must vanish. Thus we must have
 
uYt (t, x) + λY uY (t, eγ x) − uY (t, x) − (eγ − 1)xuYx (t, x) = 0.
The terminal condition is given by
uY (T, x) = f Y (x).
The contract follows a martingale evolution given by
 
duY (t, XY (t)) = uY (t, eγ XY (t−)) − uY (t, XY (t−)) d(N (t) − λY t).
(10.25)
If we consider a replicating portfolio for this contract of the form
Pt = ∆X (t) · X + ∆Y (t) · Y,
where ∆X (t) is the number of units of an asset X, and ∆Y (t) is the number
of units of an asset Y , we also have
dPY (t) = ∆X (t)dXY (t) = ∆X (t)(eγ − 1)XY (t−)d(N (t) − λY t). (10.26)
Should the price and the dynamics of the replicating portfolio P be identical
to the price and the dynamics of the contract V , we must also have
dPY (t) = dVY (t) = duY (t, XY (t)) .
Comparing (10.25) with (10.26), we get the following representation for the
hedge:
uY (t, eγ XY (t−)) − uY (t, XY (t−))
∆X (t) = .
(eγ − 1)XY (t−)
This hedging position should be held at all times, whether they are jump
times or not, and thus we can also write
uY (t, eγ XY (t)) − uY (t, XY (t))
∆X (t) = .
(eγ − 1)XY (t)
The hedging position ∆Y (t) in the asset Y is given by
∆Y (t) = uY (t, XY (t)) − ∆X (t) · XY (t)
uY (t, eγ XY (t)) − uY (t, XY (t))
= uY (t, XY (t)) − · XY (t)
(eγ − 1)XY (t)
eγ · uY (t, XY (t)) − uY (t, eγ XY (t))
= .
eγ − 1
This concludes the proof for the price function uY . The result for the price
function uX is proved in a similar way.
 Jump Models 251

10.4 European Call Option in Geometric Poisson Model

The price and the hedging formulas can be further simplified for a European
call option with payoff (XT − K · YT )+ . The price of an asset X in terms of
a reference asset Y is given by


XY (T ) = XY (t) exp γ · N (T − t) − (eγ − 1)λY (T − t)


= XY (t) exp γ · N (T − t) + (e−γ − 1)λX (T − t) ,

where λX = eγ λY in the geometric Poisson model. The price of a European

call option V is given by the Black–Scholes formula

Vt = PX Y
t (XY (T ) ≥ K) · X − K · Pt (XY (T ) ≥ K) · Y.

The goal of this subsection is to determine the probabilities PX

t (XY (T ) ≥ K)
and PYt (XY (T ) ≥ K). The event

XY (T ) ≥ K (10.27)

is equivalent to


N (T − t) ≤ 1
γ · − log 1
K · XY (t) + (eγ − 1)λY (T − t) (10.28)

when γ < 0, and to



N (T − t) ≥ 1
γ · − log 1
K · XY (t) + (eγ − 1)λY (T − t) (10.29)

when γ > 0. When γ < 0, we have

PX
t (XY (T ) ≥ K)
 
 
= P γ1 · − log 1
K · XY (t) + (eγ − 1)λY (T − t) ; λX (T − t) , (10.30)

and

PYt (XY (T ) ≥ K)
 
 
= P γ1 · − log 1
K · XY (t) + (eγ − 1)λY (T − t) ; λY (T − t) , (10.31)

where P(x; λ) is the cumulative distribution function of the Poisson random

variable with intensity λ defined as
⌊x⌋
X λk
P(x; λ) = e−λ · .
k!
k=0
252 Stochastic Finance: A Numeraire Approach

FIGURE 10.2: The price of a call option VY with payoff (XT − K · YT )+ as

a function of XY and time to maturity. The parameters are K = 21 , γ = −0.02,
λY = 300.

The price of the European call option in the case of γ < 0 is given by the
formula

VY (t) = XY (t) · P d; λX (T − t) − K · P d; λY (T − t) , (10.32)

where

h   i
d= 1
γ · log XYK(t) + (eγ − 1)λY (T − t) . (10.33)

Similarly when γ > 0, the price of the European call option is given by

VY (t) = XY (t) · (1 − P d; λX (T − t) ) − K · (1 − P d; λY (T − t) ). (10.34)

The price function uY (t, x) is plotted in Figure 10.2.

Let us determine the hedging portfolio for the European call option. Ac-
 Jump Models 253

cording to the Theorem 10.2, the hedging formulas are given by

uY (t, eγ XY (t)) − uY (t, XY (t))

∆X (t) =
(eγ − 1)XY (t)
   
eγ XY (t) · PX XY (T ) ≥ K|eγ XY (t) − K · PY XY (T ) ≥ K|eγ XY (t)
=
(eγ − 1)XY (t)
   
XY (t) · PX XY (T ) ≥ K|XY (t) − K · PY XY (T ) ≥ K|XY (t)
−
(eγ − 1)XY (t)
   
eγ · PX XY (T ) ≥ K|eγ XY (t) − PX XY (T ) ≥ K|XY (t)
=
(eγ − 1)
   
PY XY (T ) ≥ K|eγ XY (t) − PY XY (T ) ≥ K|XY (t)
−K ·
(eγ − 1)XY (t)

for the case of the hedging position in the asset X, and as

eγ · uY (t, XY (t)) − uY (t, eγ XY (t))

∆Y (t) =
eγ − 1
   
e XY (t) · P XY (T ) ≥ K|XY (t) − Keγ · PY XY (T ) ≥ K|XY (t)
γ X

=
eγ − 1
   
eγ XY (t) · PX XY (T ) ≥ K|eγ XY (t) − K · PY XY (T ) ≥ K|eγ XY (t)
−
eγ − 1
   
PX XY (T ) ≥ K|XY (t) − PX XY (T ) ≥ K|eγ XY (t)
γ
= e XY (t) ·
eγ − 1
   
eγ PY XY (T ) ≥ K|XY (t) − PY XY (T ) ≥ K|eγ XY (t)
−K ·
eγ − 1

for the case of the hedging position in the asset Y .

The formulas for ∆X (t) and ∆Y (t) can be further simplified if we express
them in terms of a cumulative distribution function of the Poisson random
variable. Consider the case when γ < 0. Then we have
 
P XY (T ) ≥ K|XY (t) = P(d; λ),

and
 
P XY (T ) ≥ K|eγ XY (t) = P(d − 1; λ).
254 Stochastic Finance: A Numeraire Approach

Therefore we can write

   
eγ · PX XY (T ) ≥ K|eγ XY (t) − PX XY (T ) ≥ K|XY (t)
∆X (t) =
(eγ − 1)
   
PY XY (T ) ≥ K|eγ XY (t) − PY XY (T ) ≥ K|XY (t)
−K ·
(eγ − 1)XY (t)
e · P(d − 1; λ (T − t)) − P(d; λX (T − t))
γ X
=
(eγ − 1)
P(d − 1; λY (T − t)) − P(d; λY (T − t))
−K ·
(eγ − 1)XY (t)
= P(d; λX (T − t))
eγ

+ γ · P(d − 1; λX (T − t)) − P(d; λX (T − t))
e −1
K

− γ · P(d − 1; λY (T − t)) − P(d; λY (T − t))
(e − 1)XY (t)
eγ X
−λX (T −t) (λ (T − t))
⌊d⌋
= P(d; λX (T − t)) − · e ·
eγ − 1 ⌊d⌋!
K Y (λY (T − t))⌊d⌋
+ · e−λ (T −t) ·
(eγ − 1)XY (t) ⌊d⌋!
= P(d; λX (T − t))
(λY (T − t))⌊d⌋ −λY (T −t) K  
γ(⌊d⌋+1−d)
− ·e · · 1 − e .
⌊d⌋! XY (t)(1 − eγ )
In contrast to the geometric Brownian motion model, the hedging position
in the asset X in the geometric Poisson process model, ∆X (t), is not equal to
PX X X
t (XY (T ) ≥ K) = P(d; λ (T − t)). While ∆ (t) is a smooth function of the
X
price of the underlying asset XY (t), Pt (XY (T ) ≥ K) is a piecewise constant
function. Figure 10.3 illustrates this situation.

The difference between PX (XY (T ) ≥ K) and ∆X (t) is equal to

(λY (T − t))⌊d⌋ −λY (T −t) K  
·e · γ
· 1 − eγ(⌊d⌋+1−d) .
⌊d⌋! XY (t)(1 − e )
Since 0 < (⌊d⌋ + 1 − d) ≤ 1, the factor (1 − eγ(⌊d⌋+1−d) ) is at least zero and
at most 1 − eγ . Therefore
(λY (T − t))⌊d⌋ −λY (T −t) K  
γ(⌊d⌋+1−d)
0< ·e · · 1 − e
⌊d⌋! XY (t)(1 − eγ )
h   i
1 K
· log +(eγ −1)λY (T −t)
(λY (T − t)) γ XY (t) Y K
≤  h   i  · e−λ (T −t) · .
Γ γ1 · log XYK(t) + (eγ − 1)λY (T − t) + 1 XY (t)
 Jump Models 255

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7
XY Price

FIGURE 10.3: Comparison of ∆X (smooth function) and PX (XY (T ) ≥

K) (step function) for parameters γ = −0.02, λY = 20, and T = 1.

Figure 10.4 illustrates the difference between PX (XY (T ) ≥ K) and ∆X (t).

The difference can be anywhere between zero and the upper bound derived
in the previous computation. The exact value depends on the difference
⌊d⌋ + 1 − d.

The hedging position ∆Y (t) is given by

∆Y (t) = VY (t) − ∆X (t)XY (t)

= −K · P(d, λY (T − t))
(λY (T − t))⌊d⌋ −λY (T −t) K  
γ(⌊d⌋+1−d)
+ ·e · · 1 − e .
⌊d⌋! (1 − eγ )
Figures 10.5 and 10.6 show the hedging positions in the assets X and Y as
a function of the price XY and time to maturity. The hedging positions in
the geometric Poisson process model are similar to the hedging positions in
the geometric Brownian motion model. Figure 10.7 shows a sample path of
a geometric Poisson process together with the price of the European option
with payoff (XT − K · YT )+ . Figure 10.8 shows the corresponding hedging po-
sitions in the assets X and Y . Since the option expires worthless, the terminal
hedging positions are zero. Note that the prices and the hedging positions ex-
hibit jumps. Figure 10.9 shows a different sample path of a geometric Poisson
process, but this time the European option expires in the money. Thus the
terminal hedging positions are given by ∆X (T ) = 1, and ∆Y = −K as seen
from Figure 10.10.
256 Stochastic Finance: A Numeraire Approach

0.04

0.035

0.03

0.025

0.02

0.015

0.01

0.005

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
XY Price

FIGURE 10.4: Difference between PX (XY (T ) ≥ K) and ∆X for parame-

ters γ = −0.02, λY = 120, and T = 1.

10.5 Lévy Models with Multiple Jump Sizes

The above results easily generalize to the case of jumps with different sizes
and intensities. Let us assume that the jump process has independent and
stationary increments, and let µ(dx, dt) denote a random measure associated
with jumps of the price process.
R t R The random measure has the following in-
terpretation. The quantity 0 A µ(dx, ds) represents the number of jumps of
sizes in the set A that happened in the time interval [0, t]. Let us denote
Z tZ
ν(t, A) = E µ(dx, ds), (10.35)
0 A

which is the expected number of jumps of sizes in the set A. This is known as
a compensator. The process
Z tZ ∞
(µ(dx, dt) − ν(dx, dt)) (10.36)
0 −∞

is a martingale, and it can serve as a model of the market noise. Since the
process has independent increments, it is time homogeneous, and we can also
 Jump Models 257

FIGURE 10.5: The hedging position in the asset X of a call option V

with payoff (XT − K · YT )+ as a function of XY and time to maturity. The
parameters are K = 12 , γ = −0.02, λY = 300.

FIGURE 10.6: The hedging position in the asset Y of a call option V

with payoff (XT − K · YT )+ as a function of XY and time to maturity. The
parameters are K = 12 , γ = −0.02, λY = 300.
258 Stochastic Finance: A Numeraire Approach

0.7

0.6

0.5

0.4
Price

0.3

0.2

0.1

0
0 1/4 1/2 3/4 1
Time

FIGURE 10.7: A sample path of the price of an asset XY (top) and the
price of a corresponding call option VY (bottom) with payoff (XT − K · YT )+
as a function of time. The parameters are K = 21 , γ = −0.02, λY = 300,
T = 1.

0.8

0.6

0.4
Hedge

0.2

−0.2

−0.4
0 1/4 1/2 3/4 1
Time

FIGURE 10.8: A sample path of the hedging position ∆X (t) and ∆Y (t)
that corresponds to the previous example as a function of time. The parame-
ters are K = 12 , γ = −0.02, λY = 300, T = 1. The option expires out of the
money, and thus the terminal positions are zero.
 Jump Models 259

0.8

0.7

0.6

0.5
Price

0.4

0.3

0.2

0.1

0
0 1/4 1/2 3/4 1
Time

FIGURE 10.9: A sample path of the price of an asset XY (top) and the
price of a corresponding call option VY (bottom) with payoff (XT − K · YT )+
as a function of time. The parameters are K = 21 , γ = −0.02, λY = 300,
T = 1.

1.2

0.8

0.6

0.4
Hedge

0.2

−0.2

−0.4

−0.6
0 1/4 1/2 3/4 1
Time

FIGURE 10.10: A sample path of the hedging position ∆X (t) and ∆Y (t)
that corresponds to the previous example as a function of time. The param-
eters are K = 12 , γ = −0.02, λY = 300, T = 1. The option expires in the
money, and thus the terminal positions are fully invested in the underlying
assets.
260 Stochastic Finance: A Numeraire Approach

write ν(dx, dt) = ν(dx)dt, so the dynamics of the jump process are determined
by the sizes of the jumps, not by time.

We have previously considered market noise driven by Brownian motion

or a Poisson process. The market noise in (10.36) is a generalization of a
compensated Poisson process, which corresponds to the choice of

ν(dx, dt) = λδ(1) (x)dx,

where δ(c) is the Dirac delta function (A.1). A compensated compound Poisson
process corresponds to
n
X
ν(dx, dt) = λi δ(ci ) (x)dx,
i=1

which means that the process makes jumps of sizes ci with intensities λi ,
i = 1, . . . , n. This is just a sum of n compensated Poisson processes with
the corresponding jump sizes and intensities. However, the jump process can
take a more general form; it can even exhibit infinitely large activity for small
enough jumps, R∞meaning that ν(x) may converge to infinity for x → 0. It is
possible that −∞ ν(dx) = ∞, which represents the case of infinite activity of
R∞
jumps, but we must assume that −∞ (|x|2 ∧ 1)ν(dx) < ∞ in order to prevent
explosions in the market noise. An example of a compensator that may have
infinite jump activity is for instance
( exp(−G|x|)
C |x|1+Y , x < 0
ν(x) =
C exp(−M|x|)
|x|1+Y , x > 0,

which corresponds to a CGMY model.

Let us assume that the price process is driven by the jump process with a
random measure µ(dx, dt) and a compensator ν(dx). Then both price pro-
cesses XY (t) and YX (t) are driven by the same jumps, but if one price
jumps up, the inverse price jumps down accordingly. Thus the jump mea-
sure µ(dx, dt) driving the price processes differs only in the sign of the jump:
µY (dx, dt) = µX (−dx, dt). The main difference is that the reference assets Y
and X place different intensity on the jumps, so ν Y represents the compensator
associated with the reference asset Y , while ν X represents the compensator
associated with the reference asset X. Let us find the relationship between
ν Y and ν X .

