Mathematical Finance: Emanuela Rosazza Gianin Carlo Sgarra

UNITEXT 149
Emanuela Rosazza Gianin

Carlo Sgarra
Mathematical
Finance
Theory Review and Exercises
Second Edition
UNITEXT
La Matematica per il 3+2
Volume 149
Editor-in-Chief
Alfio Quarteroni, Politecnico di Milano, Milan, Italy
École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
Series Editors
Luigi Ambrosio, Scuola Normale Superiore, Pisa, Italy
Paolo Biscari, Politecnico di Milano, Milan, Italy
Ciro Ciliberto, Università di Roma “Tor Vergata”, Rome, Italy
Camillo De Lellis, Institute for Advanced Study, Princeton, New Jersey, USA
Victor Panaretos, Institute of Mathematics, École Polytechnique Fédérale de
Lausanne (EPFL), Lausanne, Switzerland
Lorenzo Rosasco, DIBRIS, Università degli Studi di Genova, Genova, Italy
Center for Brains Mind and Machines, Massachusetts Institute of Technology,
Cambridge, Massachusetts, US
Istituto Italiano di Tecnologia, Genova, Italy
The UNITEXT - La Matematica per il 3+2 series is designed for undergraduate
and graduate academic courses, and also includes advanced textbooks at a research
level.
Originally released in Italian, the series now publishes textbooks in English
addressed to students in mathematics worldwide.
Some of the most successful books in the series have evolved through several
editions, adapting to the evolution of teaching curricula.
Submissions must include at least 3 sample chapters, a table of contents, and
a preface outlining the aims and scope of the book, how the book fits in with the
current literature, and which courses the book is suitable for.
For any further information, please contact the Editor at Springer:
francesca.bonadei@springer.com
THE SERIES IS INDEXED IN SCOPUS
***
UNITEXT is glad to announce a new series of free webinars and interviews
handled by the Board members, who rotate in order to interview top experts in their
field.
Access this link to subscribe to the events:
https://cassyni.com/events/TPQ2UgkCbJvvz5QbkcWXo3
Emanuela Rosazza Gianin • Carlo Sgarra
Mathematical Finance
Theory Review and Exercises
Second Edition
Emanuela Rosazza Gianin Carlo Sgarra
Dipartimento di Statistica e Metodi Dipartimento di Matematica
Quantitativi Politecnico di Milano
Università di Milano-Bicocca Milano, Italy
Milano, Italy
ISSN 2038-5714 ISSN 2532-3318 (electronic)

UNITEXT
ISSN 2038-5722 ISSN 2038-5757 (electronic)
La Matematica per il 3+2
ISBN 978-3-031-28377-2 ISBN 978-3-031-28378-9 (eBook)
https://doi.org/10.1007/978-3-031-28378-9
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2013, 2023
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family
To Francesco and Teresa
Preface to the Second Edition
This new edition of the textbook includes some new exercises both solved and
proposed. Some new material has been added to the theoretical introduction of a
few chapters. A couple of exercises has been removed since they were looking too
similar to other exercises already illustrated. Most important is the correction of
several misprints present in the previous edition in spite of the detailed checking we
tried to perform before sending our manuscript for printing. We detected all these
misprints during lectures, often thanks to our students or to our teaching assistants,
who pointed out some notational or numerical inconsistencies. We express our
gratitude to all of them, they are too many to mention them explicitly.
We included also some new references related to subjects developed quite recently
and to other textbooks published almost at the same time of ours or afterwards.
We hope this textbook will be useful to colleagues looking for auxiliary application-
oriented material for their courses on Mathematical Finance and to their students.
Milano, Italy Emanuela Rosazza Gianin

October 2022 Carlo Sgarra
vii
Preface to the First Edition
This exercise textbook is the result of several years of experience in teaching

courses on Quantitative Finance at our Universities. Although part of the content
was already published in an Italian edition a few years ago (2007), this new version
has been substantially modified: the exercise collection is much more consistent
and three new chapters have been included: one chapter on Arbitrage Theory for
discrete-time models, which is a relevant issue in many courses on Mathematical
Finance, another on Risk Measures, and one on option pricing in models beyond the
classical Black-Scholes framework. The subjects covered by the textbook include
option-pricing methods based both on Stochastic Calculus and Partial Differential
Equations, and some basic notions about portfolio optimization and risk theory.
Many textbooks on option pricing focus either on stochastic methods or on PDE
methods. The aim of the present collection of exercises is to provide examples
of both approaches and try to illustrate their mutual interplay with the aid of
the fundamental notions. Some of the exercises can be solved by students with
a rather basic mathematical background (essentially, we mean an introductory
course on Calculus and a basic course in Probability), while others require a more
developed mathematical knowledge (a course on Differential Equations and an
introductory course on stochastic processes). The book can be used in courses
with mathematical finance contents aimed at graduate students of Engineering,
Economics, and Mathematics.
The material proposed has been organized in 12 chapters. The first chapter
reviews the basic notions of probability and stochastic processes through specific
examples, while the second chapter illustrates the fundamental results of (static)
portfolio optimization. Chapter 3 presents applications of the basic option-pricing
techniques in a discrete-time framework, mainly in the binomial setting. Chapter 4
provides examples of the fundamental results of arbitrage theory in a discrete-time
setting. Chapter 5 focuses on the main results of the Black-Sholes model related to
European option-pricing and hedging; examples involving both static and dynamic
hedging strategies are illustrated. Chapter 6 presents exercises on some standard
methods in partial differential equations, which turn out to be helpful in solving
valuation problems of derivatives in continuous-time models. Chapter 7 deals with
ix
x Preface to the First Edition
more complex derivative products, namely American options: valuation and hedging
problems for these options do not admit explicit solutions, except in very few
(somehow trivial) cases; approximate solutions can be found by applying suitable
numerical techniques. We present a few examples in a discrete-time setting and
simple applications of the basic notions. Chapter 8 deals with valuation and hedging
of Exotic options; a huge number of different kind of derivative contracts belong
to this class, but we focused on the most common type of contracts, in particular
those for which an explicit solution for the valuation problem exists in a diffusion
setting: Barrier, Lookback, geometric Asian options; moreover, we provide several
examples of valuation in a binomial setting. Chapter 10 provides applications of
the valuation results for interest rate derivatives: the concern is mainly on short-rate
models, but a few examples on the so-called change of numéraire technique are
included. Chapter 11 attempts to illustrate how the derivatives valuation problem
can be attacked in models that drop some of the main assumptions underlying the
Black-Scholes model: a few examples, mainly involving affine stochastic volatility
models, are provided together with simple examples of jump-diffusion models; all
these are typically incomplete market models, and some specific assumptions about
the risk-neutral measure adopted by the market in order to assign an arbitrage-
free price to contingent claim must be made; the theoretical issues arising in this
framework go far beyond the purpose of the present textbook, so we decided to limit
our description to the most basic (and popular) models, in which these problems
can be avoided by making simple, but reasonable assumptions. Chapter 12 presents
some applications of the most important notions related to risk measures; these seem
to play an increasingly relevant role in many introductory courses in mathematical
finance, so we decided to include some examples in our exercise collection.
We have to thank several people for reading the manuscript and providing
useful comments on the material included: we thank Fabio Bellini, Marco Frittelli,
Massimo Morini, Paolo Verzella for reading the first draft of the textbook, and
Andrea Cosso, Daniele Marazzina, Lorenzo Mercuri for reading the final version
and suggesting important modifications. We thank our colleague Giovanni Cutolo
for the invaluable help and assistance on LaTeX and the graphical packages
necessary to edit the manuscript. We thank Francesca Ferrari and Francesca Bonadei
of Springer-Verlag for providing highly qualified editorial support. Finally we want
to thank all our colleagues and students who offered any kind of help or comment
in support and encouragement to the present work.
Milano, Italy Emanuela Rosazza Gianin

May 2013 Carlo Sgarra
Contents
1 Short Review of Probability and of Stochastic Processes . . . . . . . . . . . . . . 1

1.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Solved Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Proposed Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Portfolio Optimization in Discrete-Time Models. . . . . . . . . . . . . . . . . . . . . . . . 17
2.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Binomial Model for Option Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4 Absence of Arbitrage and Completeness of Market Models . . . . . . . . . . . 63
4.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5 Itô’s Formula and Stochastic Differential Equations . . . . . . . . . . . . . . . . . . . 89
5.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6 Partial Differential Equations in Finance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7 Black-Scholes Model for Option Pricing and Hedging Strategies. . . . . 131
7.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
xi
xii Contents
8 American Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

8.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
9 Exotic Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
9.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
10 Interest Rate Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
10.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11 Pricing Models Beyond Black-Scholes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
11.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
12 Risk Measures: Value at Risk and Beyond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
12.1 Review of Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Chapter 1
Short Review of Probability
and of Stochastic Processes
1.1 Review of Theory
Given a probability space .(Ω, F , P ), where .Ω denotes a non-empty set, .F a .σ -

algebra and P a probability measure on .Ω:
• a random variable (r.v.) is a function .X : Ω → R such that .{ω ∈ Ω : .X (ω) ∈
A} ∈ F for any Borel set A in .R;
• a stochastic process is a family .(Xt )t≥0 of random variables defined on
.(Ω, F , P ).
The stochastic process .(Xt )t≥0 is said to be a discrete-time stochastic process

if t takes values in .N and a continuous-time stochastic process if t takes values
in .R+ .
• A filtration on .Ω is a family .(Ft )t≥0 of .σ -algebras on .Ω such that .Fu ⊆ Fv for
any .u ≤ v.
A stochastic process .(Xt )t≥0 is called adapted to the filtration .(Ft )t≥0 if, for any
.s ≥ 0, .Xs is .Fs -measurable.
Given a stochastic process .(Xt )t≥0 , a filtration .(Ft )t≥0 is said to be generated by
.(Xt )t≥0 if, for any .s ≥ 0, .Fs is the smallest .σ -algebra that makes .Xs measurable.
A filtration can be interpreted as the evolution of the information available up to

a given time.
For a more detailed and exhaustive treatment of the notions recalled here, we
refer to the books of Mikosch [32] and Ross [39].
The following families of random variables and stochastic processes are widely
used in Mathematical Finance.
Bernoulli Random Variable
A random variable X has a Bernoulli distribution if it assumes only two values
(typically 1 and 0) with probability p and .(1 − p) respectively. In such a case, we
write .X ∼ B(p).
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

E. Rosazza Gianin, C. Sgarra, Mathematical Finance,
La Matematica per il 3+2 149, https://doi.org/10.1007/978-3-031-28378-9_1
2 1 Short Review of Probability and of Stochastic Processes
As a consequence, the expected value and the variance of X are equal, respec-
tively, to .E[X] = p and .V (X) = p(1 − p).
Binomial Random Variable and Binomial Process
A binomial random variable .Yn counts the number of successes in a series of n
independent trials, where p is the probability of success in any one trial. In such a
case, .Yn ∼ Bin(n; p). n
.Yn can be written as a sum .Yn = i=1 Xi where .(Xi )i=1,...,n are independent and
identically distributed (i.i.d.), .Xi ∼ B(p) and .Xi = 1 denotes a success in trial i.
The expected value and the variance of a binomial random variable .Yn ∼ Bin(n; p)
are, respectively, .E[Yn ] = np and .V (Yn ) = np(1 − p).
The sequence .(Yn )n∈N is called a binomial process.
Poisson Random Variable and Poisson Process
A random variable Z has a Poisson distribution with parameter .λ > 0 (in
symbols, .Z ∼ P oi(λ)) if it takes values in .N and its probability mass function
is given by
e−λ λk
P (Z = k) =
. , ∀k = 0, 1, 2, . . . , n, . . . .
k!
for any .n ∈ N. Consequently, the expected value and the variance of Z are equal,
respectively, to .E[Z] = λ and to .V (Z) = λ.
Furthermore, recall that a Poisson random variable can be obtained by taking the
limit of a sequence of binomial random variables as .p → 0, .n → +∞ and with
.pn = λ.
Setting .λ = νt (for .t ≥ 0), .ν can be understood as the rate or average number of

arrivals per unit of time.
A process .(Zt )t≥0 is said to be a Poisson process of rate .ν if .Z0 = 0 and if all
the increments .Zt − Zs (for .0 ≤ s ≤ t) are independent and identically distributed
as a Poisson with parameter .ν(t − s).
If .(Zt )t≥0 is a Poisson process of rate .ν counting the number of arrivals and T
denotes the time between two arrivals, then T has an exponential distribution with
parameter .ν > 0 (.T ∼ Exp (ν)). The density function of T is given by
fT (t) = νe−νt , ∀t > 0.