The price process XY (t) can be written as

Z ∞  
dXY (t) = (ex − 1) · XY (t−) µY (dx, dt) − ν Y (dx)dt , (10.37)
−∞
 Jump Models 261

which is a generalization of Equation (10.6). The integral is over different

jump sizes. From Ito’s formula, the inverse price process YX (t) satisfies
Z ∞  
dYX (t) = (e−x − 1) · YX (t−) µY (dx, dt) − ex ν Y (dx)dt , (10.38)
−∞
or
Z ∞  
dYX (t) = (ex − 1) · YX (t−) µX (dx, dt) − e−x ν Y (−dx)dt (10.39)
−∞

after switching the sign. From the symmetry between X and Y , this equation
can be written as
Z ∞  
dYX (t) = (ex − 1) · YX (t−) µX (dx, dt) − ν X (dx)dt . (10.40)
−∞

Comparing the two above equations, we conclude that the relationship of the
Lévy measures ν Y and ν X is given by

ν X (x) = e−x ν Y (−x). (10.41)

Let us consider a price process XY (t) of the form

Z ∞  
dXY (t) = (ex − 1) · XY (t−) µY (dx, dt) − ν Y (dx)dt . (10.42)
−∞

The corresponding inverse price process satisfies

Z ∞  
dYX (t) = (ex − 1) · YX (t−) µX (dx, dt) − ν X (dx)dt , (10.43)
−∞

where ν X (x) = e−x ν Y (−x).

THEOREM 10.3
The price function uY (t, x) = EY [VY (T )|XY (t) = x] satisfies integro-
differential equation
Z ∞
 
uYt (t, x) + ν Y (dy) uY (t, ey x) − uY (t, x) − (ey − 1)xuYx (t, x) = 0
−∞
(10.44)
with the terminal condition
uY (T, x) = f Y (x), (10.45)
and the price function uX (t, x) = EX [VX (T )|YX (t) = x] satisfies integro-
differential equation
Z ∞
X
 
ut (t, x) + ν X (dy) uX (t, ey x) − uX (t, x) − (ey − 1) xuX
x (t, x) = 0
−∞
(10.46)
262 Stochastic Finance: A Numeraire Approach

with the terminal condition

uX (T, x) = f X (x). (10.47)

PROOF Let us show the partial integro-differential equation for uY . Ac-

cording to Ito’s formula for the jump processes, we have
Z t
uY (t, XY (t)) = uY (0, XY (0)) + uYt (s, XY (s−)) ds
0
Z t
+ uYx (s, XY (s−)) dXY (s)
0
X h
+ uY (s, XY (s)) − uY (s, XY (s−))
0≤s≤t
i
−uYx (s, XY (s−)) ∆XY (s)
Z t
= uY (0, XY (0)) + uYt (s, XY (s−)) ds
0
Z tZ ∞
+ ux (s, XY (s−)) (ey − 1)XY (s−)
Y
0 −∞
 
· µY (dy, ds) − ν Y (dy)ds
Z tZ ∞
 Y
+ u (s, ey XY (s−)) − uY (s, XY (s−))
0 −∞

−uYx (s, XY (s−)) (ey − 1)XY (s−) µY (dy, ds)
Z t
Y
= u (0, XY (0)) + uYt (s, XY (s−)) ds
0
Z tZ ∞

+ ν Y (dy) −(ey − 1)XY (s−)uYx (s, XY (s−))
0 −∞


+ uY (s, ey XY (s−)) − uY (s, XY (s−)) ds
Z tZ ∞


+ uY (s, ey XY (s−)) − uY (s, XY (s−))
0 −∞
 
· µY (dy, ds) − ν Y (dy)ds .

We have subtracted
Z ∞
 Y 
u (s, ey XY (s−)) − uY (s, XY (s−)) ν Y (dy)ds
−∞
 Jump Models 263

from the µY (dy, ds) term in order to obtain the martingale part of the equality,
and added it back to the corresponding ds term. Since u (t, XY (t)) is a PY -
martingale, the corresponding ds term that is not a part of the martingale
µY (dy, ds) − ν Y (dy)ds must vanish. Thus we have the following identity:
Z ∞
 
uYt (t, x) + ν Y (dy) uY (t, ey x) − uY (t, x) − (ey − 1)xuYx (t, x) = 0
−∞
(10.48)
The terminal condition is given by

uY (T, x) = f Y (x). (10.49)

The Lévy Jump Model is in general incomplete with the exception of a

single jump size model.

References and Further Reading

Jump models of the price date back to Merton (1976) who assumed normal
distribution of the jump sizes. Kou (2002) suggested exponential distribution
of the jump sizes, leading to a better fit with empirical data. Carr et al.
(2002) suggested an even more general class of jump models that also covers
exponential distribution of the jump sizes, which also covers processes with
finite and infinite activity. Symmetry analysis of the price processes driven
by jumps appears in Jamshidian (2007) and in Fajardo and Mordecki (2006).
A standard reference for stochastic calculus with jump processes is Protter
(2005). The reader can also refer to the monographs on financial modeling
with jump processes by Schoutens (2003), and Cont and Tankov (2003). Cont
and Voltchkova (2005b,a) study numerical implementations of partial integro-
differential equations that arise in pricing of financial contracts. Carr et al.
(2001) discuss hedging in incomplete markets.

Exercises
10.1 Show that the following processes are martingales.
(a) (N (t) − λt)2 − λt.
(b) exp (γN (t) − (eγ − 1)λt) .
264 Stochastic Finance: A Numeraire Approach

10.2 Assume that the price process XY (t) follows

dXY (t) = (eγ − 1) · XY (t−)d(N (t) − λY t).

Use Ito’s formula for the jump processes to determine

d log(XY (t)).

10.3 Use Ito’s formula for the jump processes to determine

d[XY (t)]α ,

where XY (t) is a geometric Poisson process.

10.4 Consider a geometric Poisson model for two no-arbitrage assets X and
Y , where the price follows

dXY (t) = (eγ − 1) · XY (t−)d(N (t) − λY t).

(a) Compute the price of a contract that pays off VT = I(N (T ) = 0) · YT (in
terms of a reference asset Y , and the probability measure PY determined
by λY ). This contract gives its holder the asset Y if there is no jump in
the price at time T .
(b) Double check the result from (a) by computing the price of VX (T ) using
the reference asset X. Use the relationship between λY and λX .
(c) Compute the hedge of the contract: ∆X (t) and ∆Y (t).
(d) Compute the price of a contract that pays off VT = I(N (T ) = k) · YT for
general k ∈ N together with ∆X (t) and ∆Y (t).

10.5 Consider the geometric Poisson model with the price evolution

dXY (t) = (eγ − 1)XY (t−)d(N (t) − λY t).

(a) Find the price of a contract that pays off a unit of an asset X when
N (T ) = k.
(b) Find the price of a contract that pays off a unit of an asset Y when
N (T ) = k.
(c) Which of the two contracts is more valuable? Find a condition for k for
which the contract that pays off the asset X is more valuable (inequality
in terms of γ, λY , T, and XY (0)).

10.6 Show that uY (t, x) = 1 and uY (t, x) = x are solutions of the integro-
differential equation (10.44). An analogous result is valid for the integro-
differential equation (10.46). What are the financial contracts that correspond
to these solutions? How would you hedge these contracts?
 Jump Models 265

10.7 Consider the average asset A defined in (9.1).

(a) Find
dAY (t)
in the geometric Poisson process model.

(b) Let uY (t, XY (t), AY (t)) be the price of the Asian option with payoff
f Y (XY (T ), AY (T )). Using the Ito’s formula for jumps (with 2 spa-
tial variables), find the integro-differential equation for uY (t, x, y). Show
that uY (t, x, y) = 1, uY (t, x, y) = x, and uY (t, x, y) = y are solutions
of this equation. Also check that when the contract does not depend on
A, the price function uY (t, x, y) does not depend on the variable x, and
the integro-differential equation simplifies to (10.44).
(c) Find
dAX (t)
and the integro-differential equation for uX (t, x, y). Show that
uX (t, x, y) = 1, uX (t, x, y) = x, and uX (t, x, y) = y are solutions of
this equation. Note that when the payoff does not depend on the asset
Y , the integro-differential equation does not depend on the variable x,
and the equation simplifies.
The price processes YA and XA do not preserve the stationarity property, and
thus the equation for uA cannot be written in terms of a compensator ν A as
the compensator does not exist.
Appendix A
Elements of Probability Theory

A.1 Probability, Random Variables

This section summarizes some basic concepts in probability theory. Prob-
ability is a triplet (Ω, F , P), where Ω is a set of outcomes (also known as
sample space), F is a set of events, and P is the probability measure. We
denote by ω individual outcomes from the set Ω. The set of events F includes
combinations of outcomes, and thus each event from F is a subset of Ω. More-
over, the set of events F includes the set of all outcomes Ω, and is closed under
complements and countable unions. In mathematical notation
∞
[
A ∈ F ⇒ (Ω \ A) ∈ F , Ai ∈ F ⇒ Ai ∈ F .
i=1

A probability measure assigns a value on the interval [0, 1] to events in F with

the following restrictions: the probability of the set of all outcomes Ω isSone,
and when the sets Ai are disjoint (Ai ∩ Aj = ∅, i 6= j), probability of i Ai
is the sum of probabilities of individual events:
∞
[ ∞
X
P(Ω) = 1, P( Ai ) = P(Ai ).
i=1 i=1

Example A.1
Consider a coin toss. The two possible outcomes are “Head” and “Tail,”
so Ω = {Head, Tail}. The set of events is F = {∅, {Head}, {Tail}, Ω}. The
probability measure P is determined by the value of P({Head}) = p ∈ [0, 1],
because P({Tail}) + P({Head}) = P(Ω) = 1, and thus P({Tail}) = 1 − p. Note
that we may have different probability measures determined by the value p,
that comes with the same set of outcomes Ω and the same set of events F .

What is the purpose of assigning probability to events F rather than as-

signing the probability directly to individual outcomes ω? We have seen in
Example A.1 that the probability of the individual outcomes determines the
probability of all events, but this does not work in general. One example of
why it is not sufficient to assign probability just to individual outcomes is

267
268 Stochastic Finance: A Numeraire Approach

a situation when the set of outcomes is uncountable (such as the real line).
In this case the probability of each individual outcome may be zero, and it
would be impossible to reconstruct the probability of events from outcomes
with zero probability. For that one has to start with events that contain more
than a critical fraction of outcomes, such as intervals in the case of continuous
distributions on the real line.

Note that the set of outcomes may have nonnumerical values, such as in the
case of the coin toss when the set of outcomes is Ω = {Head, Tail}. When the
outcomes are assigned a number X, then we say X is a random variable.
Formally, a random variable is a mapping

X :Ω→R

with the property that for each Borel set B,

{X ∈ B} = {ω ∈ Ω; X(ω) ∈ B}

is an event from the set of events F . This assures that the probability of the
event {X ∈ B} is well defined. Every Borel set can be obtained from closed
intervals [a, b] by taking complements, or countable unions of intersections of
these sets.

We call
F (x) = P(X ≤ x),
a cumulative distribution function. The cumulative distribution function
already determines P(X ∈ B) for all Borel sets, in which case we talk about
the distribution of a random variable. The derivative f of the cumulative
distribution function, if it exists, is known as a density

f (x)dx = dF (x).

In the case when the cumulative distribution function has jumps (in the case
of discrete distributions), the density function does not exist in the classical
sense as a function. However, it exists in terms of linear functionals, measures
that operate on a space of functions. For instance the Dirac delta δ(c) (x) is
defined by the following relationship
Z ∞
g(x)δ(c) (x)dx = g(c). (A.1)
−∞

The Dirac delta itself is not a function, but one can view it as a graph that
has zero value outside c, but when evaluated at c it behaves like an impulse
of a size 1. Indeed, we have
Z c+ǫ
δ(c) (x)dx = 1
c−ǫ
 Elements of Probability Theory 269

for any ǫ > 0.

The expectation of a random variable is defined as

Z Z ∞ Z ∞
EX = X(ω)dP(ω) = xdF (x) = xf (x)dx.
Ω −∞ −∞

Example A.2 Uniform Distribution

The discrete uniform distribution on the interval [0, 1] is determined by the
following probabilities
k 1
P(X = n) = n+1 , k = 0, 1, . . . , n.

This can be viewed as a density using the Dirac deltas:

1
f (x) = n+1 · δ(k/n) (x).

The expectation of this random variable is given by

Z Z ∞ n
X
k 1
EX = X(ω)dP(ω) = x · f (x)dx = n · n+1 = 12 .
Ω −∞ k=0

The continuous uniform distribution has density

f (x) = I([0, 1])(x);

i.e., it is equal to 1 on the interval [0, 1], and 0 everywhere else. The expectation
is 12 , which is the same as in the discrete case.

Example A.3 Binomial Distribution

The binomial distribution B(n, p) is given by the following probabilities:
 
n k
P(X = k) = p · (1 − p)n−k , for k = 0, 1, . . . , n.
k

In terms of the density function, this can be expressed as

 
n k
f (x) = p · (1 − p)n−k δ(k) (x).
k

Its expectation is given by

Z Z ∞ n
X  
n k
EX = X(ω)dP(ω) = x · f (x)dx = k· p · (1 − p)n−k = n · p.
Ω −∞ k
k=0
270 Stochastic Finance: A Numeraire Approach

Example A.4 Poisson Distribution

The Poisson distribution is given by the following probabilities:

λk
P(X = k) = e−λ · ,
k!
where λ > 0 is a parameter. We can write the density of Poisson distribution
as
λk
f (x) = e−λ · δ(k) (x).
k!
Its expectation is given by
Z Z ∞ ∞
X λk
EX = X(ω)dP(ω) = x · f (x)dx = k · e−λ · = λ.
Ω −∞ k!
k=0

Example A.5 Normal Distribution

The normal distribution N (µ, σ 2 ) has the following density:
 
1 (x − µ)2
f (x) = √ · exp − .
σ 2π 2σ 2

Its expectation is given by

Z Z ∞
EX = X(ω)dP(ω) = x · f (x)dx
Ω −∞
Z ∞  
1 (x − µ)2
= x · √ · exp − = µ.
−∞ σ 2π 2σ 2

Example A.6 Exponential Distribution

The exponential distribution has a density given by

f (x) = λe−λx I(x>0) dx.

Its expectation is equal to

Z Z ∞ Z ∞
1
EX = X(ω)dP(ω) = x · f (x)dx = x · λe−λx dx = .
Ω −∞ 0 λ
 Elements of Probability Theory 271

Example A.7 Gamma Distribution

The gamma distribution is given by the density
x
−
xα−1 e β
f (x) = I(x>0) dx.
Γ(α)β α
Its expectation is equal to
Z Z Z x
∞ ∞ −
xα−1 e β
EX = X(ω)dP(ω) = x · f (x)dx = x· dx = α · β.
Ω −∞ 0 Γ(α)β α

A.2 Conditional Expectation

A σ-algebra is a set of events that is closed under complements, countable
unions, and intersections. We denote by σ(As ) the smallest σ-algebra that
contains all sets As . Let X be a random variable. By information of a random
variable σ(X) we mean the σ-algebra generated by the sets
{ω ∈ Ω : X(ω) ∈ B}
for any Borel set B. Borel sets on a real line are open intervals, or sets that are
created by taking countable unions or intersections of the previously obtained
sets, starting from the open intervals.

DEFINITION A.1 Conditional Expectation

The conditional expectation of a random variable X with respect to the in-
formation set F , E[X|F ], is a random variable that satisfies the following
properties:
1. σ(E[X|F ]) ⊆ F ,
2. for each A ∈ F , E[E[X|F ] · IA ] = E[X · IA ].

Property 1 of the conditional expectation means that the information of the

conditional expectation as a random variable is not larger than the informa-
tion set F . Property 2 ensures that the conditional expectation has the same
average as the original random variable on sets that belong to F .

Example A.8
Consider a set of outcomes that corresponds to 2 coin tosses: Ω =
{HH, HT, T H, T T }. Let us assume that the coin toss is fair, and all the out-
272 Stochastic Finance: A Numeraire Approach

comes have the same probability 14 : P(HH) = P(HT ) = P(T H) = P(T T ) = 14 .

Define a random variable X with the following property:
X = (1st coin resulted in H) × 1 + (2nd coin toss resulted in H) × 2,
or in other words,

X(HH) = 3, X(HT ) = 1, X(T H) = 2, X(T T ) = 0.

Let us consider three times: t = 0 before any coin toss, t = 1 after the first coin
toss but before the second coin toss, and t = 2 after the second coin toss. At
time t = 0, there is no information available, and thus the corresponding infor-
mation set is trivially F0 = {∅, Ω}. At time t = 1, it is possible to distinguish
between events {HH, HT } and {T H, T T }, but it is not possible to distinguish
between {HH} and {HT }, or between {T H} and {T T } since the second coin
toss has not yet been observed. Thus the information set F1 contains events
F1 = {∅, Ω, {HH, HT }, {T H, T T }}. At time t = 2, we have a complete infor-
mation set that can distinguish the events {HH}, {HT }, {T H}, {T T }. Thus
the information set is given by F2 = σ({HH}, {HT }, {T H}, {T T }).

Let us determine the conditional expectations E[X|F0 ](ω), E[X|F1 ](ω),

E[X|F2 ](ω). Let us start with E[X|F0 ](ω). From Property 1 of the condi-
tional expectation, we must have

σ(E[X|F0 ]) ⊆ F0 = {∅, Ω}.

This means E[X|F0 ](ω) must be a constant for all ω ∈ Ω. From Property 2 of
the conditional expectation, using A = Ω and IΩ = 1, we have

E[E[X|F0 ]] = E[X].

Since E[X|F0 ] must be a constant, we conclude that

1 1 1
E[X|F0 ] = E[X] = (3 · 4 +1· 4 +2· 4 + 0 · 41 ) = 32 .

Conditional expectation E[X|F1 ](ω) must satisfy Property 1, namely

σ(E[X|F1 ]) ⊆ F1 = {∅, Ω, {HH, HT }, {T H, T T }}.

The information set is richer than in the case of F0 , which means that
the conditional expectation can distinguish between events {HH, HT } and
{T H, T T }, so its value may differ on them. However, the conditional expec-
tation does not distinguish events {HH} and {HT }, or {T H} and {T T },
meaning that

E[X|F1 ](HH) = E[X|F1 ](HT ) = C1 ; E[X|F1 ](T H) = E[X|F1 ](T T ) = C2 .

 Elements of Probability Theory 273

This ensures Property 1. Let us compute C1 and C2 using Property 2. Let us

consider A = {HH, HT }. We get
1 1
E[E[X|F1 ] · IA ] = E[X|F1 ](HH) · 1 · 4 + E[X|F1 ](HT ) · 1 · 4
1 1 C1
+ E[X|F1 ](T H) · 0 · 4 + E[X|F1 ](T T ) · 0 · 4 = 2 .

1 1 1 1
E[X · IA ] = X(HH) · 1 · 4 + X(HT ) · 1 · 4 + X(T H) · 0 · 4 + X(T T ) · 0 · 4
1 1 1 1
=3·1· 4 +1·1· 4 +2·0· 4 + 0 · 0 · 4 = 1.

Since
E[E[X|F1 ] · IA ] = E[X · IA ],
C1
we must have 2 = 1, or in other words, C1 = 2. We can similarly determine
C2 by considering A = {T H, T T } and using Property 2:
1 1
E[E[X|F1 ] · IA ] = E[X|F1 ](HH) · 0 · 4 + E[X|F1 ](HT ) · 0 · 4
1 1 C2
+ E[X|F1 ](T H) · 1 · 4 + E[X|F1 ](T T ) · 1 · 4 = 2 .

1 1 1 1
E[X · IA ] = X(HH) · 0 · 4 + X(HT ) · 0 · 4 + X(T H) · 1 · 4 + X(T T ) · 1 · 4
1 1 1 1
= 3·0· 4 +1·0· 4 +2·1· 4 +0·1· 4 = 12 .
Since
E[E[X|F1 ] · IA ] = E[X · IA ],
C2 1
we must have 2 = 2, or in other words, C2 = 1. The values of E[X|F1 ] are
listed in Table A.1.

As for E[X|F2 ], from Property 1 we get

σ(E[X|F2 ]) ⊆ F2 = σ({HH}, {HT }, {T H}, {T T }),
meaning that the conditional expectation can distinguish between all out-
comes ω from Ω. We can write
E[E[X|F2 ] · I(ω) ] = E[X|F2 ](ω) · P(ω),
and
E[X · I(ω) ] = X(ω) · P(ω).
From Property 2, we must have
E[X|F2 ](ω) · P(ω) = X(ω) · P(ω),
or in other words
E[X|F2 ] = X,
meaning that the conditional expectation is the original random variable X
itself.
274 Stochastic Finance: A Numeraire Approach
TABLE A.1: Conditional expectation.
ω X(ω) E[X|F0 ](ω) E[X|F1 ](ω) E[X|F2 ](ω)
3
HH 3 2 2 3
3
HT 1 2 2 1
3
TH 2 2 1 2
3
TT 0 2 1 0

A.2.1 Some Properties of Conditional Expectation

Let us mention some important properties of the conditional expectation.
Linearity:
E[aX + bY |F ] = aE[X|F ] + bE[Y |F ].

Independence: If X is integrable and independent of F :

E[X|F ] = E[X].

Taking out what is known: If σ(Y ) ⊆ F

E[X · Y |F ] = Y · E[X|F ].

The special situation when X = 1 reads as

E[Y |F ] = Y.

Tower property: If G ⊆ F

E[E[X|F ]|G] = E[X|G].

Projection: The random variables X − E[X|F ] and E[X|F ] have zero corre-
lation:
E [[X − E[X|F ]] · E[X|F ]] = 0.

A.3 Martingales
A process M(t) is called a martingale if

E[M(t)|Fs ] = M(s). (A.2)

 Elements of Probability Theory 275

Let us introduce the following notation. When the filtration Ft represents

information up to time t, we will simply write

Et [V ] := E[V |Ft ].

The martingale property can be rewritten as

Es [M(t)] = M(s).

Let us show several processes that are martingales. Let X(i) be a random
variable (
1 p = 12 ,
X(i) =
−1 1 − p = 21 .
Pn
Then S(n) = i=1 X(i) is a martingale, and S 2 (n) − n is a martingale. In
order to prove the martingale property, it is enough to show that

En [M(n + 1)] = M(n).

In particular,

En [S(n + 1)] = En [S(n) + X(n + 1)] = S(n) + E[X(n + 1)] = S(n).

The second equality follows from the fact that S(n) is known at time n, and
X(n + 1) is independent of Fn , and thus the conditional expectation becomes
a simple expectation. As for the second statement, we have
2
En [S 2 (n + 1) − (n + 1)] = En [(S(n) + X(n + 1)) ] − (n + 1)
= En [S 2 (n) + 2S(n) · X(n + 1) + X 2 (n + 1)] − (n + 1)
= S 2 (n) + 2S(n) · E[X(n + 1)] + E[X 2 (n + 1)] − (n + 1)
= S 2 (n) + 1 − n − 1 = S 2 (n) − n.

The process S(n) is known as a discrete random walk. One can think
about this process as a basic market noise which comes from playing a fair
game with outcomes 1 and −1. At each time step, the player bets one dollar
and has a fifty percent chance of winning an extra dollar, and a fifty percent
chance of losing the stake. The process S(n) measures the player’s winning or
losing balance after n steps.

Martingales represent fair games. The First Fundamental Theorem of Asset

Pricing (Chapter 1) states that prices must be martingales (under a certain
measure associated with the reference asset) in order to prevent an opportu-
nity for risk-free profit. Intuitively, the market cannot be beaten by playing
such a game. In particular, the expected value of the martingale stays the
same, so we have
E[S(n)] = S(0) = 0.
276 Stochastic Finance: A Numeraire Approach

A natural question is whether one can beat the market (or the coin tossing
game in this case) by taking a more sophisticated strategy. The player’s con-
trols are twofold: the size of the bet, and when to quit playing the game. Let
us first check the varying size of the bet. The player bets ∆(n) at time n, so
the resulting profit or loss P from betting satisfies
n
X
P (n) = ∆(i) · X(i).
i=1

The process P (n) is still a martingale

En [P (n + 1)] = En [P (n) + ∆(n + 1) · X(n + 1)]

= P (n) + ∆(n + 1) · E[X(n + 1)] = P (n).

The crucial part is that the size of the bet, ∆(n + 1), is known at time n
(before the outcome of the bet X(n + 1)), so that it can be taken out of the
conditional expectation. This is always the case in real markets or in casino
games (think about roulette). If ∆(n + 1) could be set after observing the
outcome of X(n + 1), this would lead to a possibility of a risk-free P
profit. One
can takePfor instance ∆(n + 1) = X(n + 1), in which case P (n) = ni=1 ∆(i) ·
n
X(i) = i=1 X 2 (i) = n.

Example A.9 Doubling strategy

A particular case of a betting strategy is a doubling strategy. The player’s
initial bet is ∆(1) = 1. If he wins, he would collect one dollar (two dollars
won minus one dollar bet), and quit the game. If he loses, he would double his
previous bet (∆(2) = 2). If he wins in the second round, he would collect one
dollar (four dollars won minus three dollars bet). If he loses again, he would
double his bets, and continue in this procedure. After n steps, the probability
that he won by that time is 1 − 21n . The winning is always one dollar:

2n − (1 + 2 + · · · + 2n−1 ) = 1.

The chance that he has not won by time n is 21n , in which case the total loss
is equal to
−(1 + 2 + · · · + 2n−1 ) = −(2n − 1).
Thus we have
(
1
1 p=1− 2n ,
P (n) =
−(2n − 1) 1 − p = 21n .

While limn→∞ P(P (n) = 1) = 1, we still have E[P (n)] = 0. The probability of
winning is indeed large and converges to one, but one still does not beat the
 Elements of Probability Theory 277

expected value of the game. The reason is the unlikely event that one would
not win a single time in n trials, but the loss corresponding to this case is
astronomical. This strategy would eventually lead to a bankruptcy.

The second control over the game from the perspective of a player is to
choose a time when to stop. Let us introduce the following concept.

DEFINITION A.2 A stopping time τ is a random time with the prop-

erty that
{τ ≤ t} ∈ Ft . (A.3)

The condition {τ ≤ t} ∈ Ft means that at every moment t, one is able to

tell if the event {τ ≤ t} has happened or not. Being unsure about this event
is not an option. The simplest example of a stopping time is when τ = N ,
where N is a constant. Practical examples of stopping times include the time
when a price of a stock crosses a certain level, time of the arrival of a train,
or the time of getting married. These examples illustrate that one can tell if
the event happened or not. For instance, one can always tell in principle if he
or she is married or not married, although in this case one may not be certain
about it, like after a good party in Las Vegas. This was the case of a basket-
ball player Dennis Rodman who got married to Carmen Electra in the state
of intoxication, and had no recollection of the event (this marriage was later
annulled). This is also a good illustration of asymmetrical information since
the time of his marriage was not a stopping time with respect to Rodman’s
information set, but it was a stopping time with respect to the information
set of the state of Nevada (they could still tell if he was married or not). A
simpler example of a time that is not a stopping time is for instance a half-life
of the event. The information about the exact time of a half-life will be known
only after observing the event in the future, and one is not sure if the half-life
time already happened or not.

Stopping times play an important role in finance since one can construct
contracts that pay off at such stopping times. For instance, barrier options
become dead or alive depending on the price hitting a certain level, where
the time of hitting is a stopping time. Contracts based on the stopping times
are at least in legal terms well defined, as opposed to contracts where one is
uncertain if the contractual party already qualifies for a payoff or not.

Let us return to the question if one can beat the casino (at least in a fair
game) by adopting a strategy that involves a stopping time. Take for instance
the first time τ when one is winning a dollar, defined by

τ = min{n ≥ 0 : S(n) = 1}.

278 Stochastic Finance: A Numeraire Approach

Clearly, P (τ ) = 1, and we have that

1 = E[S(τ )] 6= S(0) = 0.
Moreover, P(τ < ∞) = 1, which is not obvious, but still true. This means that
one can potentially beat a fair game by waiting for the first time when his
profit from the game is positive. However, this win does not come easily; it
turns out that E[τ ] = ∞, so the expected profit per expected time to reach it
is still zero. The random variable τ is finite, but it has an infinite expectation.

In more practical situations, such as in the case of a bounded stopping

time, one cannot beat a fair game. For this result to hold, the value of the
game cannot take large values on small probability events. More precisely, the
following theorem holds.

THEOREM A.1 Optional Sampling Theorem

Let M be a martingale, and τ be a stopping time with P(τ < ∞) = 1. If
E[|M(τ )|] < ∞, and
lim E[|M(n)|I(τ > n)] = 0, (A.4)
n→∞

then
E[M(τ )] = E[M(0)]. (A.5)

The condition (A.4) ensures that the martingale does not dissipate in
√ space.
For instance, in the case of a discrete random walk S(n), P(τ > n) ≈ n, and
E[|S(n)|I(τ > n)] does not go to zero.

Example A.10 Gambler’s ruin problem

Suppose that a player starts with a dollars (S(0) = a), and plays a fair game
so that his wealth follows a discrete random walk. He stops in two situations:
he either loses all his money, so that S(n) = 0, or he wins enough to quit, so
that S(n) = N . The stopping time is given by
τ = min{n ≥ 0 : S(n) = 0 or S(n) = N }.
Since S(n) is bounded (it stays between 0 and N ), the conditions of the
Optional Sampling Theorem are satisfied, and we have
E[S(τ )] = E[S(0)] = a.
In this case,
E[S(τ )] = N · P(S(τ ) = N ),
and therefore
a
P(S(τ ) = N ) = N.
 Elements of Probability Theory 279

A.4 Brownian Motion

This section introduces a Brownian motion (also known as a Wiener Pro-
cess) and shows its basic properties.

DEFINITION A.3 Brownian Motion

Brownian motion W (t) is a process with the following properties:

1. W (t) − W (s) is independent of Fs , s < t,

2. W (t) − W (s) has a normal N (0, t − s) distribution,

3. W (t) is a continuous process.

It is not obvious that a process with these properties exists, but the answer
is indeed affirmative. Brownian motion can be constructed as a limit of a
discrete random walk in the following sense. Let S(n) be a discrete random
walk, and define a process W (n) by
1
W (n) (t) = √ S(nt)
n

for t of the form nk , where k is an integer. If t is not in this form, we define

W (n) (t) as a linear interpolation between its values at the nearest points nk .
When t > s are equal to nk for integer k, we have that W (n) (t) − W (n) (s) is
independent of the information set Fs . Furthermore,
 
(n) (n) 1 1
E[W (t) − W (s)] = E √ S(nt) − √ S(ns) = 0,
n n

and
 
Var W (n) (t) − W (n) (s) = E[W (n) (t) − W (n) (s)]2
 2 " nt
#2
1 1 1 X
= E √ S(nt) − √ S(ns) = E √ · X(k)
n n n
k=ns
nt nt
1 X   1 X 1
= E X(k)2 = 1 = · [nt − ns] = t − s.
n n n
k=ns k=ns

When n → ∞, the increments remain independent of Fs , they will be nor-

mally distributed because of the Central Limit Theorem with the mean 0 and
the variance t − s, and the limiting process will be continuous.
280 Stochastic Finance: A Numeraire Approach

REMARK A.1 Martingales of Brownian motion

1 2
The following processes are martingales: W (t), W (t)2 −t, eσW (t)− 2 σ t . This
can be seen from the following relationships.

Es [W (t)] = Es [(W (t) − W (s)) + W (s)] = Es [W (t) − W (s)] + W (s) = W (s).

Es [W (t)2 − t] = Es [(W (t) − W (s) + W (s))2 − t]

= Es [(W (t) − W (s))2 − 2 · W (s) · (W (t) − W (s)) + W (s)2 − t]
= E[(W (t) − W (s))2 ] − 2 · W (s) · E[W (t) − W (s)] + W (s)2 − t
= (t − s) − 2 · W (s) · 0 + W (s)2 − t = W (s)2 − s.