.
The expected value and the variance of T are then equal to .E [T ] = 1

ν and .V (T ) =
1
ν2
, respectively.
1.1 Review of Theory 3
Pareto Random Variable

A random variable X has a Pareto distribution with parameters .x0 > 0 and .a > 0
if its cumulative distribution function is given by
x a
0
P (X ≤ x) = 1 −
. 1[x0 ,+∞) (x) .
x
For .a > 1, X has finite expected value .E [X] = a

a−1 x0 .
Normal (or Gaussian) Random Variable
A random variable X taking values in .R has a Gaussian or normal distribution
with parameters .μ ∈ R and .σ 2 > 0 (.X ∼ N(μ, σ 2 )) if its cumulative distribution
function (denoted by N) is given by
x 2
1 − (y−μ)
N(x) P (X ≤ x) = √
. e 2σ2 dy.
2π σ 2 −∞
As a consequence, its expected value and variance are equal, respectively, to

E[X] = μ and .V (X) = σ 2 . When .μ = 0 and .σ 2 = 1, the random variable X is
.
called standard normal. Tables providing the values of N are widely available in the
literature.
Furthermore, a normal random variable .X ∼ N(μ, σ 2 ) can be transformed into
a standard normal .Z ∼ N(0, 1) by taking .Z = (X − μ)/σ .
The Central Limit Theorem guarantees that the sum of n random variables, that
are independent and identically distributed (i.i.d.) with finite expected values and
variances, converges in distribution to a standard normal. More precisely: given a
sequence .(Xn )n∈N of i.i.d. random variables with .E (Xn) = μ and .V (Xn ) = σ 2
n
X −nμ d
for any .n ∈ N, where .μ ∈ R and .σ 2 > 0, it follows that . i=1σ √ni →n N (0, 1).
In particular, if .Xn is a binomial random variable with parameters n, .p (i.e. the
n −np
sum of n independent Bernoulli random variables with parameter p), then . √Xnp(1−p)
can be approximated by a standard normal.
By the Central Limit Theorem the following stochastic processes can be obtained
as limits of processes discussed previously.
Brownian Motion
A Brownian Motion .(Xt )t≥0 is a continuous-time stochastic process with
continuous trajectories such that
• .X0 = 0;
• its increments are stationary and independent;
• for any .t ≥ 0, .Xt ∼ N(μt, σ 2 t).
μ and .σ are called drift and diffusion, respectively. When .μ = 0 and .σ 2 = 1, the
.
Brownian motion is called standard and denoted by .(Wt )t≥0 .

By definition of Brownian motion it follows that, for any .0 ≤ s ≤ t,
Xt − Xs ∼ N(μ (t − s) , σ 2 (t − s)).
.
For a standard Brownian motion we have
. Wt − Ws ∼ N(0, t − s).
A standard Brownian motion can be obtained as limit of a suitable binomial

process. Suppose that, at any interval .Δt of time, the random variable at the previous
time may have a positive or negative increment n .Δx with the same probability p.
After .n = [t/Δt] periods, we have .Yn = i=1 Yi (Δx), where .Yi takes values
in .{−1, 1}. Such a process is called binomial random walk. By taking the limit
of .Yn (under the assumption that .Δx → 0, .Δt → 0 and .lim[(Δx)2 /Δt] = σ 2 as
.n → +∞) and applying the Central Limit Theorem, one obtains a Brownian motion
with mean 0 and variance .σ 2 t. In order to obtain a standard Brownian motion, it is

sufficient to have .lim(Δx)2 /Δt = 1.
Log-Normal Stochastic Process
A stochastic process .(St )t≥0 such that
St = exp (Xt ) ,
.
where .(Xt )t≥0 is a Brownian motion, is said to be log-normal. It follows, therefore,

that for any .0 ≤ s ≤ t

St
. ln ∼ N(μt, σ 2 t)
S0

St
ln ∼ N(μ (t − s) , σ 2 (t − s)).
Ss
As for the Brownian motion, also the log-normal process can be seen as the
limit of a suitable stochastic process. It can be obtained, indeed, as the limit (for
.tk − tk−1 = Δt → 0) of a discrete process of the form .Sk = Sk−1 X, where X is a
√ √
√ values .u = exp(σ Δt)√and .d = exp(−σ Δt)
Bernoulli random variable taking
with probabilities .pu = (1+μ Δt/σ )/2 and .pd = (1−μ Δt/σ )/2, respectively.
By passing to the limit, the process .(St )t≥0 satisfies

St
. ln ∼ N(μt, σ 2 t), (1.1)
S0
1.2 Solved Exercises 5
so that .(ln (St ) / ln (S0 ))t≥0 is a Brownian motion with drift .μ and diffusion .σ .
Such a process .(St )t≥0 is also called geometric Brownian motion with drift .μ and
diffusion .σ .
Among the different families of stochastic processes, a remarkable one is the
family of martingales.
Discrete-Time Martingales
A stochastic process .(Xn )n≥0 is said to be a discrete-time martingale with respect
to the filtration .(Fn )n≥0 if .(Xn )n≥0 is .(Fn )n≥0 -adapted and if, for any .n ≥ 0, we
have .E [|Xn |] < +∞ and
.E [ Xn+1 | Fn ] = Xn .
Continuous-Time Martingales
A stochastic process .(Xt )t≥0 is said to be a continuous-time martingale with
respect to the filtration .(Ft )t≥0 if .(Xt )t≥0 is .(Ft )t≥0 -adapted and, for any .0 ≤ s ≤
t, we have .E [|Xt |] < +∞ and
E [ Xt | Fs ] = Xs .
.
1.2 Solved Exercises
Exercise 1.1 Let (Wt )t≥0 be a standard Brownian motion.

1. Calculate P (W6 − W2 < 0; W1 > 0) and V ((W6 − W2 ) W1 ).
2. Consider the stochastic process (Xt )t≥0 defined as Xt = μt + Wt . Establish if
there exists a drift μ > 0 such that the probability P (X6 − X2 < 0; X1 > 0) is
at least equal to 20%.
3. Set (Yt )t≥0 to be the process Yt = μt + σ Wt . Find the distribution of Yt for
μ = 0 and for μ = 0.
4. Compute P (Y6 − Y2 < 0; Y1 > 0), P (Y6 − Y2 < 4σ ) and E [(Y6 − Y2 ) Y1 ] for
μ = 0.1 and σ = 0.4.
Solution
1. Since (W6 − W2 ) and W1 are independent random variables (by definition of
Brownian motion), we deduce that
P (W6 − W2 < 0; W1 > 0) = P (W6 − W2 < 0) · P (W1 > 0) .

.
−W2
W6√
By (W6 − W2 ) ∼ N (0; 4) (hence, ∼ N(0, 1)) and W1 ∼ N (0; 1), it
4
follows that
P (W6 − W2 < 0; W1 > 0) = P (W6 − W2 < 0) · P (W1 > 0)

.

W6 − W2
=P √ < 0 · P (W1 > 0)
4
1 1
= N (0) · [1 − N (0)] = · = 0.25,
2 2
where N denotes the cumulative distribution function of a standard normal.
Furthermore,

.V ((W6 − W2 ) W1 ) = E (W6 − W2 ) W1 − (E [(W6 − W2 ) W1 ]) ,
2 2 2
(1.2)
and because of the properties of standard Brownian motions we get
V ((W6 − W2 ) W1 )
. = E (W6 − W2 )2 E W12 − (E [W6 − W2 ])2 (E [W1 ])2
= V (W6 − W2 ) V (W1 ) = 4 · 1 = 4.
2. We have to establish if there exists μ > 0 such that
P (X6 − X2 < 0; X1 > 0) ≥ 0.2.

.
To this end we rewrite the left-hand side of the inequality above in terms of μ
and of the Brownian motion. This gives
P (X6 − X2 < 0; X1 > 0) = P (6μ + W6 − (2μ + W2 ) < 0; μ + W1 > 0)

.
= P (W6 − W2 < −4μ; W1 > −μ)

= P (W6 − W2 < −4μ) · P (W1 > −μ)

W6 − W2
=P < −2μ · P (W1 > −μ)
2
= N (−2μ) · [1 − N (−μ)] = N (−2μ) · N (μ) ,
where the last inequality is due to the symmetry of the normal distribution.
The initial problem is, therefore, equivalent to establish if there exists a drift
μ > 0 so that N (−2μ)·N (μ) ≥ 0.2. The answer is yes, because taking μ = 0.1,
for instance, gives N (−2μ) · N (μ) = 0.227 ≥ 0.2.
3. Consider, first, the case where Yt = μt + σ Wt with μ = 0.
Recall that Wt ∼ N (0; t) for any t > 0. It follows that σ Wt is distributed as

a normal with mean
E [σ Wt ] = σ E [Wt ] = 0
.
and variance
V (σ Wt ) = σ 2 V (Wt ) = σ 2 t.
.
Hence, σ Wt ∼ N 0; σ 2 t .
In general, for an arbitrary μ ∈ R and for any t > 0, Yt = μt + σ Wt is
distributed as a normal with mean
E [Yt ] = E [μt + σ Wt ] = μt + σ E [Wt ] = μt

.
and variance

V (Yt ) = V (μt + σ Wt ) = V σ 2 Wt = σ 2 t.
.
Hence, for any t > 0, Yt ∼ N μt; σ 2 t either for μ = 0 or μ = 0.

4. Proceeding as above, we get
P (Y6 − Y2 < 0; Y1 > 0) = P (σ (W6 − W2 ) < −4μ; σ W1 > −μ)

.

W6 − W2 −2μ μ
=P < · P W1 > −
2 σ σ
μ
2μ
=N − 1−N −
σ σ
= N (−0.5) [1 − N (−0.25)] = 0.185,
and also
P (Y6 − Y2 < 4σ ) = P (σ (W6 − W2 ) < −4μ + 4σ )

.

W6 − W2 −2μ
=P < +2
2 σ

3
=N = 0.933.
2
By the properties of Brownian motions, we also find
E [(Y6 − Y2 ) Y1 ]
= E [(4μ + σ (W6 − W2 )) (μ + σ W1 )]
. = E 4μ2 + 4μσ W1 + μσ (W6 − W2 ) + σ 2 (W6 − W2 ) W1
= 4μ2 + 4μσ E [W1 ] + μσ E [W6 − W2 ] + σ 2 E [(W6 − W2 ) W1 ]
= 4μ2 + σ 2 E [(W6 − W2 ) W1 ] = 4μ2 ,
where the last inequality can be deduced in two different ways. The first is
based on the independence of (W6 − W2 ) and W1 . Then E [(W6 − W2 ) W1 ] =
E [W6 − W2 ] E [W1 ] = 0. The second comes from E [Wt Ws ] = min{s; t}; so
E [(W6 − W2 ) W1 ] = E [W6 W1 ] − E [W2 W1 ] = 1 − 1 = 0.
Exercise 1.2 Consider two geometric Brownian motions St1 t≥0 and St2 t≥0 ,
representing the prices of two stocks, with drifts μ1 , μ2 and diffusions σ1 = σ > 0,
σ2 = 2σ respectively, and with the same initial price S01 = S02 = S0 > 0.
1. Compute P St1 ≥ St2 .
2. For μ1 = 4μ2 and t = 4 years, find under which conditions on μ2 /σ one has
P S41 ≥ S42 ≥ 14 .
1
S
3. Compute E St1 − St2 and V t2 .
St
Solution Recall that for a geometric Brownian motion (Xt )t≥0 with
Xt = X0 eμt+σ Wt ,
.
one has

Xt
. ln ∼ N μ (t − s) ; σ 2 (t − s) (1.3)
Xs
for any 0 ≤ s ≤ t.
1. By (1.3) and S01 = S02 = S0 > 0 we deduce that

P St1 ≥ St2 = P S0 eμ1 t+σ Wt ≥ S0 eμ2 t+2σ Wt
.
= P (μ1 t + σ Wt ≥ μ2 t + 2σ Wt )
= P (σ Wt ≤ (μ1 − μ2 ) t)
√
Wt (μ1 − μ2 ) t
=P √ ≤
t σ
√
(μ1 − μ2 ) t
=N .
σ
2. By substituting μ1 = 4μ2 and t = 4 in the expression of P St1 ≥ St2 just

obtained, we get
√
(μ1 − μ2 ) 4 6μ2
P
. S41 ≥ S42 =N =N .
σ σ

In order to have P S41 ≥ S42 = N 6μ σ
2
≥ 0.25, the inequality 6μ
σ
2
≥
−1 ∼
N (0.25) = −0.67 should hold (see the tables of the cumulative distribution
function of the standard normal). This forces μσ2 ≥ −0.11.
s2
3. First of all, we remind that E eY = em+ 2 for Y ∼ N m, s 2 . Hence
1 2
S S
E St1 − St2 = E t · S0 − E t · S0
.
S0 S0