   
1 1
σW (t)− 2 σ2 t σ((W (t)−W (s))+W (s))− 2 σ2 t
Es e = Es e
1 2 h i
= eσW (s)− 2 σ t · E eσ(W (t)−W (s))
1 2 1 2 1 2
= eσW (s)− 2 σ t · e 2 σ (t−s)
= eσW (s)− 2 σ s .

REMARK A.2 Hitting time of Brownian motion

Let us denote by τa the first time when a Brownian motion reaches a level
a > 0:
τa = inf{t ≥ 0 : W (t) = a}.
Note that we can write

P(τa ≤ t) = P(τa ≤ t, W (t) ≥ a) + P(τa ≤ t, W (t) < a)

= 2P(τa ≤ t, W (t) ≥ a) = 2P(W (t) ≥ a)
Z ∞ y2
√1 − 2t
= 2πt
· e dy.
a

The event W (t) ≥ a already implies the event τa ≤ t. The equality P(τa ≤
t, W (t) ≥ a) = P(τa ≤ t, W (t) < a) follows from the so-called reflection
principle. Once Brownian motion W reaches a, it is equally likely that the
process will end up above the level a at a fixed future time T , or end up below
a because of the symmetry of distribution of Brownian motion. We can obtain
the density of the hitting time by taking a derivative of P(τa ≤ t) with respect
to t
∂P(τa ≤ t) a a2
f (t) = = √ · e− 2t .
∂t t 2πt
 Elements of Probability Theory 281

REMARK A.3 Quadratic variation of Brownian motion

The quadratic variation of a process X up to time T is defined as
n
X
[X , X ](T ) = lim [X (ti ) − X (ti−1 )]2 , (A.6)
kΠk→0
i=1

where Π = {t0 = 0, t1 , . . . , tn = T } is a partition of the interval [0, T ], and

kΠk = max1≤i≤n |ti − ti−1 |. A Brownian motion has the quadratic variation
equal to T
[W, W ](T ) = T. (A.7)

This is a remarkable result since each of the factors (W (ti ) − W (ti−1 )2 in

the definition of quadratic variation is random, and thus [W, W ](T ) is also a
random variable. But the resulting value of the quadratic variation, T , has
zero variance. It means that each path of a Brownian motion W has the same
quadratic variation.

We can prove the result by showing that E[W, W ](T ) = T , and

Var([W, W ](T )) = 0. Define
n
X
QΠ = [W (ti ) − W (ti−1 )]2
i=1

the sampled quadratic variation that corresponds to the partition Π so that

limkΠk→0 QΠ = [W, W ](T ). First,
" n # n
X X
2
E[QΠ ] = E [W (ti ) − W (ti−1 )] = [ti − ti−1 ] = T.
i=1 i=1

From the independence of Brownian increments,

n
X

Var(QΠ ) = Var [W (ti ) − W (ti−1 )]2 ,
i=1

so the variance of QΠ is the sum of the variances of the squared increments

of a Brownian motion. Since

h
2 i
Var [W (ti ) − W (ti−1 )]2 = E [W (ti ) − W (ti−1 )]2 − (ti − ti−1 )
= E [W (ti ) − W (ti−1 )]4 − 2(ti − ti−1 )E [W (ti ) − W (ti−1 )]2 + (ti − ti−1 )2
= 3(ti − ti−1 )2 − 2(ti − ti−1 )2 + (ti − ti−1 )2 = 2(ti − ti−1 )2 ,

one finds
n
X n
X
Var(QΠ ) = 2(ti − ti−1 )2 ≤ 2 kΠk · (ti − ti−1 ) = 2kΠk · T → 0.
i=1 i=1
282 Stochastic Finance: A Numeraire Approach

The relationship (A.7) can be also rewritten in infinitesimal form as

(dW (t))2 = dt. (A.8)

Note that the quadratic variation of a linear function t is zero, and the cross
variation of W and t is also zero. To show these results, note that
n
X
[T, T ](T ) = lim (ti − ti−1 )2 ≤ lim kΠk · T = 0,
kΠk→0 kΠk→0
i=1

and
n
X
[W, T ](T ) = lim (W (ti ) − W (ti−1 ))(ti − ti−1 )
kΠk→0
i=1
n
X
≤ lim max |W (ti ) − W (ti−1 )| (ti − ti−1 ) = 0.
kΠk→0 1≤i≤n
i=1

The term max1≤i≤n |W (ti ) − W (ti−1 )| converges to zero because of the con-
tinuity of a Brownian motion W . The two relationships can be written in the
infinitesimal form as
dt · dt = 0, dW (t) · dt = 0. (A.9)

The quadratic variation of a function f that has a continuous derivative is

zero. From the mean value theorem, there exists t∗i ∈ [ti−1 , ti ] such that
f (ti ) − f (ti−1 ) = f ′ (t∗i ) · (ti − ti−1 ).
Therefore
n
X
[f, f ](T ) = lim (f (ti ) − f (ti−1 ))2
kΠk→0
i=1
n
X
= lim (f ′ (t∗i ))2 · (ti − ti−1 )2
kΠk→0
i=1
n
X
≤ lim kΠk · (f ′ (t∗i ))2 · (ti − ti−1 )
kΠk→0
i=1
Z T
= lim kΠk · (f ′ (t))2 dt = 0.
kΠk→0 0

The fact that the quadratic variation of a Brownian motion is equal to T ,

and not to zero, implies that a Brownian motion does not have a continuous
derivative. It turns out that it does not have any derivative at any point, so
each Brownian path is “edgy.”
 Elements of Probability Theory 283

A.5 Stochastic Integration

In analogy to a discrete random walk, Brownian motion can represent the
profit or loss resulting from a fair game played in continuous time. As such,
it is a good model for market noise. In the discrete case, we considered the
situation when the player controls the size of the bet ∆(n) at time n, and his
change of wealth can be represented as
n
X
P (n) = ∆(i) · X(i).
i=1

In the discrete case, the random variable X(i) which takes values 1 or −1
represents the market noise. In the continuous case, the increment of the
Brownian motion W (ti ) − W (ti−1 ) represents the market noise. If one also
controls the size of the position in the underlying noise, the total profit or loss
resulting from this betting strategy, or trading in a market noise, is given by
n
X
I(T ) = ∆(ti−1 ) · (W (ti ) − W (ti−1 )). (A.10)
i=1

Note that the position in the underlying noise ∆(ti−1 ) should be set before
having any information about the increment W (ti ) − W (ti−1 ), so the position
∆ must be set at or before time ti−1 . Otherwise one would be able to make a
risk-free profit. Mathematically
Pn one can also consider other times t∗i ∈ [ti−1 , ti ]
¯ ∗
and define I(T ) = i=1 ∆(ti ) · (W (ti ) − W (ti−1 )), but only (A.10) makes
financial sense. If we allow for continuous changes of the position ∆(t) the
above sum will in the limit become an integral
Z T
I(T ) = ∆(t)dW (t). (A.11)
0

We can also rewrite the above relationship in a differential form as

dI(t) = ∆(t)dW (t). (A.12)

RT
It turns out that in contrast to an ordinary integral 0 ∆(t)dt, it does matter
which time t∗i ∈ [ti−1 , ti ] is taken in the evaluation of the integrand ∆(t∗i ).
Different choices of t∗i lead to different stochastic integrals. The reason is that
Brownian motion is not differentiable, so we cannot write dW (t) = W ′ (t)dt,
as there is no such thing as W ′ (t). Furthermore,
Z T n
X
|dW (t)| = lim |W (ti ) − W (ti−1 )| = ∞,
0 n→∞
i=1
284 Stochastic Finance: A Numeraire Approach

so Brownian motion takes an infinite path between any two time points. There-
fore it is not surprising that the choice of time t∗i in the evaluation of the
integrand ∆(t∗i ) makes a difference. When the integrand is evaluated at the
left end of the interval ti−1 , it is called Ito’s integral. It is the only stochastic
integral based on Brownian motion that makes financial sense.

Let us list some important properties of the Ito’s integral. We will prove
the listed properties for the integrands that are constant on a finite number
of intervals. The properties are valid for general integrands by considering an
approximating sequence of the constant integrands, and passing to the limit.

Linearity:

Z T Z T Z T
(aX(t) + bY (t))dW (t) = a X(t)dW (t) + b Y (t)dW (t).
0 0 0

Martingale property:
I(T ) is a martingale.

 
X
Et [I(T )] = Et I(t) + ∆(ti−1 ) · (W (ti ) − W (ti−1 ))
i:t≤ti−1
X
= I(t) + Et [∆(ti−1 ) · (W (ti ) − W (ti−1 ))]
i:t≤ti−1
X
= I(t) + Et Eti−1 [∆(ti−1 ) · (W (ti ) − W (ti−1 ))]
i:t≤ti−1
X  
= I(t) + Et ∆(ti−1 ) · Eti−1 [W (ti ) − W (ti−1 )]
i:t≤ti−1

= I(t).

Ito’s isometry:

"Z #
T
2 2
Var(I(T )) = E[I(T )] = E ∆(t) dt . (A.13)
0
 Elements of Probability Theory 285

This follows from

" n #2
X
2
E[I(T )] = E ∆(ti−1 ) · (W (ti ) − W (ti−1 ))
i=1
n
X
=E ∆(ti−1 )2 · (W (ti ) − W (ti−1 ))2
i=1
X
+ 2E ∆(ti−1 ) · ∆(tj−1 ) · (W (ti ) − W (ti−1 )) · (W (tj ) − W (tj−1 ))
1≤i<j≤n
n
X  
= EEti−1 ∆(ti−1 )2 · (W (ti ) − W (ti−1 ))2
i=1
X
+2 E[∆(ti−1 ) · ∆(tj−1 ) · (W (ti ) − W (ti−1 ))] · E[W (tj ) − W (tj−1 )]
1≤i<j≤n
n
" #
X  
= 2
E ∆(ti−1 ) · Eti−1 (W (ti ) − W (ti−1 ))2
i=1
n
"Z #
X   T
= E ∆(ti−1 )2 · (ti − ti−1 ) = E 2
∆(t) dt .
i=1 0

A.6 Stochastic Calculus

Stochastic calculus explains the evolution of functions of diffusion processes.
Let f (t, x) be a function for which the partial derivatives ft , fx , and fxx exist.
From Taylor’s expansion we can write

f (t + dt, W (t + dt)) − f (t, W (t)) = ft (t, W (t))dt

+fx (t, W (t)) · (W (t + dt) − W (t))
+ 21 fxx (t, W (t)) · (W (t + dt) − W (t))2
+ higher order terms.

We have already seen that (dW (t))2 = dt, and the higher order terms do not
contribute in the limit to the above expression as dt·dW (t) = 0, and (dt)2 = 0.
Therefore we have Ito’s formula
h i
df (t, W (t)) = ft (t, W (t)) + 12 fxx (t, W (t)) dt + fx (t, W (t))dW (t). (A.14)

Example A.11
Consider the process W (t)2 − t, which corresponds to the choice of f (t, x) =
x2 − t. The partial derivatives that appear in the Ito’s formula are given by
286 Stochastic Finance: A Numeraire Approach

ft (t, x) = −1, fx (t, x) = 2x, and fxx (t, x) = 2. Therefore we get

d(W (t)2 − t) = [−1 + 1

2 · 2]dt + 2W (t)dW (t) = 2W (t)dW (t).

Thus the following relationship holds

Z T
W (t)dW (t) = 21 [W (T )2 − T ].
0

Since stochastic integrals are martingales, this is another proof of the martin-
gale property of the process W (t)2 − t.

Example A.12
Consider a geometric Brownian motion S(t) = S(0) · exp(σW (t) − 21 σ 2 t),
which corresponds to the choice of f (t, x) = S(0) · exp(σx − 12 σ 2 t). The partial
derivatives of f are given by ft (t, x) = − 12 σ 2 f (t, x), fx (t, x) = σf (t, x), and
fxx (t, x) = σ 2 f (t, x). Therefore we have
 
d S(0) · exp(σW (t) − 12 σ 2 t) = [− 21 σ 2 f (t, W (t)) + 21 σ 2 f (t, W (t))]dt
+σf (t, W (t))dW (t)
= σ · S(0) · exp(σW (t) − 21 σ 2 t)dW (t),

or in other words,
dS(t) = σS(t)dW (t).
Thus we can write
Z T
S(T ) = S(0) + σS(t)dW (t),
0

showing that the geometric Brownian motion S(t) can be represented as a

stochastic integral, so in particular, it is a martingale.

Ito’s formula can be generalized for functions of a general diffusion process

as follows.

THEOREM A.2 Ito’s formula for a general diffusion process

Let X(t) be a diffusion process with dynamics

dX(t) = a(t, X(t))dt + b(t, X(t))dW (t). (A.15)

Then
h
df (t, X(t)) = ft (t, X(t)) + a(t, X(t)) · fx (t, X(t))
i
+ 21 b2 (t, X(t)) · fxx (t, X(t)) dt + b(t, X(t)) · fx (t, X(t))dW (t). (A.16)
 Elements of Probability Theory 287

PROOF We have
df (t, X(t)) = ft (t, X(t))dt + fx (t, X(t))dX(t) + 21 fxx (t, X(t))(dX(t))2
= ft (t, X(t))dt + fx (t, X(t)) · (a(t, X(t))dt + b(t, X(t))dW (t))
2
+ 1 fxx (t, X(t)) · (a(t, X(t))dt + b(t, X(t))dW (t))
h 2
= ft (t, X(t)) + a(t, X(t)) · fx (t, X(t))
i
+ 21 b2 (t, X(t)) · fxx (t, X(t)) dt + b(t, X(t)) · fx (t, X(t))dW (t).

REMARK A.4 Product rule

When X(t) and Y (t) are two diffusions, we have
d(X(t) · Y (t)) = X(t)dY (t) + Y (t)dX(t) + dX(t)dY (t).
One can show this result by applying the arguments presented above to the
function f (x, y) = x · y.

A.7 Connections with Partial Differential Equations

According to the First Fundamental Theorem of Asset Pricing, the prices of
no-arbitrage assets are martingales under the probability measure correspond-
ing to the reference asset. The Martingale Representation Theorem states that
every martingale with continuous path adapted to a filtration generated by
Brownian motion is in fact a diffusion, and it solves the following stochastic
differential equation
dM(t) = φ(t)dW (t) (A.17)
for some φ(t) that is adapted to FtW .

THEOREM A.3 Martingale Representation Theorem

A martingale M(t) with continuous paths adapted to a filtration FtW gener-
ated by a Brownian motion W admits the following representation
Z t
c c
M (t) = M (0) + φ(s)dW (s), (A.18)
0

where φ(t) is adapted to FtW .

In particular, a martingale has a zero “dt” term. Let us first identify a

class of martingales that can be written simply as a function f of time t and
288 Stochastic Finance: A Numeraire Approach

Brownian motion W (t), so the martingale is in the form M(t) = f (t, W (t)).
According to the Ito’s formula,

dM(t) = df (t, W (t)) = [ft (t, W (t)) + 21 fxx (t, W (t))]dt + fx (t, W (t))dW (t).

But since the corresponding “dt” term must be zero, the function f must
satisfy the following partial differential equation

ft (t, x) + 12 fxx (t, x) = 0. (A.19)

Example A.13
The functions f (t, x) = 1, f (t, x) = x, f (t, x) = x2 − t, or f (t, x) = exp(σx −
1 2
2 σ t) satisfy the partial differential equation (A.19). Therefore a constant 1,
Brownian motion W (t), process W (t)2 − t, or geometric Brownian motion
exp(σW (t) − 21 σ 2 t) are examples of martingales that can be expressed as a
function of time t and of a Brownian motion W (t).

We can obtain a similar connection between functions of a general diffusion

process and partial differential equations. Let X(t) be a diffusion process

dX(t) = a(t, X(t))dt + b(t, X(t))dW (t).

Consider a function f (t, X(t)) that depends on the time t and the value of
the diffusion process X(t). According to Ito’s formula we have
h
df (t, X(t)) = ft (t, X(t)) + a(t, X(t)) · fx (t, X(t))
i
+ 12 b2 (t, X(t)) · fxx (t, X(t)) dt + b(t, X(t)) · fx (t, X(t))dW (t).

Since f (t, X(t)) should be a martingale, the “dt” term must be zero, leading
to the partial differential equation

ft (t, x) + a(t, x) · fx (t, x) + 21 b2 (t, x) · fxx (t, x) = 0. (A.20)

Note that the equation (A.19) is a special case of the above partial differential
equation for the choice of a(t, x) = 0, and b(t, x) = 1.

Example A.14
When a(t, x) = 0 and b(t, x) = σx, we get

dX(t) = σX(t)dW (t),

so this case corresponds to a geometric Brownian motion. Martingales that

are functions of a geometric Brownian motion thus satisfy

ft (t, x) + 21 σ 2 x2 fxx (t, x) = 0.

 Elements of Probability Theory 289

An alternative characterization of a martingale that is a function of a dif-

fusion is via conditional expectation. Let M(T ) = g(X(T )) be the final value
of the martingale M expressed as a function of the value of X(T ). We can
define
f (t, x) = E[g(X(T ))|X(t) = x].
Note that f (t, X(t)) is a martingale, which easily follows from the tower prop-
erty

f (s, X(s)) = E[g(X(T ))|X(s)] = E[E[g(X(T ))|X(t)]|X(s)] = Es [f (t, X(t))].

Since f (t, X(t)) is a martingale, the function f (t, x) must satisfy partial dif-
ferential equation (A.20). We have proved the following theorem.

THEOREM A.4 Feynman–Kac Theorem

A function f (t, x) defined by

f (t, x) = E[g(X(T ))|X(t) = x],

where
dX(t) = a(t, X(t))dt + b(t, X(t))dW (t)
satisfies partial differential equation

ft (t, x) + a(t, x) · fx (t, x) + 21 b2 (t, x) · fxx (t, x) = 0

with the terminal condition

f (T, x) = g(x).

Example A.15
Consider the partial differential equation

ft (t, x) + µx · fx (t, x) + 21 σ 2 x2 · fxx (t, x) = 0

with the terminal condition

f (T, x) = log(x).

According to the Feynman–Kac theorem, this partial differential equation

admits a solution with stochastic representation

f (t, x) = E[log(X(T ))|X(t) = x],

where
dX(t) = µX(t)dt + σX(t)dW (t).
290 Stochastic Finance: A Numeraire Approach

The process X(t) is a geometric Brownian motion with drift µ that admits a
closed form solution

X(T ) = X(t) · exp((µ − 12 σ 2 )(T − t) + σW (T − t)),

and thus

f (t, x) = E[log(X(T ))|X(t) = x]

= E[log(X(t) · exp((µ − 21 σ 2 )(T − t) + σW (T − t)))|X(t) = x]
= log(x) + E[(µ − 21 σ 2 )(T − t) + σW (T − t)]
= log(x) + (µ − 12 σ 2 )(T − t).

One can check that f (t, x) = log(x) + (µ − 21 σ 2 )(T − t) is indeed the solution
of the partial differential equation.

References and Further Reading

A more detailed treatment of the concepts of conditional expectation and
martingales can be found in Williams (1991). The books by Lawler (2006)
and Mikosch (1999) serve as introductory references for martingales, Markov
chains, and stochastic calculus. The reader interested in a more advanced
theory of stochastic calculus should refer to Karatzas and Shreve (1991), Ok-
sendal (2007), or Revuz and Yor (2004).

Exercises
A.1 Consider a fair die toss, X which results in a value contained in the
set ({1, 2, 3, 4, 5, 6}). Let F = σ({1, 3, 5}, {2, 4, 6}). This is an information set
that corresponds to distinguishing odd/even outcomes. Determine

E[X|F ].

Compute
E[(E[X|F ] − X)2 ],

and
E[E[X|F ]].
 Elements of Probability Theory 291

A.2 Consider two consecutive coin tosses with a fair coin (P(H) = P(T ) =
1
Let the set of outcomes be given by Ω = {{HH}, {HT }, {T H}, {T T }}.
2 ).
Consider random variables S(0), S(1), and S(2) defined by

S(n) = number of heads in the first n coin tosses

for n = 0, 1, 2.
(a) Find the information sets Gn = σ(S(n)) for n = 0, 1, 2.
(b) For a random variable X given by X(HH) = 3, X(HT ) = 1, X(T H) = 2,
X(T T ) = 0, compute E[X|Gn ] for n = 0, 1, 2.
(c) Determine E[E[X|G2 ]|G1 ]. Is it equal to E[X|G1 ]? Why?
Hint: For the tower property E[E[X|G2 ]|G1 ] = E[X|G1 ] to hold in general,
we must have G1 ⊆ G2 . Check if the condition G1 ⊆ G2 holds.
hR i2
T
A.3 Determine E 0 W (t)dW (t) from Ito’s isometry.

A.4 Let W (t) be a standard Brownian motion. Using Ito’s formula, deter-
mine dW (t)4 , and compute E[W (T )4 ].

A.5 Let
S(t) = σS(t)dW (t).
Use Ito’s formula to determine
(a)
dS(t)α ,

(b)
d log(S(t)).

A.6 Define
Z(t) = exp(σW (t) − 12 σ 2 t).
Show that
Z(t) · (W (t) − σt)
is a martingale.
Hint: Apply the product rule, show that the resulting formula has a zero dt
term.
Solutions to Selected Exercises

1.2:
dSM (t) = σSM (t)dW M (t) − a(t)SM (t)dt.
1.3:
The self-financing condition is given by

(d∆Y (t)) + (d∆X (t)) · dXY (t) + (d∆X (t)) · XY (t) = 0.

(a) We have ∆Y (t) = [max0≤s≤t XY (s)], ∆X (t) = 0. Therefore

(d∆Y (t)) + (d∆X (t)) · dXY (t) + (d∆X (t)) · XY (t)

 
Y
= (d∆ (t)) = d max XY (s) 6= 0.
0≤s≤t

This portfolio is not self-financing.

h R i
t
(b) We have ∆Y (t) = 1t 0 XY (s)ds , ∆X (t) = 0. Therefore

(d∆Y (t)) + (d∆X (t)) · dXY (t) + (d∆X (t)) · XY (t)

h R i
t
= (d∆Y (t)) = d 1t 0 XY (s)ds
hR i
hR i
t t
= 0 XY (s)ds d 1t + 1t d 0 XY (s)ds
h R i
t
= − t12 0 XY (s)ds dt + 1t XY (t)dt 6= 0.

1.5:
We want to prove that

XY (t)dN (d+ ) + dXY (t)dN (d− ) − KdN (d− ) = 0.

Note that we can write

N (d+ ) = f (t, XY (t))
for a function f (t, x) given by
 √ 
f (t, x) = N σ√1T −t · log( K
x
) + 12 σ T − t ,

and
N (d− ) = g(t, XY (t))

293
294 Stochastic Finance: A Numeraire Approach

for a function g(t, x) given by

 √ 
x
g(t, x) = N σ√1T −t · log( K ) − 21 σ T − t .

Using Ito’s formula, we can write

dN (d+ ) = df (t, XY (t))

= ft (t, XY (t))dt + fx (t, XY (t))dXY (t) + 21 fxx (t, XY (t))d2 XY (t)
 
= ft (t, XY (t)) + 12 fxx (t, XY (t))σ 2 XY2 (t) dt + fx (t, XY (t))dXY (t).

The partial derivatives are given by

 
ft (t, x) = φ(d+ ) · 12 √ 1 · x
log( K ) − 1 √1
4 σ T −t ,
σ (T −t)3

fx (t, x) = φ(d+ ) · √1 · x1 ,
σ T −t
h i2 h i
fxx (t, x) = φ′ (d+ ) · √1
σ T −t
· 1
x + φ(d+ ) · − σ√1T −t · 1
x2 ,

x2
where φ(x) = √12π · e− 2 is the density of the standard normal variable.
Similarly, we can write

dN (d− ) = dg(t, XY (t))

= gt (t, XY (t))dt + gx (t, XY (t))dXY (t) + 21 gxx (t, XY (t))d2 XY (t)
 
= gt (t, XY (t)) + 12 gxx (t, XY (t))σ 2 XY2 (t) dt + gx (t, XY (t))dXY (t),

where
 
1 √ 1 x
gt (t, x) = φ(d− ) · 2 σ (T −t)3 · log( K ) + 14 σ √T1−t ,

gx (t, x) = φ(d− ) · √1 · x1 ,
σ T −t
h i2 h i
gxx (t, x) = φ′ (d− ) · √1
σ T −t
· 1
x + φ(d− ) · − σ√1T −t · 1
x2 .

Thus we can expand the expression

XY (t)dN (d+ ) + dXY (t)dN (d− ) − KdN (d− )

and get

XY (t)dN (d+ ) + dXY (t)dN (d− ) − KdN (d− ) =

 
= XY (t) (ft (t, XY (t)) + 12 fxx (t, XY (t))σ 2 XY2 (t))dt + fx (t, XY (t))dXY (t))
+ fx (t, XY (t))σ 2 XY2 (t)dt
 
− K (gt (t, XY (t)) + 21 gxx (t, XY (t))σ 2 XY2 (t))dt + gx (t, XY (t))dXY (t)) .
 Solutions to Selected Exercises 295

Let us check both the dXY (t) and dt terms in the above equality. The dXY (t)
term is
XY (t) · fx (t, XY (t)) − K · gx (t, XY (t)),
or after substitution for fx and gx ,

XY (t) · φ(d+ ) · √1 · 1
− K · φ(d− ) · √1 · 1
=
σ T −t XY (t) σ T −t XY (t)

= √1 · 1
· [XY (t) · φ(d+ ) − K · φ(d− )] .
σ T −t XY (t)

The reader should verify that we indeed have

XY (t) · φ(d+ ) − K · φ(d− ) = 0,

so the dXY (t) term is zero. The dt term is given by

 
XY (t) ft (t, XY (t)) + 12 fxx (t, XY (t))σ 2 XY2 (t) +
 
+ fx (t, XY (t))σ 2 XY2 (t) − K gt (t, XY (t)) + 21 gxx (t, XY (t))σ 2 XY2 (t) .

After substitution for the partial derivatives of f and g, we get

   
XY (t) · φ(d+ ) · 12 √ 1 3 · log XYK(t) − 14 σ √T1−t
σ (T −t)
 h i2 h i
+ XY (t) φ′ (d+ ) σ√1T −t · XY1(t) + φ(d+ ) − σ√1T −t · 1
2 (t)
XY
σ 2 XY2 (t)

+ φ(d+ ) · σ√1T −t · XY1(t) σ 2 XY2 (t)

   
− K · φ(d− ) · 21 √ 1 3 · log XYK(t) + 41 σ √ 1 T −t
σ (T −t)
 h i2 h i
′
− K · φ (d− ) · √1
σ T −t
· 1
XY (t) + φ(d− ) · − σ√1T −t · 1
2 (t)
XY
σ 2 XY2 (t).

Note that since XY (t) · φ(d+ ) − K · φ(d− ) = 0, two pairs of these terms in the
above expression cancel out. More specifically,
 
1 √ 1 XY (t)
XY (t) · φ(d+ ) · 2 σ (T −t)3 · log K
 
1 √ 1 XY (t)
− K · φ(d− ) · 2 σ (T −t)3 · log K = 0,

and
h i
XY (t) · φ(d+ ) · − σ√1T −t · 2
XY
1
σ 2 XY2 (t)
(t)
h i
− K · φ(d− ) · − σ√1T −t · 1
2 (t)
XY
σ 2 XY2 (t) = 0.
296 Stochastic Finance: A Numeraire Approach

This slightly simplifies the dt term, which after additional substitution be-
comes φ′ (x) = −x · φ(x),
h i
XY (t) · φ(d+ ) · − 41 σ √T1−t − XY (t) · d+ · φ(d+ ) · T 1−t +
h i
+ φ(d+ ) · √T1−t · σXY (t) − K · φ(d− ) · 14 σ √T1−t + K · d− · φ(d− ) · T 1−t .

Since K · φ(d− ) = XY (t) · φ(d+ ), we can replace these terms in the above
expression to get
h i
XY (t) · φ(d+ ) · √Tσ−t − 41 − σ√dT+−t + 1 − 41 + σ√dT−−t .

But since − σ√dT+−t + σ√dT−−t = − 21 , the above term sums to zero. Thus we
proved that both dt and dXY (t) terms of XY (t)dN (d+ ) + dXY (t)dN (d− ) −
KdN (d− ) are zero, and therefore the trading strategy is self-financing.

2.1:
(a)
VY (0) = EY [VY (1)] = EY [I(ω = H)] = PY (H),
1−d
V0 = PY (H) · Y0 = · Y0 .
u−d

(b) The hedging portfolio has the form

P0 = ∆X (0) · X0 + ∆Y (0) · Y0 ,
where
VY (1, H) − VY (1, T ) 1−0 YX (0)
∆X (0) = = = ,
XY (1, H) − XY (1, T ) u · XY (0) − d · XY (0) u−d
and
1
VX (1, H) − VX (1, T ) · YX (0) − 0 d
∆Y (0) = = 1
u
1 =− .
YX (1, H) − YX (1, T ) u · YX (0) − d · YX (0) u−d
Therefore the hedging portfolio is
YX (0) d
P0 = · X0 − · Y0 .
u−d u−d
Indeed,
YX (0) d 1−d
PY (0) = · XY (0) − = ,
u−d u−d u−d
YX (0) d u d
PY (1, H) = · XY (1, H) − = − = 1,
u−d u−d u−d u−d
YX (0) d d d
PY (1, T ) = · XY (1, T ) − = − = 0.
u−d u−d u−d u−d
 Solutions to Selected Exercises 297

(c)

UX (0) = EX [UX (1)] = EX [I(ω = H)] = PX (H),

1−d
U0 = PX (H) · X0 = u · · X0 .
u−d

(d) The hedging portfolio has the form

P0 = ∆X (0) · X0 + ∆Y (0) · Y0 ,

where

UY (1, H) − UY (1, T ) u · XY (0) − 0 u

∆X (0) = = = ,
XY (1, H) − XY (1, T ) u · XY (0) − d · XY (0) u−d

and

UX (1, H) − UX (1, T ) 1−0 ud · XY (0)

∆Y (0) = = 1 =− .
YX (1, H) − YX (1, T ) u · YX (0) − d1 · YX (0) u−d

Therefore the hedging portfolio is

u ud · XY (0)
P0 = · X0 − · Y0 .
u−d u−d

Indeed,

u ud · XY (0) 1−d
PX (0) = − · YX (0) = u · ,
u−d u−d u−d
u ud · XY (0) u d
PX (1, H) = − · YX (1, H) = − = 1,
u−d u−d u−d u−d
u ud · XY (0) u u
PX (1, T ) = − · YX (1, T ) = − = 0.
u−d u−d u−d u−d
298 Stochastic Finance: A Numeraire Approach

2.4:
(a) At the terminal time n = 2, we have

V$ (2, ω) = max(5, S$ (2, ω)).

Thus

V$ (2, HH) = 16, V$ (2, HT ) = V$ (2, T H) = 5, V$ (2, T T ) = 5.