= S0 · E eμ1 t+σ Wt − S0 · E eμ2 t+2σ Wt

σ 2t 2
= S0 eμ1 t+ 2 − eμ2 t+2σ t ,
since μ1 t + σ Wt ∼ N μ1 t; σ 2 t and μ2 t + 2σ Wt ∼ N μ2 t; 4σ 2 t .
As for the variance, we obtain that

St1 St1
V =V = V eμ1 t+σ Wt · e−(μ2 t+2σ Wt )
S0
·
St2
St2
S0
2 2
=V e(μ1 −μ2 )t−σ Wt
= E e(μ1 −μ2 )t−σ Wt − E e(μ1 −μ2 )t−σ Wt
.
2
σ 2t
= E e2(μ1 −μ2 )t−2σ Wt − e(μ1 −μ2 )t+ 2
2
= e2(μ1 −μ2 )t+2σ t − e2(μ1 −μ2 )t+σ t = e2(μ1 −μ2 )t+σ t eσ t − 1 ,
2 2 2
where the previous equalities are based on the fact that Wt and (−Wt ) have the
same distribution.
Exercise 1.3 Suppose that the number of shares of a given stock bought over time
(measured in minutes) follows a Poisson process of rate λ = 12 per minute.
1. Establish how many minutes are needed so that more than 36 shares are bought
with probability of at least 94.8%. Denote by n∗ this minimum number of
minutes.
2. Compute the average waiting time for the purchase of 40 shares.
3. Assume that the same stock is also sold on a different market and that the number
of its shares bought in time (measured in minutes) in such a market follows a
Poisson process of rate μ = 8 per minute, independent of the first one.
Compute the probability that the total number of shares (bought in the two
markets combined) in n∗ minutes is greater than 72.
Solution
1. Denote by Xt the number of shares bought in t minutes on the first market. We
have to find how many minutes t are needed so that P (Xt > 36) ≥ 0.948.
Denote by n∗ the smallest following integer, namely n∗ = [t] + 1. We know
that
.Xt ∼ P oi (λt) .
Hence,

36
(λt)k
P (Xt > 36) = 1 − P (Xt ≤ 36) = 1 −
. e−λt ·
k!
κ=0

36
(12t)k
= 1− e−12t · .
k!
κ=0
Since one has P (Xt > 36) = 1.44 · 10−7 for t = 1, P (Xt > 36) = 0.008
for t = 2, P (Xt > 36) = 0.456 for t = 3 and P (Xt > 36) = 0.956 for
t = 4, we deduce that the minimum number of minutes to wait in order for
P (Xt > 36) ≥ 0.948 is n∗ = 4.
2. Let Ti be the waiting time (in minutes) of the i-th purchase. It is well known that
Ti+1 − Ti ∼ Exp (λ) .

.
Hence, the average time to wait for the purchase of 40 shares (measured in
minutes) is given by
.E [T40 − T0 ] = E [T40 − T39 ] + E [T39 − T38 ] + . . . + E [T1 − T0 ]

1 1 1 40 40
= + + ... + = = = 3.33.
λ λ λ λ 12
3. Denote now by Yt the number of shares bought in t minutes on the second market.
Since Yt is assumed to be a Poisson process of intensity μ (so, Yt ∼ P oi (μt))
and Xt and Yt are assumed to be independent,
Xt + Yt ∼ P oi ((λ + μ) t)
.
for any t ≥ 0.
Since n∗ = 4 (by item 1.), it follows that
P (Xn∗ + Yn∗ > 72) = 1 − P (X4 + Y4 ≤ 72)

.

72
(4 (λ + μ))k
= 1− e−4(λ+μ) ·
k!
k=0

72
(80)k ∼
= 1− e−80 · = 0.80.
k!
k=0
Exercise 1.4 Consider a stock with current price of 8 euros. In each of the following
2 years the stock price may increase by 20% (with probability 40%) or decrease by
20% (with probability 60%).
Denote with (Sn )n=0,1,2 the process representing the evolution of the stock price
in time. S1 may then take the values S1u = S0 u and S1d = S0 d, while S2 the values
S2uu = S0 u2 , S2ud = S0 ud and S2dd = S0 d 2 , where u is the growth factor and d the
decreasing factor.
1. Is (Sn )n=0,1,2 a martingale with respect to the above probability and with respect
to the filtration generated by (Sn )n=0,1,2 ?
2. Consider the stochastic process (S̃n )n≥0 defined as
.S̃0 S0
S̃1u S1u − k; S̃1d S1d + k
S̃2 S2 .

Establish whether S̃n can be a martingale with respect to the probability
n=0,1,2
and the filtration of the previous item for a suitable k > 0.
3. Discuss if there exists a probability measure Q such that the probability that
the stock price increases (respectively, decreases) in the first year is equal to the
probability that the stock price increases (respectively, decreases) in the second
year, and such that (Sn )n=0,1,2 is a martingale with respect to Q.
4. Establish if there exist û > 1 and d̂ > 0 such that the new stock price process is
a martingale with respect to the probability measure of item 1.
Solution Since the growth factor is u = 1.2 and the decreasing factor is d = 0.8,
the stock prices evolves as follows.
11.52 = S0 u2
S0 u = 9.6

S0 = 8 7.68 = S0 ud
.
S0 d = 6.4

5.12 = S0 d 2
− − −− − − − − −− − − − − −−
0 1 year (S1 ) 2 years (S2 )
It is also easy to check that
P (S1 = 9.6) = 0.4; P (S1 = 6.4) = 0.6

P (S2 = 11.52) = 0.16; P (S2 = 7.68) = 0.48; P (S2 = 5.12) = 0.36
.
P ( S2 = 11.52| S1 = 9.6) = P ( S2 = 7.68| S1 = 6.4) = 0.4
P ( S2 = 7.68| S1 = 9.6) = P ( S2 = 5.12| S1 = 6.4) = 0.6,
and so on.
1. Let us verify if (Sn )n=0,1,2 is a martingale with respect to P .
From
E [ S1 | S0 ] = S1u p + S1d (1 − p) = 9.6 · 0.4 + 6.4 · 0.6

.
= 7.68 = S0
it follows immediately that (Sn )n=0,1,2 is not a martingale with respect to P .

2. We need to verify if the following equalities hold for some k > 0:
⎧
⎨ E S̃1 S̃0 = S̃0
.
⎩ E S̃2 S̃1 = S̃1
.
The first equality is equivalent to

E S̃1 S̃0 = S̃0
.
(9.6 − k) · 0.4 + (6.4 + k) · 0.6 = 8
7.68 + 0.2 · k = 8
k = 1.6,
hence S̃1u = S̃1d = 8. This implies that

E S̃2 S̃1 = S̃1u = 11.52 · 0.4 + 7.68 · 0.6 = 9.216 = S̃1u ,
.

so S̃n is not a martingale with respect to P .
n=0,1,2
3. Define now a probability measure Q as follows:
Q (S1 = 9.6) = q; Q (S1 = 6.4) = 1 − q

Q (S2 = 11.52) = q 2 ; Q (S2 = 7.68) = 2q (1 − q)
. Q (S2 = 5.12) = (1 − q)2
Q ( S2 = 11.52| S1 = 9.6) = Q ( S2 = 7.68| S1 = 6.4) = q
Q ( S2 = 7.68| S1 = 9.6) = Q ( S2 = 5.12| S1 = 6.4) = 1 − q.
We look now for q ∈ (0, 1) such that

EQ [ S1 | S0 ] = S0
. ,
EQ [ S2 | S1 ] = S1
or, equivalently,
EQ [ S1 | S0 ] = S0.
. (1.4)
EQ S2 | S1 = S1u = S1u. (1.5)

EQ S2 | S1 = S1d = S1d . (1.6)
By Eq. (1.4) it follows that
S0 uq + S0 d (1 − q) = S0
.
q = u−d
1−d
= 12 .
There remains therefore to check if such a q also satisfies equations (1.5)

and (1.6) or not.
Since uq + d (1 − q) = 1, we obtain that
EQ S2 | S1 = S1u = S0 u2 q + S0 ud (1 − q)
.
= S0 u [uq + d (1 − q)]
= S0 u = S1u ,
hence (1.5) is satisfied. In a similar way, it is easy to check that also (1.6) is true.
Consequently, (Sn )n=0,1,2 is a martingale with respect to the probability
measure Q defined above with q = 0.5.
4. We
need to establish if there exist û > 1 and d̂ > 0 such that the process
Ŝn defined below is a martingale with respect to P (and to the natural
n=0,1,2
filtration):
S0 û2
S0 û

Ŝ0 = S0 = 8 S0 ûd̂

.
S0 d̂

S0 d̂ 2
− − − − −− − − − − −− − − − − −−
1 year (Ŝ1 ) 2 years (Ŝ2 )
Proceeding as above, we need to check

if there exist û > 1 and d̂ > 0

satisfying E Ŝ1 Ŝ0 = Ŝ0 and E Ŝ2 Ŝ1 = Ŝ1 . Or, equivalently, if the

following system in the unknowns û, d̂ admits solutions:
⎧
⎨ S0 ûp + S0 d̂ (1 − p) = S0
. S û2 p + S0 ûd̂ (1 − p) = S0 û
⎩ 0
S0 ûd̂p + S0 d̂ 2 (1 − p) = S0 d̂
Since the previous system is equivalent to ûp + d̂ (1 − p) = 1, we find
ûp + d̂ − d̂p = 1
. 1−d̂+d̂p
û = p .
Taking,
for instance,
d̂ = 0.8, we obtain û = 1.3 > 1. The pair
û = 1.3; d̂ = 0.8 is therefore one among the (infinitely many) pairs of factors

for which the corresponding process Ŝn is a martingale with respect
n=0,1,2
to P .
Exercise 1.5 Consider the stock of Exercise 1.4 and assume that in each period the
price may increase with probability p and decrease with probability (1 − p).
1.3 Proposed Exercises 15
Let X1 and X2 be two random variables defined as

1; if S1 = S1u
X1 =
.
0; if S1 = S1d
⎧
⎨ 2; if S2 = S2uu
X2 = 1; if S2 = S2ud
⎩
0; if S2 = S2dd
1. Write P S2 = S2uu as a function of P (X2 = 2).

2. Establish if (Xn )n=1,2 is a binomial process.
Solution
1. By the definition of X2 it follows that
P (X2 = 2) = P S2 = S2uu = p2
.

P (X2 = 1) = P S2 = S2ud = 2p (1 − p)

P (X2 = 0) = P S2 = S2dd = (1 − p)2 .
2. We have to check if (Xn )n=1,2 is a binomial process, i.e. if X1 ∼ Bin (1; p) and
X2 ∼ Bin (2; p).
It is immediate to see that X1 ∼ Bin (1; p). By definition of a binomial
variable with parameters n = 2 and p and using the probabilities computed
above, it follows that X2 ∼ Bin (2; p) .
Consequently, (Xn )n=1,2 is a binomial process.
1.3 Proposed Exercises
Exercise 1.6 Let (Wt )t≥0 be a standard Brownian motion.

Consider two stocks whose associated gains (difference between current stock
price and buying price) evolve as
Xt = μt + σ Wt
.
Yt = μ1 t + σ1 Wt ,
with X0 = 1, Y0 = 10, μ = 20, μ1 = 10, σ = 10 and σ1 = 20 euros per year.

Taking into account the stocks above, consider also a derivative of value
Zt = Xt2 Yt
.
at time t.
1. Suppose we buy the derivative only if in 2 years its expected value will exceed
10X02 Y0 (that is, 10 times its current price). Decide whether we eventually buy
the derivative or not.
2. Compute the probability of having a positive net gain (Z1 − Z0 ) in 1 year or the
probability of having a net gain (Z2 − Z1 ) between the first and the second years
greater than 100 euros.
3. Compute the probability of both events occurring (Z1 − Z0 > 0 and Z2 − Z1 >
100). Are such events independent?
Exercise 1.7 Let (Wt )t≥0 be a standard Brownian motion. Consider a stock whose
price evolves as the following stochastic process (Yt )t≥0 :
Yt = Y0 eμt+σ Wt .
.
1. Compute P Yt2 ≥ Yt , when Y0 = 4, t = 2, μ = 0.1 and σ = 0.4.

2. Establish if Y2t and
Y
2
t have thesame
distribution
when Y0 = 1.
Y82 Y82
3. Compute E ln Y16 and V ln Y16 when Y0 = 1.
[Hint: remember that E [Wt Ws ] = min (s; t).]
Chapter 2
Portfolio Optimization in Discrete-Time
Models
We restrict our interest to one-period models, and in this case the “portfolio”, which
we are going to define formally below, is characterized through its composition
at the initial time, and the returns of different assets are assumed to be random
variables. A systematic exposition of the notions briefly summarized below can
be found in the textbooks by Barucci [3], Capiński and Zastawniak [10] and
Luenberger [30]. A rich collection of examples and exercises on multi-period
models is provided in the textbook [36].
Let us consider a market model where n assets are traded, and whose values at
a prescribed date are represented by n random variables .S1 , S2 , . . . , Sn . We shall
denote by .S the random vector .S = (S1 , S2 , . . . , Sn ) with components .S1 , S2 , . . . ,
.Sn .
We define portfolio a vector .v = (v1, v2 , . . . vn ) with each component .vi identi-

fying the number of units of the asset with value .Si . The portfolio value V is given
by the scalar product of the two vectors .v and .S, i.e.:

n
V = v · ST =
. vj Sj .
j =1
More often a portfolio is characterized by the vector of its relative weights

w = (w1 , w2 , . . . wn ), where the components .wi are defined by:
.
vi Si vi Si
wi
. = n . (2.1)
V j =1 vj Sj

18 2 Portfolio Optimization in Discrete-Time Models
One can immediately verify that the weights .wi sum up to 1. A negative value of
the component .wi denotes a short position in the asset with value .Si . The set of
portfolios satisfying the normalization condition on the weights .wi is called the set
of admissible portfolios.
The portfolio optimization problem consists, briefly, in finding a portfolio whose
expected utility is the maximum possible among all admissible portfolios. The utility
function whose expectation must be maximized should be chosen so to describe the
investors preferences, and it should take into account their risk-aversion attitude.
A utility function is a nondecreasing, concave function .U : R → [−∞, +∞) of
class .C 2 . Because of their simplicity, the most commonly adopted utility functions
are the following:
• exponential utility:
U (x) = 1 − exp(−αx),
. α > 0; (2.2)
• logarithmic utility:

−∞, x≤0
U (x) =
. ; (2.3)
ln(x), x>0
• power utility:

−∞, x≤0
U (x) =
. , 0 < α < 1. (2.4)
xα , x>0
In the simplest formulation, the portfolio optimization problem can be reduced

to the problem of finding a portfolio with the maximum expected return and the
minimal variance, assuming the latter is adopted as a risk measure associated to the
portfolio under consideration. A portfolio with minimum variance with respect to
all other portfolios with the same expected return, or with maximum expected return
with respect to all portfolios with the same variance, is called efficient.
If we denote by .rK the return of portfolio K, .ri (.i = 1, . . . , n) the return of asset
.Si , .μi its expected value and .σij (.i, j = 1, . . . , n) the covariance between .ri and .rj ,
then the expected return and the variance1 of portfolio K can be easily proved to be
given by:

n
n
.E(rK ) = wi E(ri ) = wi μi. (2.5)
i=1 i=1

n
n
n
n
V ar(rK ) = wi2 V ar(ri ) + wi wj Cov(ri , rj ) = wi wj σij .
i=1 i=1,...,n j =1 i=1 j =1
i=j
(2.6)
The portfolio with minimum variance (with no assigned expected return) can be
determined via the following formula:
uC−1
w=
. , (2.7)
uC−1 uT
where .u is the n-dimensional vector with components equal to 1 and .C is the

covariance matrix of returns.
The portfolio with minimum variance among those with expected return .μV can
be determined via the following formula:

1 uC−1 mT −1 uC−1 uT 1
det uC + det mC−1
μV mC−1 mT mC−1 uT μV
w=
. , (2.8)
uC−1 uT uC−1 mT
det
mC−1 uT mC−1 mT
where .m = (μ1 , μ2 , . . . μn ) is the vector of the expected returns of the n assets.

The set of efficient portfolios with expected return .μV can also be characterized
as the solution of the following:
⎧ n
⎨ j =1 σij wj − λ1 μi − λ2 = 0, i = 1, . . . , n
n
. wi μi = μV
⎩ i=1
n
i=1 wi = 1
or, equivalently,
⎧
⎨ C · wT − λ1 m − λ2 = 0
. w · mT = μV ,
⎩
w · 1T = 1
1 In contrast to the other chapters, to avoid confusion with the notation adopted for the portfolio
value we shall denote the variance of a random variable by .V ar (·) and not by .V (·).
where .λ1 , λ2 are the Lagrange multipliers relative to the last two constraint
equations. The set of efficient portfolios obtained by varying expected returns is
called the efficient frontier.
Sometimes the portfolio optimization problem includes some constraints on the
portfolio weights; the most typical example of this situation is the no-short-selling
constraint that requires all the portfolio weights to be nonnegative, i.e. .wi ≥ 0
for .i = 1, . . . , n. When the portfolio optimization problem consists in variance
minimization with a prescribed expected return, and the no-short-selling constraints
are imposed, the problem can be identified as a quadratic programming problem.
This kind of problem, in general, cannot be solved explicitly without the support of
a computer program (there are several available on the market), unless the dimension
of the problem is small enough to allow explicit calculations.
The Mutual Funds Theorem is another fundamental result of mathematical
portfolio theory: it states that given a market with n assets, it is always possible
to determine two funds (i.e. two portfolios) such that every efficient portfolio (as
far as the expected return and variance are concerned) can be expressed as a linear
combination of them. In order to compute these two portfolios, it is sufficient to find
two solutions of the optimization problem for two arbitrary values of the Lagrange
multipliers .λ1 , λ2 . In general the two solutions determined by this procedure will
violate the normalization condition on the weights, but this can be easily amended
by simply dividing the weights obtained by their sum. We shall provide an example
in order to illustrate the method just outlined.
The Capital Asset Pricing Model (CAPM) assumes as a starting point the linear
dependence of the return of a generic portfolio K (the model postulates the same
relation for all different returns of each asset) on the return .rM of a benchmark
portfolio, called market portfolio:
rK = rf + βK (rM − rf ) + εi ,
. (2.9)
where .rf is the risk-free interest rate and .εi is a zero-mean Gaussian random
variable. By simply taking the expectation of each side of the previous equality,
one obtains:
.μK = E(rK ) = rf + βK (μM − rf ). (2.10)
The (straight) regression line for the present model is identified by the coeffi-
cients .βK , .αK , where .βK and .αK can be obtained by the following formulas:
Cov(rM , rK )
βK =
.
2
. (2.11)
σM
αK = μK − βK μM . (2.12)
Exercise 2.1 Determine the expected return and the variance of the portfolio
formed by the two assets S1 , S2 with weights w1 = 0.6 and w2 = 0.4. The assets’
returns are described by the following scheme:
Scenario Probability r1 r2
ω1 0.1 −20% −10%
.
ω2 0.4 0% 20%
ω3 0.5 20% 40%
Solution Since the expected returns of the two assets are given by:
E(r1 ) = 0.1 · (−0.2) + 0.4 · 0 + 0.5 · 0.2 = 0.08

.
E(r2 ) = 0.1 · (−0.1) + 0.4 · 0.2 + 0.5 · 0.4 = 0.27,
the expected return of the portfolio K with weights w1 = 0.6 and w2 = 0.4 for the
assets S1 and S2 is then:
E(rK ) = w1 E(r1 ) + w2 E(r2 ) = 0.6 · 0.08 + 0.4 · 0.27 = 0.156.

.
In order to calculate the portfolio variance it is necessary to know the covariance

matrix of the two returns. In the present case, we obtain:
V ar(r1 ) = E r12 − (E (r1 ))2

.
= 0.1 · (−0.2)2 + 0.4 · 02 + 0.5 · (0.2)2 − (0.08)2 = 0.0176
V ar(r2 ) = E r22 − (E (r2 ))2
= 0.1 · (−0.1)2 + 0.4 · (0.2)2 + 0.5 · (0.4)2 − (0.27)2 = 0.0241

Cov(r1 , r2 ) = E (r1 r2 ) − E (r1 ) E (r2 )
= 0.1 · (−0.2) (−0.1) + 0.4 · 0 · 0.2 + 0.5 · 0.2 · 0.4 − 0.08 · 0.27
= 0.0204
Hence
V ar(rK ) = w12 V ar(r1 ) + w22 V ar(r2 ) + 2w1 w2 Cov(r1 r2 )

.
= (0.6)2 · 0.0176 + (0.4)2 · 0.0241 + 2 · 0.6 · 0.4 · 0.0204 = 0.019984.

Exercise 2.2 Suppose that the returns r1 , r2 of the two assets S1 , S2 are as in the
following scheme:
Scenario Probability r1 r2
ω1 0.2 −10% 10%
.
ω2 0.3 5% −2%
ω3 0.5 20% 15%
1. Compare the risk (assume the variance as a risk measure) of a portfolio K

composed by the two assets above with weights w1 = 0.3, w2 = 0.7, with
the risk of the two single assets considered separately.
2. By using the same table, determine the composition of a portfolio K ∗ still
consisting of the two previous assets, but with expected return equal to 9%.
Calculate the variance of the portfolio K ∗ .
Solution
1. In strict analogy with Exercise 2.1, one obtains easily
E (r1 ) = 0.095; E (r2 ) = 0.089

. V ar(r1 ) = 0.0137; V ar(r2 ) = 0.0054
Cov (r1 , r2 ) = 0.0042.
Then
Cov(r1 , r2 )
ρ12 = √
. = 0.49.
V ar(r1 )V ar(r2 )
The variance of the portfolio K is then given by:
V ar(rK ) = w12 V ar(r1 ) + w22 V ar(r2 ) + 2ρ12 w1 w2 V ar(r1 )V ar(r2 )

.
= (0.3)2 · 0.0137 + (0.7)2 · 0.0054

√
+2 · 0.49 · 0.3 · 0.7 · 0.0137 · 0.0054
= 0.0056.
2. We have to find a new portfolio K ∗ with expected return equal to 9%. More
precisely, we have to compute the weights corresponding to the two assets in the
new portfolio K ∗ .
Reminding that E(r1 ) = 0.095 and E(r2 ) = 0.089, the two required weights
w1∗ , w2∗ must satisfy the following system of equations:

w1∗ · 0.095 + w2∗ · 0.089 = 0.09
. .
w1∗ + w2∗ = 1
The solution is then given by
1 5
w1∗ =
. , w2∗ = .
6 6
The weights for the two assets S1 and S2 in the new portfolio K ∗ are then
w1∗ = 16 and w2∗ = 56 .
The variance of the return of K ∗ is then given by:
2 2
V ar(rK ∗ ) = w1∗
. V ar(r1 ) + w2∗ V ar(r2 ) + 2w1∗ w2∗ Cov(r1 r2 )
5
= (1/6)2 · 0.0137 + (5/6)2 · 0.0054 + 2 · · 0.0042 = 0.0053.
36
Exercise 2.3 Let three risky assets be given. Their returns have expectations and
covariance matrix described in the following scheme:
μ1 = 0.20 σ1 = 0.25 ρ12 = −0.20

. μ2 = 0.12 σ2 = 0.30 ρ23 = 0.50
μ3 = 0.15 σ3 = 0.22 ρ13 = 0.30
Find the minimum variance portfolio and compute its expected return and
variance.
Solution Let us recall the formula providing the weight vector w of the minimum
variance portfolio:
uC−1
w=
. ,
uC−1 uT
where C is the returns covariance matrix and u is the vector with all components
equal to 1. In the present case, we have:
⎛ ⎞
0.0625 −0.015 0.0165
.C = ⎝ −0.015 0.09 0.033 ⎠ .
0.0165 0.033 0.048
Since the inverse matrix C−1 of C is:

⎛ ⎞
21.497 8.4132 −13.174
C−1
. = ⎝ 8.4132 18.1487 −15.369 ⎠ ,
−13.174 −15.369 35.928
we can easily obtain the weights of the minimum variance portfolio:
w1 = 0.474,
. w2 = 0.317, w3 = 0.209.
The expected return and the variance of this portfolio are then
μV = w1 μ1 + w2 μ2 + w3 μ3 = 0.164
.
σV2 = w12 σ12 + w22 σ22 + w32 σ32 + 2w1 w2 Cov (r1 , r2 )
+2w1 w3 Cov (r1 , r3 ) + 2w2 w3 Cov (r2 , r3 )
= 0.0283.
Exercise 2.4 Consider a portfolio K with return rK and the market portfolio KM
with return rM , where rK and rM take the following values:
Scenario Probability rK rM
ω1 0.3 3% 8%
. ω2 0.2 2% 7%
ω3 0.3 4% 10%
ω4 0.2 1% 9%
Find the coefficients βK and αK of the regression line.