At time n = 1, we have to compare the intrinsic value and the continuation

value of the contract. This gives us
 
1
V$ (1, ω) = max max(5, S$ (1, ω)), 1+r · ET [V$ (2)](ω) .

The T-forward measure is given by

1+r−d u − (1 + r)
pT = PT (H) = , and q T = PT (T ) = .
u−d u−d

For the particular choice of the parameters u = 2, d = 21 , r = 14 , we get

pT = q T = 21 . We conclude that
 
V$ (1, H) = max max(5, 8), 1 · [ 2 · 16 + 2 · 5] = max(8, 42
1 1 1 42
5 ) = 5 .
1+ 4

The continuation value is larger than the intrinsic value, and thus it is better
to continue. Similarly,
 
1 1 1
V$ (1, T ) = max max(5, 2), 1 · [ 2 · 5 + 2 · 5] = max(5, 4) = 5.
1+ 4

The intrinsic value is larger than the continuation value, and thus it is better
to stop. Finally,
 
V$ (0) = max max(5, 4), 1 1 · [ 12 · 42
5 + 1
2 · 5] = max(5, 134 134
25 ) = 25 .
1+ 4

The continuation value is larger than the intrinsic value, and thus it is better
to continue. The optimal stopping strategy is τ ∗ (HH) = τ ∗ (HT ) = 2, and
τ ∗ (T H) = τ ∗ (T T ) = 1.
(b) The hedging portfolio is given by
42
V$ (1, H) − V$ (1, T ) −5 17
∆S (0) = = 5 = ,
S$ (1, H) − S$ (1, T ) 8−2 30
21
VS (1, H) − VS (1, T ) 1 − 25 4 232
∆M (0) = · = 20
1 1 · 5 = 75 ,
$S (1, H) − $S (1, T ) 1 + r 8 − 2
 Solutions to Selected Exercises 299
V$ (2, HH) − V$ (2, HT ) 16 − 5 11
∆S (1, H) = = = ,
S$ (2, HH) − S$ (2, HT ) 16 − 4 12
VS (2, HH) − VS (2, HT ) 1 1− 5 4 16
∆M (1, H) = · = 1 41 · = .
$S (2, HH) − $S (2, HT ) 1 + r 16 − 8
5 5

The contract should be exercised at time 1 when ω = T , and thus no hedging

is necessary on that outcome.
(c) At the terminal time n = 2, we have

V$ (2, ω) = min(5, S$ (2, ω)).

Thus

V$ (2, HH) = 5, V$ (2, HT ) = V$ (2, T H) = 4, V$ (2, T T ) = 1.

At time n = 1, we have to compare the intrinsic value and the continuation

value of the contract. This gives us

 
1
V$ (1, ω) = max min(5, S$ (1, ω)), 1+r · ET [V$ (2)](ω) .

We conclude that
 
1
V$ (1, H) = max min(5, 8), 1 · [ 12 ·5+ 1
2 · 4] = max(5, 18
5 ) = 5.
1+ 4

The intrinsic value is larger than the continuation value, and thus it is better
to stop. Similarly,

 
1
V$ (1, T ) = max min(5, 2), 1 · [ 12 ·4+ 1
2 · 1] = max(2, 2) = 2.
1+ 4

The intrinsic value is the same as the continuation value, and thus it makes
no difference to continue or to stop. Finally,

 
1
V$ (0) = max min(5, 4), 1 · [ 12 · 5 + 1
2 · 2] = max(4, 14
5 ) = 4.
1+ 4

The intrinsic value is larger than the continuation value, and thus it is better
to stop. The contract becomes trivial; it should be exercised immediately.
300 Stochastic Finance: A Numeraire Approach

2.5:
(a)

1
3 VY (2, HH) = 0

8
VY (1, H) = 9
1
3
2
3 4
VY (2, HT ) = 3

8
VY (0) = 9

1
3 VY (2, T H) = 0
2
3
8
VY (1, T ) = 9

2
3 4
AY (2, T T ) = 3

(b)

2
3 VX (2, HH) = 0

1
VX (1, H) = 9
2
3
1
3 1
VX (2, HT ) = 3

2
VX (0) = 9

2
3 VX (2, T H) = 0
1
3
4
VX (1, T ) = 9

1
3 4
VX (2, T T ) = 3
 Solutions to Selected Exercises 301

(c)
8
VY (1, H) − VY (1, T ) −8
∆X (0) = = 9 9 = 0,
XY (1, H) − XY (1, T ) 8−2
1
VX (1, H) − VX (1, T ) −4 8
∆Y (0) = = 91 19 = ,
YX (1, H) − YX (1, T ) 8 − 2
3

VY (2, HH) − VY (2, HT ) 0 − 34 1

∆X (1, H) = = =− ,
XY (2, HH) − XY (2, HT ) 16 − 4 9
VX (2, HH) − VX (2, HT ) 0− 1 16
∆Y (1, H) = = 1 31 = ,
YX (2, HH) − YX (2, HT ) 16 − 4
9

VY (2, T H) − VY (2, T T ) 0 − 43 4
∆X (1, T ) = = =− ,
XY (2, T H) − XY (2, T T ) 4−1 9
4
VX (2, T H) − VX (2, T T ) 0 − 16
∆Y (1, T ) = = 1 3 = .
YX (2, T H) − YX (2, T T ) 4 − 1 9

(d) The probability measure PA is given by

AY (2, ω) Y
PA (ω) = · P (ω).
AY (0)
Thus we need values AY (2, ω) for ω = HH, HT, T H, T T . From the definition
of A
AY (2) = 13 (XY (0) + XY (1) + XY (2)) ,
we get


AY (2, HH) = 1
3 (XY (0) + XY (1, H) + XY (2, HH)) = 31 1 + u + u2 · XY (0),
1
AY (2, HT ) = 3 (XY (0) + XY (1, H) + XY (2, HT )) = 31 (1 + u + u · d) · XY (0),
1 1
AY (2, T H) = 3 (XY (0) + XY (1, T ) + XY (2, T H)) = 3 (1 + d + d · u) · XY (0),


AY (2, T T ) = 1
3 (XY (0) + XY (1, T ) + XY (2, T T )) = 1
3 1 + d + d2 · XY (0).

Furthermore, from the martingale property of AY , we get

AY (1) = EY1 [AY (2)] = EY1 [ 31 (XY (0) + XY (1) + XY (2))] = 13 ·XY (0)+ 32 ·XY (1),

and therefore
1 2


AY (1, H) = XY (1, H) = 3 + 3 ·u · XY (0),
1 2


AY (1, T ) = XY (1, T ) = 3 + 3 ·d · XY (0).

Similarly,

AY (0) = EY [AY (2)] = EY [ 13 (XY (0) + XY (1) + XY (2))] = XY (0).

302 Stochastic Finance: A Numeraire Approach

Therefore
 2
AY (2, HH) Y 1−d
PA (HH) = · P (HH) = 13 (1 + u + u2 ) · ,
AY (0) u−d
   
AY (2, HT ) Y 1−d u−1
PA (HT ) = · P (HT ) = 31 (1 + u + u · d) · · ,
AY (0) u−d u−d
   
AY (2, T H) Y 1−d u−1
PA (T H) = · P (T H) = 31 (1 + d + d · u) · · ,
AY (0) u−d u−d
 2
A AY (2, T T ) Y 1 2 u−1
P (T T ) = · P (T T ) = 3 (1 + d + d ) · .
AY (0) u−d
Similarly,
 
AY (1, H) Y 1−d
PA (H) = · P (H) = 13 (1 + 2u) · ,
AY (0) u−d
 
AY (1, T ) Y u−1
PA (T ) = · P (T ) = 31 (1 + 2d) · .
AY (0) u−d
This gives us conditional probabilities
   
PA (HH) 1 + u + u2 1−d
PA (HH|H) = = · ,
PA (H) 1 + 2u u−d
   
A PA (HT ) 1+u+u·d u−1
P (HT |H) = A = · ,
P (H) 1 + 2u u−d
   
PA (T H) 1+d+d·u 1−d
PA (T H|T ) = A = · ,
P (T ) 1 + 2d u−d
   
PA (T T ) 1 + d + d2 u−1
PA (T T |T ) = A = · .
P (T ) 1 + 2d u−d

(e) For the choice of the parameters u = 2 and d = 21 we get

 
1−d 5
PA (H) = 13 (1 + 2u) · = ,
u−d 9
 
u−1 4
PA (T ) = 13 (1 + 2d) · = ,
u−d 9
and
   
1 + u + u2 1−d 7
PA (HH|H) = · = ,
1 + 2u u−d 15
   
1+u+u·d u−1 8
PA (HT |H) = · = ,
1 + 2u u−d 15
   
1+d+d·u 1−d 5
PA (T H|T ) = · = ,
1 + 2d u−d 12
   
1 + d + d2 u−1 7
PA (T T |T ) = · = .
1 + 2d u−d 12
 Solutions to Selected Exercises 303

7
15 VA (2, HH) = 0

2
VA (1, H) = 15
5
9
8
1
15 VA (2, HT ) = 4

2
VA (0) = 9

5
12 VA (2, T H) = 0
4
9
1
VA (1, T ) = 3

7
4
12 VA (2, T T ) = 7

3.1:
(a) From f X (x) = f Y ( x1 ) · x we get

[f X (x)]′ = f Y ( x1 ) − 1
x · [f Y ( x1 )]′ ,

and

[f X (x)]′′ = − x12 · [f Y ( x1 )]′ + 1

x2 · [f Y ( x1 )]′ + 1
x3 · [f Y ( x1 )]′′ = 1
x3 · [f Y ( x1 )]′′ ≥ 0.

3.2:
Let us compute

d2+ d2−
√1 − 2 √1 − 2
XY (t) · 2π
· e −K · 2π
· e .

Note that

   √ 2
d2± = √1
σ T −t
· log XYK(t) ± 21 σ T − t
h  i2  
= 1
σ2 (T −t) · log XYK(t) + 41 σ 2 (T − t) ± log XYK(t) .
304 Stochastic Finance: A Numeraire Approach

Thus
d2+ d2−
XY (t) · √12π · e− 2 −K · · e− 2
√1
2π
 h  i2 
XY (t) 1 2
= √1 1
· exp 2σ2 (T log + σ (T − t)
2π −t) K 8
h      i
× XY (t) · exp − 21 log XYK(t) − K · exp 21 log XYK(t)
 h  i2 
XY (t) 1 2
= √1 1
· exp 2σ2 (T −t) log + 8 σ (T − t)
2π K
 q q 
K XY (t)
× XY (t) · XY (t) − K · K

= 0.

3.4:
(a) We have seen that the price VY (t) is given by

VY (t) = N (d− ) ,

where   √
XY (t)
d± = √1 · log ± 21 σ T − t.
σ T −t K

Thus
∂VY (t)
∆X (t) = = φ(d− ) · √1
σ T −t
· YX (t),
∂XY (t)
and
∆Y (t) = VY (t) − ∆X (t) · XY (t) = N (d− ) − φ(d− ) · √1
σ T −t
.
The hedging portfolio takes the following form
h i h i
PtV = φ(d− ) · σ√1T −t · YX (t) · X + N (d− ) − φ(d− ) · √1
σ T −t
· Y.

(b) The UX (t) price is given by

UX (t) = N (d+ ).

Therefore
∂UX (t)
∆Y (t) = = −φ(d+ ) · √1
σ T −t
· XY (t),
∂YX (t)
and
∆X (t) = UX (t) − ∆Y (t) · YX (t) = N (d+ ) + φ(d+ ) · √1
σ T −t
.
The hedging portfolio is given by
h i h i
PtU = N (d+ ) + φ(d+ ) · σ√1T −t · X + −φ(d+ ) · √1
σ T −t
· XY (t) · Y.
 Solutions to Selected Exercises 305

(c) The hedge of the contract W = U − K · V is given by

Pt = P U (t) − K · P V (t)
h i
= N (d+ ) + φ(d+ ) · σ√1T −t − K · φ(d− ) · σ√1T −t · YX (t) · X
h i
+ −φ(d+ ) · σ√1T −t · XY (t) − K · N (d− ) + K · φ(d− ) · σ√1T −t · Y
= [N (d+ )] · X + [−K · N (d− )] · Y.

We have used the identity

XY (t) · φ(d+ ) − K · φ(d− ) = 0

that appears in Exercise 3.2.

4.1:
Let us determine the evolution of dB$T (t) first. We have
 R 
T
B$T (t) = exp − t f (t, u)du ,

and thus according to Ito’s formula

h R i h R i
T T
dB$T (t) = B$T (t)d − t f (t, u)du + 12 B$T (t)d2 − t f (t, u)du .

We also have
h R i Z T
T
d − t f (t, u)du = f (t, t)dt − df (t, u)du
t
Z T
= r(t)dt + [σdW M (t)]du
t
= r(t)dt + σ(T − t)dW M (t),

implying

dB$T (t) = [r(t) + 21 σ 2 (T − t)2 ]B$T (t)dt + σ(T − t)B$T (t)dW M (t).

Therefore
T
dBM (t) = 12 σ 2 (T − t)2 BM
T T
(t)dt + σ(T − t)BM (t)dW M (t).

The bond B T price with respect to the money market has positive drift, and
thus arbitrage is possible. It cannot be locked by trading in only two assets
since there is still a noise term dW M (t), but it can be locked by trading in
three assets, two bonds B T1 , B T2 , and the money market M . Let’s find a
306 Stochastic Finance: A Numeraire Approach

portfolio of the form B T2 + ∆T1 (t)B T1 whose price with respect to the money
market M has a positive dt term and no noise term.
 
T2 T1
d BM (t) + ∆T1 (t)BM (t)
h i
T2 T1
= 12 σ 2 (T2 − t)2 BM (t) + ∆T1 (t) 21 σ 2 (T1 − t)2 BM (t) dt
h i
T2 T1
+ σ(T2 − t)BM (t) + ∆T1 (t)σ(T1 − t)BM (t) dW M (t).

When
T2
T 2 − t BM (t) T2 − t T2
∆T1 (t) = − · T1 =− · BB T1 (t),
T1 − t BM (t) T1 − t
the noise term dW M (t) cancels, and we get
 
T2 T1 T2
d BM (t) + ∆T1 (t)BM (t) = 12 σ 2 (T2 − t)(T2 − T2 )BM (t)dt.

The dt term is positive when T2 > T1 . Should the arbitrage portfolio start
with
P0 = B T2 + ∆T1 (0) · B T1 + ∆M (0) · M = 0,
we must have
T2
∆M (0) = ( TT21 − 1) · BM (0).
Therefore a portfolio with the following positions
   
T2 T1 M T2 − t T2 T2 T2
∆(t) = (∆ (t), ∆ (t), ∆ (t)) = 1, − · BB T1 (t), − 1 · BM (0)
T1 − t T1
is an arbitrage portfolio.

5.1:
(a)

VY (t) = EYt [VY (T )]

h  α i
= EYt [XY (T )]α = EYt [XY (t)]α · exp(σW Y (T − t) − 21 σ 2 (T − t))
= [XY (t)]α · exp(− 12 ασ 2 (T − t)) · EYt [exp(σαW Y (T − t))]
= [XY (t)]α · exp( 21 α(α − 1)σ 2 (T − t)).
Therefore
uY (t, x) = xα · exp( 21 α(α − 1)σ 2 (T − t)).

(b) If we substitute uY (t, x) = h(t) · xα into the partial differential equation

uYt (t, x) + 21 σ 2 x2 uYxx (t, x) = 0, we get

h′ (t)xα + 21 σ 2 x2 g(t) · α(α − 1) · xα−2 = 0,

 Solutions to Selected Exercises 307

or after dividing by xα

h′ (t) + 21 σ 2 α(α − 1) · h(t) = 0,

which is an ordinary differential equation for h(t) with a terminal condition

h(T ) = 1. The solution is given by

h(t) = exp( 12 α(α − 1)σ 2 (T − t)),

and thus uY is given by

uY (t, x) = xα · exp( 21 α(α − 1)σ 2 (T − t)),

which confirms the result from (a).

(c) The hedging positions are given by

∆X (t) = uYx (t, XY (t)) = α[XY (t)]α−1 · exp( 21 α(α − 1)σ 2 (T − t)),

and

∆Y (t) = uY (t, XY (t)) − XY (t) · uYx (t, XY (t))

= (1 − α)[XY (t)]α · exp( 21 α(α − 1)σ 2 (T − t)).