Solution From the table above we obtain
μK = E(rK ) = 0.027; μM = E(rM ) = 0.086;

.
2 = 0.000124;
σM Cov(rK , rM ) = 0.000058.
By applying directly the formulas providing the coefficients βV and αV , we

obtain
Cov(rK , rM ) 0.000058
βV =
.
2
= = 0.467
σM 0.000124
αK = μK − βK μM = 0.027 − 0.467 · 0.086 = −0.01316.
Exercise 2.5
1. Consider two assets S1 , S2 with jointly normally distributed returns of expected
value
√ √ E(r1 ) = μ1 = 0.20, E(r2 ) = μ2 = 0.16, σ1 =
and standard deviation
V ar(r1 ) = 0.30, σ2 = V ar(r2 ) = 0.36, respectively, and with correlation
coefficient ρ12 = −0.5.
Determine the portfolio maximizing the expected utility, assumed to be the

exponential utility function:
U (x) = 1 − exp(−αx)
.
with parameter α = 0.04.

2. Consider now a new portfolio J made by the same asset S1 of the previous
item and another asset B (riskless) with return rB = 0.06. Find the optimal
composition of the new portfolio with respect to the same utility function as in
item 1. (exponential utility).
Solution
1. Denote by K the portfolio with weights w1 = x (in the asset S1 ) and w2 = 1 − x
(in the asset S2 ). The expectation and the variance of the return of K are given
by
E[rK ] = E [w1 r1 + w2 r2 ] = 0.20 · x + 0.16 · (1 − x) = 0.16 − 0.04x

.
V ar(rK ) = V ar (w1 r1 + w2 r2 )
= (0.3)2 x 2 + (0.36)2 (1 − x)2 − 2 · 0.5 · 0.30 · 0.36 · x(1 − x)
= 0.3276 · x 2 − 0.828 · x + 0.1296.
If a random variable Z is normally distributed, then exp(Z) is distributed

according to a log-normal and the following relation holds:

1
.E exp(Z) = exp E(Z) + V ar(Z) .
2
Since maximizing the expectation of the exponential utility is equivalent to

maximizing the quantity:

1
E[U (Z)] = 1 − E exp(−αZ) = 1 − exp −αE(Z) + α 2 V ar(Z) ,
.
2
under the assumption of jointly normally distributed returns of both assets, it is

enough to maximize the following expression:
1
E[Z] − αV ar(Z).
.
2
In the present case Z = rK , so we need to find the value x maximizing the

second-order polynomial:
g(x) = 0.16 − 0.04x − 0.04

2 (0.3276x − 0.828x + 0.1296)
2
.
= 0.1574 − 0.0234x − 0.0066x 2 .
It is easy to obtain that the optimal point is x ∗ = −1.776.

2. The expected return of the new portfolio J , identified via its weights w1 = y and
w2 = 1 − y, is then:
E[rJ ] = E [w1 r1 ] + w2 rB = 0.20 · y + (1 − y) · 0.06,

. (2.13)
and its variance is:
. V ar(rJ ) = (0.3)2 · y 2 . (2.14)
The new function to be maximized is now the following:
0.04
f (y) = 0.14·y+0.06−
. (0.3 · y)2 = −0.0018·y 2 +0.14·y+0.06. (2.15)
2
It is immediate to verify that the maximum point is y ∗ = 38.89. The correspond-

ing weight for the riskless asset is 1 − y ∗ = −37.89, that corresponds to a short
position in the non-risky asset B.
Exercise 2.6 Consider the market consisting of three assets S1 , S2 , S3 with returns
r1 , r2 , r3 , uncorrelated and all with unit variance, with expectations E[r1 ] = 1/4,
E[r2 ] = 1/2, E[r3 ] = 3/4, respectively.
Determine the efficient portfolio with expected return r̂, under the constraint of
no short selling (i.e. wi ≥ 0 for i = 1, 2, 3).
Solution Since unilateral constraints are imposed, the problem cannot be solved by
means of a system of equations. It is a quadratic optimization problem with linear
constraints. Since the number of assets involved is small, the problem can be solved
by considering all different pairs of assets.
Denote by K the portfolio with weight wi in the asset Si (for i = 1, 2, 3) and by
rK its return. Since assets’ returns are uncorrelated and all with unit variance, the
expectation and the variance of rK are given by
E[rK ] = w1 E[r1 ] + w2 E[r2 ] + w3 E[r3 ]

.
1 1 3
= w1 ·+ w2 · + (1 − w1 − w2 ) ·
4 2 4
3 w1 w2
= − − ;
4 2 4
V ar(rK ) = V ar (w1 r1 + w2 r2 + w3 r3 )
= w12 V (r1 ) + w22 V (r2 ) + w32 V (r3 )
= w12 + w22 + (1 − w1 − w2 )2
= 2w12 + 2w22 + 1 − 2w1 − 2w2 + 2w1 w2 .
Set x w1 and y w2 .
Our goal is to find a portfolio with return r̂ and with minimum variance. Notice,
however, that (because of the no-short-selling constraints) only target returns r̂ ∈
[ 14 ; 34 ] are admissible.
By the condition on the target return, we get
3 x y
. − − = r̂,
4 2 4
so y = 3−2x−4r̂. We have thus to minimize (under the no-short-selling constraints)

the following
V ar(rK ) = 2x 2 + 2y 2 + 1 − 2x − 2y + 2xy
.
= x 2 + 2(3 − 2x − 4r̂)2 + 1 − 2x − 2(3 − 2x − 4r̂) + 2x(3 − 2x − 4r̂)

= 6x 2 − 16x + 24x r̂ + 12 + 32r̂ 2 − 40r̂ f (x).
Since f (x) = 12x − 16 + 24r̂, the weights

⎧
⎨ x ∗ = 3 − 2r̂
4
. y = 3 − 2x ∗ − 4r = 13
∗ (2.16)
⎩ ∗
z = 1 − x ∗ − y ∗ = 2r̂ − 2
3
are optimal if admissible. It is easy to check that the optimal weights above satisfy
all the constraints if and only if 13 ≤ r̂ ≤ 23 . In such a case, the efficient portfolio
weights are the following:
4 1 2
w1 =
. − 2r̂, w2 = , w3 = 2r̂ − ,
3 3 3
and the portfolio mean square deviation turns out to be σ = 31/9 − 8r̂ + 8r̂ 2 .
Take 23 < r̂ ≤ 34 . Since f (x) = 12x − 16 + 24r̂ ≥ 0 for any x ∈ [0, 1], it is
easy to check that the efficient portfolio weights are the following:
w1 = 0, w2 = 3 − 4r, w3 = 4r̂ − 2
.
√
and the portfolio mean square deviation turns out to be σ = 13 − 40r̂ + 32r̂ 2 .
Take now 23 < r̂ ≤ 34 . Since for any x ∈ [0, 1], it is easy to check that the
efficient portfolio weights are the following:
w1 = 0, w2 = 3 − 4r̂, w3 = 4r̂ − 2,
.
√
Finally, take 14 ≤ r̂ < 13 . In order to have admissible weights one should have
⎧
⎨ x ∈ [0, 1]
. y = 3 − 2x − 4r ∈ [0, 1] (2.17)
⎩
z = 1 − x − y = x + 4r̂ − 2 ∈ [0, 1]
so 2 − 4r̂ ≤ x ≤ 32 − 2r̂. Since 43 − 2r̂ ≤ 2 − 4r̂ for r̂ ≤ 1

3, we deduce that the
efficient portfolio weights are the following:
w1 = 2 − 4r̂, w2 = 4r̂ − 1, w3 = 0
.
√
Exercise 2.7 Consider a portfolio composed by five assets S1 , S2 , S3 , S4 , S5 with
expected returns E[r1 ] = 0.183, E[r2 ] = 0.085, E[r3 ] = 0.121, E[r4 ] = 0.112,
E[r5 ] = 0.096, respectively, and with return covariance matrix
⎛ ⎞
0.0690 0.0279 0.0186 0.0222 −0.0069
⎜ 0.0279 0.0168 0.0078 ⎟
⎜ 0.042 0.0066 ⎟
⎜ ⎟
.C = ⎜ 0.0186 0.0066 0.054 0.0234 −0.0081 ⎟ . (2.18)
⎜ ⎟
⎝ 0.0222 0.0168 0.0234 0.102 −0.168 ⎠
−0.0069 0.0078 −0.0081 −0.168 0.078
Determine two funds whose linear combinations generate any efficient portfolio.
Solution The two funds we are looking for must belong to the set of portfolios
with minimum variance. We briefly recall that the efficient portfolio (with expected
return μK ) weights must satisfy the following linear system:

C · wT − λ1 m − λ2 = 0,
. (2.19)
w · mT = μK ,
together with the normalization condition:

n
. wj = 1. (2.20)
j =1
Here wi are the portfolio weights, w in vector notation, and μi = E[ri ] the expected
returns of assets Si , m in vector notation. C is the returns’ covariance matrix. The
Two Funds theorem implies that, in order to compute an efficient portfolio for every
value of μK , it is sufficient to determine two solutions of (2.19) and (2.20), since
every other solution is a linear combination of them. A simple method to determine
two solutions is that of specifying two possible values for the Lagrange multipliers
λ1 and λ2 , for example we can pick the solution with λ1 = 0 and λ2 = 1, and
the solution with λ1 = 1 and λ2 = 0. If we look for the solution with Lagrange
multipliers λ1 = 0, λ2 = 1, we have to solve the following linear system:
C · v1 T = 1.
. (2.21)
The solution can be easily found to be the following:
v1 = (2.1191; 26.1615; 17.6417; −8.5490; −5.9895).

.
In order to compute the relative weights w1,i we must impose the normalization
condition on the components of the vector v1 , i.e. we must divide each component
by their sum:
vi
w1,i = n
. .
j =1 vj
We obtain then:
w1 = (0.0721; 0.8233; 0.6004; −0.2909; −0.2038).

.
In order to determine the solution with λ1 = 1 and λ2 = 0 we must solve the

second linear system:
C · v2 T = m.
. (2.22)
We obtain:
v2 = (2.0736; 0.9003; 1.7540; −1.0116; −0.6725),

.
that, after normalization, provides the second fund expressed by its relative weights:
w2 = (0.6813; 0.2958; 0.5763; −0.3324; −0.2210).

.
Exercise 2.8 Given two risky assets whose values are normally distributed with
S1 ∼ N (3; 0.5), S2 ∼ N(4; 0.7) and with correlation ρ12 = 0.5, determine the
efficient frontier and the optimal portfolio with respect to the exponential utility
function with parameter α = 1.5.
Exercise 2.9 In the set of all admissible portfolios composed by three assets of
expected returns μ1 = 0.30, μ2 = 0.15, μ3 = 0.18, with standard deviation σ1 =
0.22, σ2 = 0.30, σ3 = 0.26, and correlations ρ12 = 0.34, ρ23 = 0.02, ρ13 = 0.25,
find the minimum variance portfolio and compute its expected return and variance.
Exercise 2.10 In the set of all admissible portfolios composed by the three assets of
the previous exercise and with expected return μV = 20%, determine the minimum
variance portfolio and compute its variance.
Exercise 2.11 Assume that the risk-free interest rate is rf = 6%, the expectation
and the variance of the market portfolio are 10% and 20%, respectively, and the
correlation between a given asset and the market portfolio return is 0.5. Compute
the expected return of the asset considered in a CAPM framework.
Exercise 2.12 Consider a market with three assets S1 , S2 , S3 with expected returns
E[r1 ] = 1/8, E[r2 ] = 1/4, E[r3 ] = 3/8, respectively. Suppose the returns are
uncorrelated and all with unit variance. Find the efficient portfolio with expected
return r̂ = 13 when short-selling is allowed and also when it is not.
Exercise 2.13 Consider a portfolio composed by four assets S1 , S2 , S3 , S4 with
expected returns E[r1 ] = 0.112, E[r2 ] = 0.076, E[r3 ] = 0.191, E[r4 ] = 0.125,
and with return covariance matrix
⎛ ⎞
0.1024 0.05376 −0.0144 0.03456
⎜ 0.05376 0.0441 0.0189 0.00756 ⎟
.C = ⎜ ⎟. (2.23)
⎝ −0.0144 0.0189 0.2025 −0.1296 ⎠
0.03456 0.00756 −0.1296 0.1296
Determine two funds able to generate every efficient portfolio by linear combi-
nation.
Chapter 3
Binomial Model for Option Pricing
In the following, we consider a market model where a non-risky asset (called bond)
and a risky asset (called stock) are available. The bond price is denoted by B, while
the stock price is denoted by S.
Let us focus, first, on a one-period model. This means that any asset on the market
has to be evaluated just at the beginning and at the end of the given time interval.
By convention, let .t = 0 and .t = 1 be the corresponding dates.
The initial prices of the (non-risky and risky) assets are known: .B0 for the bond,
.S0 for the stock. The bond price is deterministic, equal to
B1 = B0 (1 + r)
.
at .t = 1, where r is the risk-free interest rate and the rate is compounded once at the
end of the period. The stock price at .t = 1 is given by
S1 = S0 X,
.
where X indicates a Bernoulli random variable that may assume u and d, with
probability p and .(1 − p), respectively.
The market model described above (known as one-period binomial model) can
be summarized as follows:
For .d < 1 + r < u, the market above is free of arbitrage. Roughly speaking,
it is not possible to have a profit for free (see Chap. 4 for details on arbitrage on
more general market models). As the binomial model is arbitrage-free, the price
of any financial derivative (or contingent claim) on the underlying S is uniquely
determined. This fact can be seen in two different ways. The first is based on the
construction of a portfolio composed by one derivative and by a suitable number of
stock shares rendering the portfolio riskless (hence its dynamics are deterministic).

32 3 Binomial Model for Option Pricing
Bond .B0 .B1 = B0 (1 + r)

S0 u = S1u ; with prob. p
Stock .S0 .S1 =
S0 d = S1d ; with prob. (1 − p)
.t =0 .t =1
Such a strategy is called Delta-Hedging strategy. The second is based on the

construction of a portfolio (composed by a suitable number of bonds and stock
shares) “replicating” the value of the derivative, that is assuming at maturity T the
same value of the derivative in any state of the world. This last strategy is called
replicating strategy (see Chap. 4 for more details).
Under the no-arbitrage assumption, the initial cost .F (S0 ) of the derivative
written on the stock S is given by the following formula:
1
F (S0 ) =
. [qu · F (S0 u) + qd · F (S0 d)] , (3.1)
1+r
where .qu = (1+r)−d

u−d , .qd =
u−(1+r)
u−d = 1 − qu , and .F (S0 u) and .F (S0 d) denote the
payoff of the derivative when the price of the underlying is equal to .S0 u and .S0 d,
respectively. In the following, we will write .φ u = F (S0 u) and .φ d = F (S0 d).
As we could expect, the initial price of the derivative depends on its payoff .F (S1 )
at maturity.
Since .qu , .qd verify .qu , qd ≥ 0 and .qu + qd = 1, they can be seen as the new
probabilities associated to the values .u, d assumed by the random variable X. The
derivative price in formula (3.1) can be thus interpreted as the expected value of
the discounted payoff of the derivative with respect to the new probability measure
induced by .qu , .qd . In other words
1
F (S0 ) =
. EQ [F (S1 )] . (3.2)
1+r
The new probability measure Q is called risk-neutral measure or equivalent

martingale measure. By definition of .qu and .qd , the expected value (with respect
to Q) of the discounted final value of the stock is indeed equal to its current value:
1
. [qu S0 u + qd S0 d] = S0 , (3.3)
1+r
which means that the discounted price process of the stock is a martingale under Q.
The above market model can be generalized to a multi-period setting as follows.

Consider a discrete-time model where a bond and a stock are traded. In each period
(from t to .t + 1, for .t = 0, 1, . . . , T − 1), the bond price and the stock price are
assumed to evolve as in a one-period binomial model. In other words:
n up;0 down
Sn
p
1−p
(n−1) up;1 down
... Sn
p
.. ..
S0 u = S1u . .
.. ..
.
p . .
S0
.. ..
1−p . .
S0 d = S1d
1 up;(n−1) down
1−p Sn
. . . p
1−p
0 up;n down
Sn
The market described above is known as multi-period binomial model.

By the Markov property fulfilled by the model, it is easy to check that under the
risk-neutral measure Q
1 1
Sk =
. EQ [Sl |Sk ] = EQ [Sl |Fk ] , 0 ≤ k ≤ l ≤ n. (3.4)
(1 + r)l−k (1 + r)l−k
Once discounted, the process .(Sk )k=0,1,...,n is then a martingale under Q and with
respect to the filtration .(Fk )k=0,1,...,n generated by .(Sk )k=0,1,...,n .
The pricing method for the derivative .F (S) is based on the assumption that,
in the binomial model, any derivative is attainable, namely there always exists a
strategy consisting in a suitable number of bonds and of stocks replicating exactly
the derivative value at any time and in any state of the world. Or, equivalently, there
always exists a portfolio strategy composed by a suitable amount of the derivative
and the underlying, so to “completely hedge” the risk. Such a property is called
completeness of the market (see Chap. 4 for more details on completeness). More
precisely, a market model is said to be complete if any derivative can be attained (or
replicated).
Both the one-period and the multi-period binomial market models described
above are complete. In the multi-period binomial model, the initial price .F (S0 )
of the derivative is given by the expected value of the discounted payoff under the
risk-neutral measure Q, i.e.
1 1
n
n k n−k
F (S0 ) =
. EQ [F (Sn )] = k n−k
k qu qd F (S0 u d ). (3.5)
(1 + r)n (1 + r)n
k=0
Necessary and sufficient conditions for a (quite general) market model to be free
of arbitrage and/or complete are given by the Fundamental Theorems of the Asset
Pricing. What just recalled in a binomial model can be seen as a particular case of
these conditions.
The First Fundamental Theorem of the Asset Pricing establishes that, in “suit-
able” market models, no-arbitrage and existence of an equivalent martingale
measure are equivalent conditions.
For a market that is free of arbitrage, the Second Fundamental Theorem of the
Asset Pricing guarantees the uniqueness of the equivalent martingale measure when
the market model is complete.
As already underlined, the binomial model is complete. Nevertheless, if at any
time the risky asset could assume not only two but three (or more) different values,
such new market model would not any longer be complete. As we will see in
Chap. 4, in such a case, indeed, an arbitrary derivative could not be attainable only
by means of the bond B and of the risky asset S.
European Put and Call options (sometimes called vanilla options) and American
options, i.e. options that can be exercised at any time between the initial 0 and
the maturity T , can be easily evaluated in a binomial model. Concerning European
options, it is sufficient to apply the arguments above to the payoff of a European
Call (respectively, Put) option with strike K, maturity T and written on the stock S:
FCall (ST ) = (ST − K)+ ,

. resp. FP ut (ST ) = (K − ST )+ .
It is also worth to mention that the price (at time .t = 0, 1, . . . , T ) of a European

Call option (.Ct ) and of a European Put option (.Pt ) with the same characteristics are
related by the so called Put-Call Parity:
K
Ct − Pt = St −
. . (3.6)
(1 + r)T −t
The pricing of some other options whose value depends not only on the
underlying value at maturity, but also on its value at intermediate times between
0 and T , will be illustrated in some exercises of the present chapter and of the next
ones.
Based on the arguments above, it is also possible to obtain the strategy able to
make the portfolio riskless between two consecutive times: it would be composed
by a long position in the derivative and by .Δ = (Fu − Fd )/(u − d) short positions

in the underlying.
For a detailed treatment and further details on the subject, we refer to Björk [6],
Hull [25], Pliska [37] and Ross [39], among others. A rich collection of examples
and exercises on multi-period models is provided in the textbook [36].
Exercise 3.1 Different options having the same stock as underlying are available
on the market. Suppose that the risk-free interest rate is 4% per year, that the current
stock price is 20 euros and that such price may go up or down by 25% in each of the
next 2 semesters.
1. Assume that the stock price may go up or down in each semester with (objective)
probability of 50%. Verify whether the exercise of a European Call option with
strike of 18 euros and with maturity of 6 months is more likely than the exercise
of a Call option as the one above, but with maturity of 1 year.
2. Compare the prices of the Call options of item 1.
3. Discuss whether the options prices of items 1. and 2. would change if:
(a) the objective probability of an increase in the stock price was of 80%
(consequently, 20% of a decrease);
(b) if the strike was of 20 euros. If yes, compute the price of a European Call
option with maturity 6 of months and with strike of 20 euros.
Solution We deduce that the stock price evolves as follows:
31.25 = S0 u2

S0 u = 25

S0 = 20 18.75 = S0 ud

.
S0 d = 15

11.25 = S0 d 2
− − −− − − − − −− − − − − −−
0 6 months T = 1 year
where u = 1.25 (growth factor) and d = 0.75.

1. Remember that a European Call option is exercised when the underlying price is
greater (or equal) than the strike. With this fact in mind, it is easy to compute the
probability to exercise the options above.
Before doing that, it is worth to underline that the behavior of the stock price
under the “real” probability is
25 p

20

15 1−p
− − −− − − − − − − − − −− − − − − − − − − − − −−
0 T = 6 months with probability
.
31.25 p2

20 → 18.75 2p(1 − p)

11.25 (1 − p)2
−−− − − − − − − − − −− − − − − − − − − − − −−
0 T = 1 year with probability
where p = 1 − p = 0.5.
It follows that the option with maturity (T ) of 6 months and with strike of 18
euros is exercised with probability
P (ST ≥ K) = P (ST ≥ 18) = P (ST = 25) = 0.5,

.
while the option with maturity (T ) of 1 year and with strike of 18 euros is
exercised with probability
P (ST ≥ K) = P (ST ≥ 18)

.
= P ({ST = 18.75} ∪ {ST = 31.25})

= p2 + 2p (1 − p) = (0.5)2 + 2 · 0.5 · 0.5 = 0.75.
The exercise of a European Call option with strike of 18 euros and with maturity
of 1 year is then more likely than the exercise of a Call option as the one above
but with maturity of 6 months.
2. In order to compare the prices of the options taken into account, we look for the
“risk-neutral” probability Q that will be useful to evaluate the options. Such a
probability Q corresponds to:
√
(1 + r)1/2 − d 1.04 − 0.75
.qu = = = 0.54
u−d 1.25 − 0.75
qd = 1 − qu = 0.46,
because r = 0.04 is the annual interest rate and the time intervals are of 6 months.
For the European Call (A) with maturity T = 6 months we get:
S0 u = 25 u =7
φA
qu
S0 = 20
. qd
S0 d = 15 d =0
φA
− − −− − − − − −− − − − − − − −−
0 T = 6 months Payoff of option A
since the payoff of a European Call is given by

ST − K; if ST ≥ K
. (ST − K)+ = .
0; if ST < K
It follows that the initial price of the European Call option A is given by
1 1
C0A =
. qu · φA
u
+ qd · φA
d
=√ [0.54 · 7 + 0] = 3.71 euros.
(1 + r)1/2 1.04
For the European Call (B) with maturity T = 1 year, we have
S0 u2 = 31.25 φBuu = 13.25

qu
S0 u = 25
qu qd
S0 = 20 S0 ud = 18.75 φBud = 0.75
. qd qu
S0 d = 15
qd
S0 d 2 = 11.25 φBdd = 0
− − −− − − − − −− − − − − − − −− − − − − − − − − −−
0 6 months 1 year Payoff of option B
To compute the initial cost of such an option we can proceed in two different
ways:
• translate the problem in a one-period setting by considering only the initial
date (0) and the maturity T . Consequently,
0 T = 1 year with probability payoff

−−− − − − − −− − − − − − − − − −− − − − − −−
31.25 (qu )2 13.25 = φBuu
.
20 −→ 18.75 2qu qd 0.75 = φBud

11.25 (qd )2 0 = φBdd
It follows that the initial price of the European Call B is given by
1 2 uu
C0B =
. qu φB + 2qu qd φBud + qd2 φBdd
1+r
1
= (0.54)2 · 13.25 + 2 · 0.54 · 0.46 · 0.75 + 0 = 4.07 euros;
1.04
• proceed backwards in time as illustrated below.
u at the
We first consider the sub-tree in bold and compute the “price” C6m
node “S6m ” at time t = 6 months.
u
S0 u2 = 31.25 φBuu = 13.25

S0 u = 25

S0 = 20 S0 ud = 18.75 φBud = 0.75
.
S0 d = 15

S0 d 2 = 11.25 φBdd = 0
− − −− − − − − −− − − − − − − −− − − − − − − −−
This reduces to option pricing in a one-period model. Hence
1
u
C6m
. = qu · φBuu + qd · φBud
(1 + r)1/2
1
= √ [0.54 · 13.25 + 0.46 · 0.75] = 7.3545.
1.04
The same procedure can be applied to the sub-tree below:
S0 u2 = 31.25 φBuu = 13.25