5.2:
First, in order to have L1−α Rα = X on the barrier, we must have
L1−α RYα (t) = XY (t) = Ler(T −t) . Solving for α from

L1−α RYα (t) = L1−α [XY (t)]α exp( 21 α(α − 1)σ 2 (T − t)) = XY (t)

leads to α = − σ2r2 . Similarly, on the barrier we have

PX X 1
t (XY (T ) ≥ K) · X = Pt ( K ≥ YX (T )) · X


1 1−α
= PX
t K ≥ [YX (T )] 1−α
·X


1 1−α
= PX
t K ≥ RX α
(T ) · X
 
(α)

1 1−α α XRα (T )
= Pt K ≥ RX (t) · · L1−α Rα
XRα (t)
  
1−α
(α) L2
= Pt K ≥ XR (T ) · L1−α Rα
α


1−α  1−α α
(α) α
= Pt RX (T ) ≥ LK2 ·L R

1−α

(α)
= Pt [YX (T )]1−α ≥ LK2 · L1−α Rα
 
(α) L2
= Pt K ≥ X Y (T ) · L1−α Rα .
308 Stochastic Finance: A Numeraire Approach

The important points in this problem are to realize that the two relevant
measures are just PX and P(α) , and an observation that RX
α
(T ) = [YX (T )]1−α .

5.3:
 √ 
P(α) (XY (T ) ≥ K) = N √1
σ T −t
· log( XYK(t) ) + (α − 12 )σ T − t .

5.4:
The price of V is given by the Black-Scholes formula
(2)2
Vt = Pt (RX (T ) ≥ K) · Rt2 − KPX (RX
2
(T ) ≥ K) · Xt .
(2)
2
Let us determine Pt (RX (T ) ≥ K) and PX 2
t (RX (T ) ≥ K). The second prob-
ability PX
t (RX
2
(T ) ≥ K) is simply

PX 2 X 2 X
t (RX (T ) ≥ K) = Pt ([XY (T )] · YX (T ) ≥ K) = Pt (XY (T ) ≥ K)
√
= N ( σ√1T −t · log( XYK(t) ) + 12 σ T − t).

2 (2)
The first probability Pt (RX (T ) ≥ K) is given by
(2)2 (2) (2)
Pt (RX (T ) ≥ K) = Pt (XY (T ) ≥ K) = Pt ([XY (T )]2 ≥ K 2 )
(2) (2)
= Pt (RY2 (T ) ≥ K 2 ) = Pt ( K12 ≥ YR2 (T ))
(2)
= Pt ( K12 ≥ YR2 (t) · exp(2σW (2) (T − t) − 2σ 2 (T − t)))
 √ 
(2) XY (t) W (2) (T −t)
= Pt √1 · log( ) + 3
σ T − t ≥ √
σ T −t K 2 T −t
 √ 
1 XY (t) 3
= N σ√T −t · log( K ) + 2 σ T − t .

The hedging portfolio P is given by

√
Pt = [N ( σ√1T −t · log( XYK(t) ) + 23 σ T − t)] · Rt2
√
+[−K · N ( σ√1T −t · log( XYK(t) ) + 21 σ T − t)] · Xt
h √ 2
= N ( σ√1T −t · log( XYK(t) ) + 23 σ T − t) · [2XY (t)eσ (T −t) ]
√ i
−K · N ( σ√1T −t · log( XYK(t) ) + 21 σ T − t) · Xt
h √ 2
i
+ −N ( σ√1T −t · log( XYK(t) ) + 23 σ T − t) · [XY (t)]2 eσ (T −t) · Yt .

We have expressed the hedge for R2 in terms of the assets X and Y using the
explicit formulas from Exercise 5.1.

9.1:
The solutions 1, x, and y represent contracts with payoffs YT , XT , and AT .
 Solutions to Selected Exercises 309

10.2:
Let f (x) = log(x). Then f ′ (x) = x1 , and from Ito’s formula we get
Z t
1
log(XY (t)) = log(XY (0)) + · (eγ − 1) · XY (s−)d(N (s) − λY s)
0 X Y (s−)
X  1

+ log(XY (s)) − log(XY (s−)) − ∆XY (s)
XY (s−)
0≤s≤t
Z t
= log(XY (0)) + (eγ − 1)d(N (s) − λY s)
0
Z t
+ [γ − (eγ − 1)] dN (s)
0
Z t Z t
= log(XY (0)) + γdN (s) + (eγ − 1)d(−λY s).
0 0

Rewriting these dynamics in differential form, we get

d log(XY (t)) = γdN (t) − (eγ − 1)λY dt.

10.3:
 
α αγ α α(eγ − 1) Y
d[XY (t)] = [e − 1] · [XY (t−)] d N (t) − αγ λ t .
e −1

10.4:
(a)
VY (t) = EYt [VY (T )] = EYt [I(N (T ) = 0)]
= PYt (N (T ) = 0) = PYt ((N (T ) − N (t) = 0) · I(N (t) = 0))


= exp −λY (T − t) · I(N (t) = 0).
(b)
VX (t) = EX X
t [VX (T )] = Et [YX (T )I(N (T ) = 0)]


= EX
t YX (t) · exp −(e
−γ
− 1)λX (T − t) · I(N (T ) = 0)


= YX (t) · exp −(e−γ − 1)λX (T − t) · PX t (N (T ) = 0)
−γ X


= YX (t) · exp −e · λ (T − t) · I(N (t) = 0).
Note that using the change of numeraire formula we have
VY (t) = VX (t) · XY (t)


= YX (t) · exp −e−γ λX (T − t) · I(N (t) = 0) · XY (t)


= exp −e−γ λX (T − t) · I(N (t) = 0)


= exp −λY (T − t) · I(N (t) = 0),
310 Stochastic Finance: A Numeraire Approach

so the two prices are indeed consistent.

(c) The hedging portfolio Pt = ∆X (t)·X + ∆Y (t)·Y takes the following form:
 
uY (t, eγ XY (t)) − uY (t, XY (t))
Pt = ·X
(eγ − 1)XY (t)
 X 
u (t, e−γ YX (t)) − uX (t, YX (t))
+ · Y.
(e−γ − 1) YX (t)

The price of the contract is zero after the first jump, so the values of
uY (t, eγ XY (t)) and uX (t, e−γ YX (t)) are both zero. Thus
 

X uY (t, eγ XY (t)) − uY (t, XY (t)) exp −λY (T − t) · I(N (t) = 0)
∆ (t) = =− ,
(eγ − 1)XY (t) (eγ − 1)XY (t)

and

uX (t, e−γ YX (t)) − uX (t, YX (t))

∆Y (t) =
(e−γ − 1) YX (t)


YX (t) · exp −e−γ λX (T − t) · I(N (t) = 0)
=−
(e−γ − 1) YX (t)
Y


exp −λ · (T − t) · I(N (t) = 0)
=− .
(e−γ − 1)

Thus the hedging portfolio has the form

Y Y
e−λ (T −t)
· I(N (t) = 0) γ e
−λ ·(T −t)
· I(N (t) = 0)
Pt = − · X + e · · Y.
(eγ − 1)XY (t) (eγ − 1)

(d) Let N (t) = m ≤ k. Then

VY (t) = EYt [VY (T )] = EYt [I(N (T ) = k)]

= PYt (N (T ) = k) = PYt (N (T ) − N (t) = k − m)
 k−m
Y

λY (T − t)
= exp −λ (T − t) · ,
(k − m)!

 k−m
Y

λY (T − t)
VX (t) = VY (t) · YX (t) = YX (t) · exp −λ (T − t) · .
(k − m)!

For the hedging portfolio we need the post-jump prices uY (t, eγ XY (t−)) and
uX (t, e−γ YX (t−)). When k > m, the number of jumps left to reach N (T ) = k
is reduced by one (now N (t) = m + 1 after the jump), and so we have the
 Solutions to Selected Exercises 311

same formula as above with k − m − 1 instead of k − m. Therefore

 Y 
X u (t, eγ XY (t)) − uY (t, XY (t))
∆ (t) =
(eγ − 1)XY (t)
Y [λY (T −t)]k−m−1 Y [λY (T −t)]k−m
e−λ (T −t)
· (k−m−1)! − e−λ (T −t)
· (k−m)!
=
(eγ − 1)XY (t)
 Y k−m−1
−λY (T −t) λ (T − t) k − m − λY (T − t)
=e · · ,
(k − m)! (eγ − 1)XY (t)
and
uX (t, e−γ YX (t)) − uX (t, YX (t))
∆Y (t) =
(e−γ − 1) YX (t)
 k−m−1 k−m

Y [λY (T −t)] −λY (T −t) [λ (T −t)]
Y
YX (t) e−λ (T −t) · (k−m−1)! − e · (k−m)!
=
(e−γ − 1) YX (t)
 
k−m−1
Y λY (T − t) k − m − λY (T − t)
= e−λ (T −t) · · .
(k − m)! (e−γ − 1)

10.7:
(a)
Z ∞  
dAY (t) = ¯ X (t−) · (ex − 1) · XY (t−) µY (dx, dt) − ν Y (dx)dt ,
∆
−∞
RT
¯X
where ∆ (t) = µ(ds).
t
(b)
Z "
∞

uYt (t, x, y) + ¯ X (t)(eξ − 1)x + y
ν (dξ) uY t, eξ x, ∆
Y
−∞
#
Y
− u (t, x, y) − uYx ξ
(t, x, y) · (e − 1)x − uYy ¯X ξ
(t, x, y) · ∆ (t)(e − 1)x = 0.

(c)
Z ∞  
 
dAX (t) = ¯ X (t−) · (ex − 1) · µX (dx, dt) − ν X (dx)dt ,
AX (t−) − ∆
−∞

Z "
∞

uX
t (t, x, y) +
¯ X (t)
ν X (dξ) uX t, eξ x, eξ y − (eξ − 1)∆
−∞
#
−u X
(t, x, y) − uX ξ
− 1)x − uX ¯X ξ
x (t, x, y) · (e y (t, x, y) · (y − ∆ (t))(e − 1) = 0.
312 Stochastic Finance: A Numeraire Approach

The reduced equation becomes

Z "
∞

uX ¯ X (t)
ν (dξ) uX t, eξ y − (eξ − 1)∆
X
t (t, y) +
−∞
#
X
− u (t, y) − uX ¯X ξ
· (y − ∆ (t))(e − 1) = 0.
y (t, x, y)
References

Arrow, K. J. and G. Debreu (1954). Existence of an equilibrium for a com-

petitive economy. Econometrica 22 (3), 265–290.
Bachelier, L. (1900). Théorie de la spéculation. Ann. Sci. École Norm. Sup. 17,
21–86.
Barone-Adesi, G. and R. Whaley (1987). Efficient analytic approximation of
American option values. Journal of Finance 42 (2), 301–320.
Baxter, M. and A. Rennie (1996). Financial Calculus. Cambridge University
Press.
Bayraktar, E. and H. Xing (2010). Pricing Asian options for jump diffusions.
Mathematical Finance.
Bensoussan, A. (1984). On the theory of option pricing. Acta Applicandae
Mathematicae 2 (2), 139–158.
Bjork, T. (2004). Arbitrage Theory in Continuous Time (2 ed.). Oxford
University Press, USA.
Black, F. (1976). The pricing of commodity contracts. Journal of Financial
Economics 3 (1-2), 167–179.
Black, F. and M. Scholes (1973). The pricing of options and corporate liabil-
ities. Journal of Political Economy 81 (3), 637–654.
Boyarchenko, S. and S. Levendorskii (2002). Perpetual American options
under Levy processes. SIAM Journal on Control and Optimization 40 (6),
1663–1696.
Boyle, P. (1977). Options: A Monte Carlo approach. Journal of Financial
Economics 4 (3), 323–338.
Boyle, P., M. Broadie, and P. Glasserman (1997). Monte Carlo methods for
security pricing. Journal of Economic Dynamics & Control 21 (8–9), 1267–
1321.
Boyle, P. P. (1988). A lattice framework for option pricing with two state
variables. The Journal of Financial and Quantitative Analysis 23 (1), 1–12.
Boyle, P. P., J. Evnine, and S. Gibbs (1989). Numerical evaluation of multi-
variate contingent claims. The Review of Financial Studies 2 (2), 241–250.

313
314 Stochastic Finance: A Numeraire Approach

Brace, A., D. Gatarek, and M. Musiela (1997). The market model of interest
rate dynamics. Mathematical Finance 7 (2), 127–155.
Brekke, K. (1997). The numeraire matters in cost-benefit analysis. Journal
of Public Economics 64 (1), 117–123.
Brigo, D. and F. Mercurio (2006). Interest Rate Models – Theory and Practice:
With Smile, Inflation and Credit (2 ed.). Springer.
Broadie, M. and J. Detemple (1996). American option valuation: New bounds,
approximations, and a comparison of existing methods. Review of Financial
Studies 9 (4), 1211–1250.
Broadie, M. and J. Detemple (1997). The valuation of American options on
multiple assets. Mathematical Finance 7 (3), 241–286.
Broadie, M. and P. Glasserman (1996). Estimating security price derivatives
using simulation. Management Science 42 (2), 269–285.
Broadie, M. and P. Glasserman (1997). Pricing American-style securities using
simulation. Journal of Economic Dynamics & Control 21 (8-9), 1323–1352.
Broadie, M., P. Glasserman, and S. Kou (1997). A continuity correction for
discrete barrier options. Mathematical Finance 7 (4), 325–349.
Brown, H., D. Hobson, and L. Rogers (2001). Robust hedging of barrier
options. Mathematical Finance 11 (3), 285–314.
Buchen, P. and O. Konstandatos (2005). A new method of pricing lookback
options. Mathematical Finance 15 (2), 245–259.
Cameron, R. H. and W. T. Martin (1944). Transformation of Wiener integrals
under translations. Ann. Math 45, 386–396.
Carr, P. (1998). Randomization and the American put. Review of Financial
Studies 11 (3), 597–626.
Carr, P. and J. Bowie (1994). Static simplicity. Risk 7, 44–50.
Carr, P. and A. Chou (1997). Breaking barriers. Risk 10, 139–145.
Carr, P., K. Ellis, and V. Gupta (1998). Static hedging of exotic options.
Journal of Finance 53 (3), 1165–1190.
Carr, P., H. Geman, and D. Madan (2001). Pricing and hedging in incomplete
markets. Journal of Financial Economics 62 (1), 131–167.
Carr, P., H. Geman, D. B. Madan, and M. Yor (2002). The fine structure
of asset returns: an empirical investigation. Journal of Business 75 (2),
305–332.
Carr, P. and R. Lee (2009). Put-call symmetry: Extensions and applications.
Mathematical Finance 19 (4), 523–560.
 References 315

Carr, P. P., R. Jarrow, and R. Myneni (1992). Alternative characterizations

of American put options. Mathematical Finance 2 (2), 87–106.
Cerny, A. (2009). Mathematical Techniques in Finance: Tools for Incomplete
Markets (2 ed.). Princeton University Press.
Cherubini, U., E. Luciano, and W. Vecchiato (2004). Copula Methods in
Finance. John Wiley.
Clement, E., D. Lamberton, and P. Protter (2002). An analysis of a least
squares regression method for American option pricing. Finance and
Stochastics 6 (4), 449–471.
Cont, R. and P. Tankov (2003). Financial Modelling with Jump Processes.
Chapman & Hall.
Cont, R. and E. Voltchkova (2005a). A finite difference scheme for option
pricing in jump diffusion and exponential levy models. SIAM Journal on
Numerical Analysis 43 (4), 1596–1626.
Cont, R. and E. Voltchkova (2005b). Integro-differential equations for option
prices in exponential Levy models. Finance and Stochastics 9 (3), 299–325.
Cox, J. C., J. E. Ingersoll, and S. A. Ross (1985). A theory of the term
structure of interest rates. Econometrica 53 (2), 385–407.
Cox, J. C., S. A. Ross, and M. Rubinstein (1979). Option pricing: A simplified
approach. Journal of Financial Economics 7 (3), 229–263.
Cox, J. C. and M. Rubinstein (1985). Options Markets. Prentice Hall.
Curran, M. (1994). Valuing Asian and portfolio options by conditioning on
the geometric mean price. Management Science 40 (12), 1705–1711.
Dana, R. A. and M. Jeanblanc (2007). Financial Markets in Continuous Time
(2 ed.). Springer.
Delbaen, F. and W. Schachermayer (1996). The fundamental theorem of asset
pricing for unbounded stochastic processes. Mathematische Annalen 312,
215–250.
Delbaen, F. and W. Schachermayer (2006). The Mathematics of Arbitrage.
Springer Finance. Springer-Verlag.
D’Halluin, Y., P. Forsyth, and G. Labahn (2005). A semi-Lagrangian ap-
proach for American-Asian options under jump diffusion. SIAM Journal
on Scientific Computing 27 (1), 315–345.
Duffie, D. (2001). Dynamic Asset Pricing Theory (3 ed.). Princeton University
Press, Princeton, New Jersey.
Duffy, D. J. (2006). Finite Difference Methods in Financial Engineering: A
316 Stochastic Finance: A Numeraire Approach

Partial Differential Equation Approach. Wiley.