S0 u = 25

S0 = 20 S0 ud = 18.75 φBud = 0.75
.
S0 d = 15

S0 d2 = 11.25 φBdd = 0
− − −− − − − − −− − − − − − − −− − − − − −−
It follows that:
1
d
C6m
. = qu · φBud + qd · φBdd
(1 + r)1/2
1
= √ [0.54 · 0.75 + 0] = 0.397.
1.04
Finally, there remains to deduce the initial price of the Call from the prices
u and C d just computed. In other words,
C6m 6m
u = 7.3545
C6m qu
C0B =???
.
d
C6m = 0.397 qd
− − − − −− − − − − − − −− − − − − − − −−
0 t = 6 months with prob.
Also in the present case the option pricing reduces to the pricing in a one-
period setting, hence:
1
C0B =
. qu · C6m
u
+ qd · C6m
d
(1 + r)1/2
1
= √ [0.54 · 7.3545 + 0.46 · 0.397] = 4.07 euros.
1.04
3.
(a) The prices of options A and B studied in items 1.–2. do not change on varying
the real probability. Such a probability, indeed, does not influence the option
pricing.
(b) The price of the European Call option with maturity of 6 months decreases
as the strike increases.
Recall that the initial price of the European Call option A with maturity
T = 6 months and with strike of K = 18 euro is C0A = 3.85 euros.
Consider now the same option as before except for the strike, now of
K̃ = 20 euros. We get
S0 u = 25 φ̃ u = 5
qu
S0 = 20
. qd
S0 d = 15 φ̃ d = 0
− − −− − − − − −− − − − − − − − − − − − − − − −−
0 T = 6 months Payoff of option with strike K̃ = 20
Consequently, the price of the European Call (as A) but with strike K̃ is
given by
1 1
C̃0 =
. qu · φ̃ u + qd · φ̃ d = √ [0.54 · 5 + 0] = 2.65 < C0A .
(1 + r)1/2 1.04
Exercise 3.2 Different options having the same stock as underlying are available
on the market. Suppose that the risk-free interest rate is 4% per year, that the current
stock price is 20 euros and that such price may go up or down by 25% in each of the
next 3 semesters.
1. Evaluate a European Call option with strike of 18 euros and with maturity of
18 months.
2. Evaluate the corresponding Put option.
3. Consider the Call and Put options above.
(a) How many shares of the stock would you be able to buy, and how much
money could you invest in a bank account at the initial time, if you sold eight
Call options and bought one Put option (supposing it is not possible to buy
fractions of shares)?
(b) Compute the profit (or the loss) of the investment above if the stock price
grows in the next 3 semesters. What about the profit (or the loss) if the stock
price grows in 2 semesters and decreases in one semester?
4. Evaluate the Call option of item 1. when the length of the periods is 1 year instead
one semester.
Solution
1. Recall that, when the model consists of n periods (months, years, . . . ) and rp
denotes for the interest rate per period1 (monthly, annual, . . . ), we have:
F (S0 ) = initial option price

.
1
n
n k n−k (k) up;(n−k) down
= n qu · qd · φ ,
1 + rp k=0 k
where φ (k) up;(n−k) down stands for the payoff of the option when the price of the
underlying has moved k times up and (n − k) times down.
In the present case, we have that time periods are of 6 months and that the
price of the underlying evolves as follows
39.0625 φ uuu = 21.0625

31.25

25 23.4375 φ uud = 5.4375

20 18.75
.
15 14.0625 φ udd = 0

11.25

8.4375 φ ddd = 0
−− − − −− − − −− − − − − −− − − − − − − −−
0 6 months 1 year 18 months Payoff of the Call
1 Iftime and interest rate rp refer to a different unit of time, it is enough to transform the interest
rate according to compounding. For instance, if ryear is the annual interest rate and time intervals
are months, then the monthly interest rate (equivalent to ryear ) is given by
1/12
.rmonth = 1 + ryear − 1.
The monthly interest rate (equivalent to r = 0.04 per year) is rm =

(1 + r)1/12 − 1 = (1.04)1/12 − 1, so (1 + rsem )3 = (1 + r)3/2 . From the
arguments above it follows that
C0 = price of the European Call with maturity T (18 months )

.= 3/2 (qu ) · φ
1 3 uuu + 3 (q )2 q · φ uud + 3q (q )2 · φ udd + (q )3 · φ ddd
(1+r) u d u d d
= 1
(0.54)3 · 21.0625 + 3 · (0.54)2 · 0.46 · 5.4375 + 0 + 0 = 5.19,
(1.04)3/2
where qu = 0.54 and qd = 0.46 are the same as in Exercise 3.1.

2. We can evaluate the European Put option in two different ways.
First, we can proceed as above. Since the payoff of the European Put option is

K − ST ; if ST ≤ K
. (K − ST )+ = max{K − ST ; 0} = ,
0; if ST > K
it follows that
39.0625 (qu )3 ut = 0
φPuuu

23.4375 3 (qu )2 qd ut = 0
φPuud
20
. 14.0625 3qu (qd )2 ut = 3.9375
φPudd

8.4375 (qd )3 ut = 9.5625
φPddd
− − −− − − − − − − −− − − − − −− − − − − − − −−
0 T = 18 months with probability Payoff of the Put
Hence
P0 = price of the European Put with maturity T

.
1
= (qu )3 · φPuuu
ut + 3 (qu ) qd · φP ut
2 uud
(1 + r)3/2

+3qu (qd )2 · φPudd
ut + (qd )3
· φ ddd
P ut
1
= 0 + 0 + 3 · 0.54 · (0.46)2 · 3.9375 + (0.46)3 · 9.5625
(1.04)3/2
= 2.15 euros.
An alternative way to evaluate the Put option above is to apply the Put-Call
Parity true for European options with the same maturity and strike and written
on the same underlying, that is
K
C0 − P0 = S0 −
. .
(1 + r)T
Due to the equality above, we obtain immediately
K ∼
P0 = C0 − S0 +
. = 2.15 euros.
(1 + r)T
3. By the previous items, C0 = 5.19 and P0 = 2.15 euros.

(a) If we sell eight Call options and buy one Put option as above, we dispose of
π = 8 · C0 − P0 = 8 · 5.19 − 2.15 = 39.37 euros.

.
Under the restrictions imposed, we can buy only one share of the underlying
(its initial price is indeed S0 = 20 euros) and invest the remaining budget of
19.37 euros at risk-free interest rate.
(b) If the stock price grows in each of the next three semesters (hence, ST =
S0 u3 = 39.0625 euros), the profit (or loss) of our strategy is
8 call sold −8 (ST − K)+ = −8 · φ uuu = −168.5

1 put bought (K − ST )+ = 0
. Profit/loss by the stock (ST − S0 ) = 19.0625
Cash invested in a bank account 19.37 · (1.04)3/2 = 20.54375
T otal prof it / loss −128.894 euro
If the stock price grows in two of the next three semesters and decreases in
one (hence ST = S0 u2 d = 23.4375 euros), the profit (or loss) of our strategy
is
8 calls sold −8 (ST − K)+ = −8 · φ uud = −43.5

1 put bought (K − ST )+ = 0
. Profit/loss by the stock (ST − S0 ) = 3.4375
Cash invested in a bank account 19.37 · (1.04)3/2 = 20.54375
T otal prof it / loss −19.5188 euros
4. With annual periods instead of semestral ones, the stock price would evolve as
before:
39.0625 φ uuu = 21.0625

31.25

25 23.4375 φ uud = 5.4375

20 18.75
.
15 14.0625 φ udd = 0

11.25

8.4375 φ ddd = 0
−−− − − −− − − −− − − − − −− − − − − − − −−
0 1 year 2 years 3 years Payoff of the Call
while the maturity and the equivalent martingale measure Q would change. For
annual periods, indeed, Q would correspond to
(1 + r) − d 1.04 − 0.75
qu∗ =
. = = 0.58
u−d 1.25 − 0.75
qd∗ = 1 − qu∗ = 0.42,
hence the price of the European Call with maturity 3 years would be given by
1 2
qu∗ · φ uuu + 3 qu∗ qd∗ · φ uud
(3y) 3
C0
. =
(1 + r)3
2 3
+3qu∗ qd∗ · φ udd + qd∗ · φ ddd
1
= (0.58)3 · 21.0625 + 3 · (0.58)2 · 0.42 · 5.4375
(1.04)3
+0 + 0] = 5.70.
Exercise 3.3 Consider a stock whose current price is S0 = 40 euros. In each of the
next two 4-month periods, the stock price may move up by 4% or down by 2%. The
risk-free rate available on the market is of 4% per year.
Compute the initial price of an option with maturity of T = 8 months and with
payoff
+
(K − S0 )2 (K−S0 )2
0; ST <
.φ = ST − = 10
(K−S0 )2 (K−S0 )2
10 ST − 10 ; ST ≥ 10
when the strike K is of 60 euros.

Solution We know that on each of the next 4-month periods the stock price may
increase by a growth factor u = 1.04 or decrease by a factor d = 0.98. Accordingly,
the stock price evolves as below:
43.264 = S0 u2

41.6

40 40.768 = S0 ud
.
39.2

38.416 = S0 d 2
−−− − − −− − − − − − − −−
0 4 months 8 months
The payoff of the option may thus take the following values:
+ +
(60 − 40)2 202
.φ uu
= STuu − = 43.264 − = 3.264
10 10
+ +
(60 − 40)2 202
φ ud
= STud − = 40.768 − = 0.768
10 10
+ +
(60 − 40)2 202
φ dd
= STdd − = 38.416 − = 0.
10 10
Consequently,
43.264 = S0 u2 3.264 = φ uu

41.6

40 40.768 = S0 ud 0.768 = φ ud
.
39.2

38.416 = S0 d 2 0 = φ dd
−−− − − −− − − − − − − −− − − − − − − −−
0 4 months 8 months Payoff of the option
The risk-neutral probability Q to be used for option pricing corresponds to:
(1 + r)4/12 − d
.qu = = 0.55
u−d
qd = 1 − qu = 0.45,
since periods are of 4 months while r is the annual interest rate.

From the arguments above, the option with payoff φ now costs
1
F (S0 ) =
. (qu )2 φ uu + 2qu qd φ ud + (qd )2 φ dd
(1 + r)T
1
= (0.55)2 · 3.264 + 2 · 0.55 · 0.45 · 0.768 + 0 = 1.33 euros.
(1 + r)8/12
Exercise 3.4 We take a short position in a European Call option with maturity
4 months and with strike of 20 euros, having a stock with current price of 20 euros
as underlying. In the next 4 months, the stock price may increase by a growth factor
u = 1.2 or decrease by a factor d = 0.8. The risk-free interest rate available on the
market is of 4% per year.
1. Find the replicating strategy of the option.
2. Compute the initial price of the option.
3. Suppose now that the stock price does not follow a binomial model any more,
but that in 4 months it may either increase by a growth factor u = 1.2, decrease
by a factor d = 0.8 or remain unchanged. Discuss whether it is still possible
to replicate the Call option above only with the underlying and with cash (to be
deposited or borrowed).
Solution
1. The stock price and the option payoff are the following:
S0 u = 24 φu = 4

S0 = 20
.
S0 d = 16 φd = 0
− − −− −−−−−−− − − − − −−
0 T = 4 months Payoff
In order to find the replicating strategy of the option we need to find a number
Δ of shares of the underlying and an amount x of cash so that the seller of
the option holding such a portfolio can be “covered” against losses (due to the
option) at maturity and in any state of the world. We are then looking for Δ and
x such that

ΔS0 u + x (1 + r)T = φ u
. .
ΔS0 d + x (1 + r)T = φ d
The previous system of linear equations is equivalent to

ΔS0 u − ΔS0 d = φ u − φ d
. ,
ΔS0 d + x (1 + r)T = φ d
whose solution is given by:

φ u −φ d
Δ= S0 (u−d)
. .
x= 1
φ d − ΔS0 d
(1+r)T
The replicating strategy is then composed as below:

φ u −φ d
Δ= S0 (u−d) = 20(1.2−0.8) = 0.5
4−0
.
x= 1
[0 − 0.5 · 20 · 0.8] = −7.9
(1.04)4/12
To be covered against losses deriving from the exercise of the option, the seller
has to buy 0.5 shares of the stock and to borrow 7.9 euros.
Since the cost of such a strategy is equal to
ΔS0 + x = 0.5 · 20 − 7.9 = 2.1,

.
the No Arbitrage principle implies that the Call price should be equal to the cost
of the replicating strategy, i.e. C0 = ΔS0 + x = 2.1.
2. We have already computed the option price above by means of the replicating
strategy.
An alternative way to compute C0 is to proceed as in Exercises 3.1 and 3.2.
By the no arbitrage principle, the result has to coincide with the one above.
4/12 −d
Since the equivalent martingale measure corresponds to qu = (1+r) u−d =
0.53 and qd = 1 − qu = 0.47, we obtain indeed
1
C0 =
. qu φ u + qd φ d
(1 + r)T
1
= [0.53 · 4 + 0] = 2.10 euros.
(1.04)4/12
3. The stock price is now assumed to evolve as follows:
S0 u = 24 φu = 4