Dufresne, D. (2000). Laguerre series for Asian and other options. Mathemat-
ical Finance 10 (4), 407–428.
Eberlein, E. and A. Papapantoleon (2005). Equivalence of floating and fixed
strike Asian and lookback options. Stochastic Processes and Their Appli-
cations 115 (1), 31–40.
El Karoui, N., M. Jeanblanc-Picque, and S. Shreve (1998). Robustness of the
Black and Scholes formula. Mathematical Finance 8 (2), 93–126.
Fajardo, J. and E. Mordecki (2006). Symmetry and duality in Levy markets.
Quantitative Finance 6 (3), 219–227.
Figlewski, S. and B. Gao (1999). The Adaptive Mesh Model: A new approach
to efficient option pricing. Journal of Financial Economics 53 (3), 313–351.
Filipovic, D. (2008). Optimal numeraires for risk measures. Mathematical
Finance 18 (2), 333–336.
Filipovic, D. (2009). Term-Structure Models: A Graduate Course. Springer
Verlag.
Flemming, J., S. Turnovsky, and M. Kemp (1977). Choice of numeraire and
certainty price in general equilibrium-models of price uncertainty. Review
of Economic Studies 44 (3), 573–583.
Fouque, J. P. and C. H. Han (2003). Pricing Asian options with stochastic
volatility. Quantitative Finance 3 (5), 353–362.
Fouque, J. P., G. Papanicolaou, and R. Sircar (2000). Derivatives in Financial
Markets with Stochastic Volatility. Cambridge University Press.
Garman, M. and S. Kohlhagen (1983). Foreign currency option values. Journal
of International Money and Finance 2 (3), 231–237.
Gatheral, J. (2006). The Volatility Surface: A Practitioner’s Guide. Wiley
Finance. Wiley.
Geman, H., N. El Karoui, and J. C. Rochet (1995). Changes of numeraire,
changes of probability measure and option pricing. Journal of Applied Prob-
ability 32, 443–458.
Geman, H. and M. Yor (1993). Bessel processes, Asian options, and perpetu-
ities. Mathematical Finance 3 (4), 349–375.
Gerber, H. and E. Shiu (1996). Martingale approach to pricing perpetual
American options on two stocks. Mathematical Finance 6 (3), 303–322.
Geske, R. and H. Johnson (1984). The American put option valued analyti-
cally. Journal of Finance 39 (5), 1511–1524.
 References 317

Girsanov, I. V. (1960). On transforming a certain class of stochastic processes

by absolutely continuous substitution of measures. Theory Probability Ap-
plications 5, 285–301.
Glasserman, P. (2003). Monte Carlo Methods in Financial Engineering.
Springer.
Goldman, M. B., H. B. Sossin, and M. A. Gatto (1979). Path dependent
options: “buy at the low, sell at the high”. Journal of Finance 34 (5),
1111–1127.
Gourieroux, C., J. Laurent, and H. Pham (1998). Mean-variance hedging and
numeraire. Mathematical Finance 8 (3), 179–200.
Harrison, J. M. and D. Kreps (1979). Martingales and arbitrage in multiperiod
securities markets. Journal of Economic Theory 20, 381–408.
Harrison, J. M. and S. Pliska (1981). Martingales and stochastic integrals in
the theory of continuos trading. Stochastic Processes and Their Applica-
tions 15, 313–316.
Haug, E. G. (1997). The Complete Guide to Option Pricing Formulas.
McGraw-Hill.
Haugh, M. and L. Kogan (2004). Pricing American options: A duality ap-
proach. Operations Research 52 (2), 258–270.
Heath, D., R. Jarrow, and A. Morton (1992). Bond pricing and the term struc-
ture of interest rates: A new methodology for contingent claims valuation.
Econometrica 60 (1), 77–105.
Henderson, V. and R. Wojakowski (2002). On the equivalence of floating- and
fixed-strike Asian options. Journal of Applied Probability 39 (2), 391–394.
Heston, S. L. (1993). A closed-form solution for options with stochastic volatil-
ity with applications to bond and currency options. The Review of Financial
Studies 6 (2), 327–343.
Hiriart-Urruty, J.-B. and C. Lemarechal (1993). Convex Analysis and Mini-
mization Algorithms I: Fundamentals. Springer.
Hobson, D. (1998). Robust hedging of the lookback option. Finance and
Stochastics 2 (4), 329–347.
Hoogland, J. and C. D. D. Neumann (2000). Asians and cash dividends:
exploiting symmetries in pricing theory. Working Paper .
Hoogland, J. and C. D. D. Neumann (2001a). Local scale invariance and
contingent claim pricing I. International Journal of Theoretical and Applied
Finance 4 (1), 1–21.
Hoogland, J. and C. D. D. Neumann (2001b). Local scale invariance and con-
318 Stochastic Finance: A Numeraire Approach

tingent claim pricing II: Path-dependent contingent claims. International

Journal of Theoretical and Applied Finance 4 (1), 23–43.
Hull, J. and A. White (1990). Pricing interest-rate-derivative securities. Re-
view of Financial Studies 3 (4), 573–592.
Hull, J. C. (2008). Options, Futures, and Other Derivatives (7 ed.). Prentice
Hall.
Ito, K. (1944). Stochastic integral. Proceedings of the Imperial Academy,
Tokyo 20, 519–524.
Jaeckel, P. (2002). Monte Carlo Methods in Finance. Wiley.
Jaillet, P., D. Lamberton, and B. Lapeyre (1990). Variational-inequalities and
the pricing of American options. Acta Applicandae Mathematicae 21 (3),
263–289.
James, J. and N. Webber (2000). Interest Rate Modelling: Financial Engi-
neering. John Wiley & Sons.
Jamshidian, F. (1989). An exact bond option formula. The Journal of Fi-
nance 44 (1), 205–209.
Jamshidian, F. (1997). LIBOR and swap market models and measures. Fi-
nance and Stochastics 1 (4), 293–330.
Jamshidian, F. (2007). Exchange options. Working Paper , 1–22.
Jeanblanc, M., M. Yor, and M. Chesney (2009). Mathematical Methods for
Financial Markets. Springer Finance. Springer-Verlag.
Johansson, P. (1998). Does the choice of numeraire matter in cost-benefit
analysis? Journal of Public Economics 70 (3), 489–493.
Johnson, H. (1987). Options on the maximum or the minimum of several
assets. Journal of Financial and Quantitative Analysis 22 (03), 277–283.
Joshi, M. S. (2008). The Concepts and Practice of Mathematical Finance.
Cambridge University Press.
Karatzas, I. (1988). On the pricing of American options. Applied Mathematics
and Optimization 17 (1), 37–60.
Karatzas, I. and C. Kardaras (2007). The numeraire portfolio in semimartin-
gale financial models. Finance and Stochastics 11 (4), 447–493.
Karatzas, I. and S. E. Shreve (1991). Brownian Motion and Stochastic Cal-
culus (2 ed.). Springer.
Karatzas, I. and S. E. Shreve (2001). Methods of Mathematical Finance.
Springer.
 References 319

Korn, R., E. Korn, and G. Kroisandt (2010). Monte Carlo Methods and Models
in Finance and Insurance. Chapman & Hall.
Kou, S. (2002). A jump-diffusion model for option pricing. Management
Science 48 (8), 1086–1101.
Kunitomo, N. and M. Ikeda (1992). Pricing options with curved boundaries.
Mathematical Finance 2 (4), 275–298.
Lawler, G. F. (2006). Introduction to Stochastic Processes. Chapman & Hall.
Levendorskii, S. (2004). Early exercise boundary and option prices in Levy
driven models. Quantitative Finance 4 (5), 525–547.
Lewis, A. L. (2000). Option Valuation Under Stochastic Volatility. Finance
Press.
Linetsky, V. (2004). Spectral expansions for Asian (average price) options.
Operations Research 52 (6), 856–867.
Long, J. (1990). The numeraire portfolio. Journal of Financial Eco-
nomics 26 (1), 29–69.
Longstaff, F. and E. Schwartz (2001). Valuing American options by simula-
tion: A simple least-squares approach. Review of Financial Studies 14 (1),
113–147.
Magdon-Ismail, M. and A. Atiya (2004). Maximum drawdown. Risk 17 (10),
99–102.
Magdon-Ismail, M., A. Atiya, A. Pratap, and Y. Abu-Mostafa (2004). On the
maximum drawdown of a Brownian motion. Journal of Applied Probabil-
ity 41, 147–161.
Margrabe, W. (1976). A theory of forward and futures prices. Wharton School
of Business, Preprint .
Margrabe, W. (1978). The value of an option to exchange one asset for an-
other. The Journal of Finance 33 (1), 177–186.
McKean, H. P. (1965). A free boundary problem for the heat equation aris-
ing from a problem in mathematical economics. Industrial Management
Review 6 (2), 32–39.
Merton, R. (1973). Theory of rational option pricing. Bell Journal of Eco-
nomics and Management Science 4, 141–183.
Merton, R. C. (1969). Lifetime portfolio selection under uncertainty: The
continuous-time case. The Review of Economics and Statistics 51 (3), 247–
257.
Merton, R. C. (1976). Option pricing when underlying stock returns are
320 Stochastic Finance: A Numeraire Approach

discontinuous. Journal of Financial Economics 3, 125–144.

Merton, R. C. (1992). Continuous-Time Finance. Wiley-Blackwell.
Mikosch, T. (1999). Elementary Stochastic Calculus With Finance in View.
World Scientific Publishing Company.
Milevsky, M. and S. Posner (1998). Asian options, the sum of lognormals, and
the reciprocal gamma distribution. Journal of Financial and Quantitative
Analysis 33 (3), 409–422.
Musiela, M. and M. Rutkowski (2008). Martingale Methods in Financial Mod-
elling (2 ed.). Springer.
Neftci, S. N. (2008). Principles of Financial Engineering (2 ed.). Academic
Press.
Nielsen, J. and K. Sandmann (2003). Pricing bounds on Asian options. Jour-
nal of Financial and Quantitative Analysis 38 (2), 449–473.
Oksendal, B. (2007). Stochastic Differential Equations: An Introduction with
Applications (6 ed.). Springer.
Papell, D. and H. Theodoridis (2001). The choice of numeraire currency
in panel tests of purchasing power parity. Journal of Money, Credit and
Banking 33 (3), 790–803.
Pellser, A. (2000). Efficient Methods for Valuing Interest Rate Derivatives.
Springer Verlag.
Pham, H. (1997). Optimal stopping, free boundary, and American option
in a jump-diffusion model. Applied Mathematics and Optimization 35 (2),
145–164.
Platen, E. (2004). A class of complete benchmark models with intensity-based
jumps. Journal of Applied Probability 41 (1), 19–34.
Platen, E. (2006). A benchmark approach to finance. Mathematical Fi-
nance 16 (1), 131–151.
Pospisil, L. and J. Vecer (2008). PDE methods for the maximum drawdown.
Journal of Computational Finance 2 (12), 59–76.
Pospisil, L. and J. Vecer (2010). Portfolio sensitivity to changes in the maxi-
mum and the maximum drawdown. Quantitative Finance 10 (6), 617–627.
Protter, P. E. (2005). Stochastic Integration and Differential Equations (2
ed.). Springer.
Rebonato, R. (2004). Volatility and Correlation: The Perfect Hedger and the
Fox. Wiley.
Revuz, D. and M. Yor (2004). Continuous Martingales and Brownian Motion.
 References 321

Springer.
Rogers, L. (2002). Monte Carlo valuation of American options. Mathematical
Finance 12 (3), 271–286.
Rogers, L. and Z. Shi (1995). The value of an Asian option. Journal of Applied
Probability 32 (4), 1077–1088.
Rubinstein, M. and E. Reiner (1991). Breaking down barriers. Risk 4 (9),
28–35.
Sadr, A. (2009). Interest Rate Swaps and Their Derivatives: A Practitioner’s
Guide (Wiley Finance). Wiley.
Samuelson, P. A. (1965). Proof that properly anticipated prices fluctuate
randomly. Industrial Management Review 6 (2), 41–49.
Samuelson, P. A. (1973). Mathematics of speculative price. SIAM Re-
view 15 (1), 1–42.
Schoutens, W. (2003). Levy Processes in Finance: Pricing Financial Deriva-
tives. Wiley.
Schroder, M. (1999). Changes of numeraire for pricing futures, forwards, and
options. Review of Financial Studies 12 (5), 1143–1163.
Shiryaev, A. N. (1999). Essentials of Stochastic Finance: Facts, Models, The-
ory. World Scientific Publishing Company.
Shreve, S. E. (2004a). Stochastic Calculus for Finance. I. Springer Finance.
Springer-Verlag.
Shreve, S. E. (2004b). Stochastic Calculus for Finance. II. Springer Finance.
Springer-Verlag.
Stulz, R. M. (1982). Options on the minimum or the maximum of two risky
assets : Analysis and applications. Journal of Financial Economics 10 (2),
161 – 185.
Tavella, D. and C. Randall (2000). Pricing Financial Instruments: The Finite
Difference Method. Wiley.
Tsitsiklis, J. and B. Van Roy (2001). Regression methods for pricing complex
American-style options. Neural Networks, IEEE Transactions on 12 (4),
694–703.
Vasicek, O. (1977). An equilibrium characterization of the term structure.
Journal of Financial Economics 5 (2), 177–188.
Vecer, J. (2001). A new PDE approach for pricing arithmetic average Asian
options. Journal of Computational Finance 4 (4), 105–113.
Vecer, J. (2002). Unified pricing of Asian options. Risk 15 (6), 113–116.
322 Stochastic Finance: A Numeraire Approach

Vecer, J. (2006). Maximum drawdown and directional trading. Risk 19 (12),

88–92.
Vecer, J. and M. Xu (2004). Pricing Asian options in a semimartingale model.
Quantitative Finance 4 (2), 170–175.
Williams, D. (1991). Probability with Martingales. Cambridge University
Press.
Wilmott, P. (2006). Paul Wilmott on Quantitative Finance 3 Volume Set (2
ed.). Wiley.
Wystup, U. (2008). Foreign exchange symmetries. Working Paper , 1–16.
Zhang, X. (1997). Numerical analysis of American option pricing in a jump-
diffusion model. Mathematics of Operations Research 22 (3), 668–690.
Zhu, S.-P. (2006). An exact and explicit solution for the valuation of American
put options. Quantitative Finance 6 (3), 229–242.
Zvan, R., P. Forsyth, and K. Vetzal (1998). Penalty methods for American
options with stochastic volatility. Journal of Computational and Applied
Mathematics 91 (2), 199–218.
Zvan, R., K. Vetzal, and P. Forsyth (2000). PDE methods for pricing barrier
options. Journal of Economic Dynamics & Control 24 (11-12), 1563–1590.