S0 = 20 → S0 · 1 = 20 φm = 0
.
S0 d = 16 φd = 0
− − −− − − − − − − −− − − − − −−
0 T = 4 months Payoff
Verifying the existence of a replicating strategy for the option is equivalent

to verifying the existence of a number Δ of shares of the underlying and of an
amount x of cash such that
⎧
⎨ ΔS0 u + x (1 + r) = φ u
T
. ΔS0 + x (1 + r)T = φ m .
⎩
ΔS0 d + x (1 + r)T = φ d
Since the previous system (of 3 linear equations in 2 variables) is equivalent to

the following ones
⎧ ⎧
⎨ ΔS0 u + x (1 + r) = φ u
T ⎪
⎨ ΔS0 u + x (1 + r) = φ
T u
. ΔS + x (1 + r) = φ m
T ΔS0 + x (1 + r)T = φ m
⎩ 0 ⎪
⎩ Δ = φ u −φ d
ΔS0 (u − d) = φ u − φ d S0 (u−d)
⎧ ⎧
⎨ x (1 + r) = φ u − ΔS0 u ⎨ x (1 + r) = −8
T T
. x (1 + r)T = φ m − ΔS0 x (1 + r)T = −10 ,

⎩ ⎩
Δ = 12 Δ = 12
it is evident that there is no solution. Consequently, it is not possible to replicate

the option only by means of the underlying and of cash invested at risk-free rate.
From the arguments above it follows that the market model is incomplete (i.e.
not any option is attainable) and that the option price is not uniquely determined,
in general.
Exercise 3.5 We take a short position in a European Call option with maturity
2 years and with strike of 25 euros, having a stock with current price of 40 euros
as underlying. In each of the next 2 years, the stock price may increase by a growth
factor u = 1.5 or decrease by a factor d = 0.5, while the risk-free interest rate is of
4% per year.
1. Find the replicating trading strategy of the option above and deduce the option
price.
2. Find the replicating strategy of a portfolio formed by two short positions in a
Call with maturity of 1 year, in one long position in a Put with maturity of 1 year
(where both Call and Put options have strike of 25 euros and are written on the
underlying described above) and in one long position in the underlying.
3. Discuss whether the replicating strategy and the option price change if everything
(underlying dynamics, strike, maturity, . . . ) stays the same except for the risk-free
rate that increases to 6% per year.
Solution
1. The stock price evolves as follows:
90 = S0 u2

S0 u = 60

S0 = 40 30 = S0 ud
.
S0 d = 20

10 = S0 d 2
− − −− −−−−− −−−−−−−
0 1 year 2 years
where u = 1.5 (growth factor) and d = 0.5.

First of all, let us find the risk-neutral measure to be used for option pricing.
Such a measure Q corresponds to:
(1 + r) − d 1.04 − 0.5
qu =
. = = 0.54
u−d 1.5 − 0.5
qd = 1 − qu = 0.46,
because both the risk-free rate r and the periods are annual.
We will find simultaneously the option price and the replicating strategy. For
a European Call option with strike of 25 euros:
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 20

10 = S0 d 2 φ dd = 0
− − −− −−−−− − − − − − − − − − − − − − −−
0 1 year 2 years Payoff of the option
To find the replicating strategy of the option, we consider separately any

period and we proceed backwards in time.
We first consider the sub-tree in bold so to compute the option “price” C1u at
the node “S1u ” in t = 1 year.
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 30

10 = S0 d 2 φ dd = 0
− − −− − − − − −− − − − − −− − − − − −−
We have reduced to evaluate an option and/or find its replicating trading

strategy in a one-period model. We obtain that:
1 1
C1u =
. qu · φ uu + qd · φ ud = [0.54 · 65 + 0.46 · 5] = 35.96
1+r 1.04

and that the replicating strategy Δu1 , x1u at time t = 1 and at the node “S1u ” is
given by:

φ uu −φ ud
Δu1 = S1u (u−d)
= 60(1.5−0.5)
65−5
=1
.
x1u = 1
1+r φ
ud − Δ1 S1 d = 1.04
u u 1
[5 − 30] = −24.04
As expected,
.C1u = Δu1 S1u + x1u = Δu1 S0 u + x1u = 35.96.
Proceeding analogously for the sub-tree in bold below:
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 20

10 = S0 d2 φ dd = 0
− − −− − − − − −− − − − − −− − − − − −−
we get
1 1
C1d =
. qu · φ ud + qd · φ dd = [0.54 · 5 + 0] = 2.6
1+r 1.04
and
φ ud −φ dd
Δd1 = = 20(1.5−0.5)
5−0
= 0.25
. S1d (u−d)
x1d = 1
1+r φ
dd − Δ S d = −2.4
d
1 1
d
Also in the present case, we obtain the same result as previously, that is
C1d = Δd1 S1d + x1d = Δd1 S0 d + x1d = 2.6.

.
Finally, we consider the first period:
C1u = 35.96 qu
C0 =???
.
C1d = 2.6 qd
− − −− − − − − −−
0 t = 1 year
It is a one-period case, hence the option price can be computed immediately

as
1
C0 =
. qu · C1u + qd · C1d
1+r
1
= [0.54 · 35.96 + 0.46 · 2.6] = 19.8 euros,
1.04
while the replicating strategy is given by

C u −C d
Δ0 = S01(u−d)
1
= 40(1.5−0.5)
35.96−2.6
= 0.834
.
−1
x0 = (1 + r) C1 − Δ0 S0 d = −13.54
d
Consequently, we find again C0 = Δ0 S0 + x0 = 19.8 euros.

Summing up: the initial price of the option is 19.8 euros and its replicating
trading strategy consists in the following. The seller of the option buys 0.834
stocks and borrows 13.54 euros at t = 0. Then, at t = 1: (a) if the stock price is
increased to S0 u = 60, he buys (1 − 0.834) = 0.166 more stocks and borrows
(24.04 − 13.54) = 10.50 euros more. So, his portfolio consists in 1 stock and in
24.04 euros borrowed. (b) If the stock price is decreased to S0 d = 20, he sells
(0.834 − 0.25) = 0.584 stocks and gives back (13.54 − 2.4) = 11.14 euros of
what he had borrowed. Thus, his portfolio consists in 0.25 stocks and in 2.4 euros
borrowed.
2. We consider now a portfolio formed by two short positions in a Call with maturity
of 1 year, by one long position in a Put with maturity of 1 year (where both
Call and Put options have strike of 25 euros and are written on the underlying
described above) and by one long position in the underlying.
Since the underlying and the Call and Put payoffs read:
S0 u = 60 u
φCall = 35 φPu ut = 0

S0 = 40

.
S0 d = 20 d
φCall =0 φPd ut = 5
− − −− − − − − −− − − − − −− − − − − −−
0 T ∗ = 1 year Call payoff Put payoff
the portfolio π to be replicated is given by
φπ = −2φCall + φP ut + ST ∗
.
⎧ u
⎨ φπ ; if S1 = S1u
=
⎩ d
φπ ; if S1 = S1d
⎧
⎨ −2φCall
u + φPu ut + S1u = −10; if S1 = S1u
=
⎩
−2φCalld + φPd ut + S1d = 25; if S1 = S1d
In order to find a replicating strategy of the above portfolio of options and

stocks, we have to look for a number Δ of shares of the underlying and for an
amount x of cash such that
∗
ΔS0 u + x (1 + r)T = φπu
. ∗ .
ΔS0 d + x (1 + r)T = φπd
In the present case, we obtain

φπu −φπd −10−25
ΔS0 u − ΔS0 d = φπu − φπd Δ= S0 (u−d) = 40(1.5−0.5) = −0.875
. ∗
ΔS0 d = φπd − x (1 + r)T x= 1+r φπ − ΔS0 d =
1 d 40.865
Consequently, the replicating strategy we are looking for consists in selling

0.875 shares of the underlying and in investing 40.865 euros in a bank account.
Such a strategy costs then ΔS0 + x = −35 + 40.865 = 5.865 euros.
3. To find the replicating trading strategy of the option (that exists for sure!), we
recall that the dynamics of the stock, as well as the option payoff, do not depend
on the risk-free rate and are the following:
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 20

10 = S0 d 2 φ dd = 0
− − −− − − − − −− − − − − −− − − − − −−
In order to find the replicating strategy we analyze separately the two periods
and we proceed backwards in time as in the previous item.
First, let us consider the sub-tree in bold so to compute the option price C1u at
the node “S1u ” and at time t = 1:
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 20

10 = S0 d 2 φ dd = 0
− − −− − − − − −− − − − − −− − − − − −−
Under the new risk-free rate r̃, the risk-neutral measure corresponds now to
(1 + r̃) − d 1.06 − 0.5

q̃u =
. = = 0.56
u−d 1.5 − 0.5
q̃d = 1 − q̃u = 0.44.

Accordingly, the replicating strategy Δu1 , x1u to be held at t = 1 and at the node
“S1u ” is now

φ uu −φ ud
Δu1 = S1u (u−d)
= 60(1.5−0.5)
65−5
=1
.
−1
x1u = (1 + r̃) φ − Δ1 S1u d
ud u = (1 + 0.06)−1 [5 − 30] = −23.58
Therefore, C1u = Δu1 S1u + x1u = Δu1 S0 u + x1u = 36.42 euros.

Analogously for the sub-tree in bold below:
90 = S0 u2 φ uu = 65

S0 u = 60

S0 = 40 30 = S0 ud φ ud = 5
.
S0 d = 20

10 = S0 d2 φ dd = 0
− − −− − − − − −− − − − − −− − − − − −−
In the present case, we get

φ ud −φ dd
Δd1 = = 20(1.5−0.5)
5−0
= 0.25
. S1d (u−d)
x1d = (1 + r̃)−1 φ dd − Δd1 S1d d = −2.36
and, consequently, C1d = Δd1 S1d + x1d = Δd1 S0 d + x1d = 2.64 euros.
Finally, we take into account the first period:
C1u = 36.42 q̃u
C0 =???
.
C1d = 2.64 q̃d

− − −− − − − − −−
0 1 year
The replicating trading strategy is given by

C1u −C1d
Δ0 = S0 (u−d) = 40(1.5−0.5) = 0.8445
36.42−2.64
. ,
x0 = 1+r̃ C1 − Δ0 S0 d = −13.44
1 d
so we deduce that C0 = Δ0 S0 + x0 = 20.34 euros.

Exercise 3.6 Consider a market where a bond (corresponding to a risk-free rate of
4% per year) and a stock are traded. The current price of the stock is S01 = 8 euros
that, in 1 year, may move up to 16 euros or to 12 euros, or down to 2 euros.
1. Consider a European Call option on the stock above, with strike of 8 euros and
with maturity of 1 year. Verify if it is possible to replicate such an option only by
means of the underlying and of cash invested (or borrowed) at risk-free rate.
2. Consider another stock of current price S02 = 10 euros and that, in 1 year, may
cost 20, 8 or 6 euros whenever the other stock costs, respectively, 16, 12 or 2
euros.
A new option is also available, with maturity T of 1 year, with the same strike
K as the Call above and with payoff
+
φ = S̃T − K ,
.

ST1 +ST2 S 1 −S 2
where S̃T = 2 − T 4 T .
Verify if it is possible to replicate the option above by means of the two stocks
and of cash invested (or borrowed). If yes, find the cost of the replicating strategy.
Solution
1. The price of the first stock and the payoff of the European Call on the stock, with
maturity of 1 year and strike of 8 euros are the following:
S11,u = 16 u
φCall =8

S01 = 8 → S11,m = 12 m
φCall =4
.
S11,d = 2 d
φCall =0
− − −− − − − − −− − − − − − − −−
0 1 year Payoff of the Call
In order to verify if the Call option above is attainable by means of the

underlying stock and of cash, we need to check if the following system has
solutions or not:
⎧
⎪ 1,u
⎨ ΔS1 + x (1 + r) = φCall
T u
1,m
. ΔS1 + x (1 + r) = φCall
T m ,
⎪
⎩ ΔS 1,d + x (1 + r)T = φ d
1 Call
where Δ represents the number of shares of the underlying to buy or sell and
x the amount of cash to be invested or borrowed.
Replacing T with 1 (year), the previous system (of 3 equations in 2 variables)
is equivalent to the following ones:
⎧ ⎧
⎪ 1,u
⎨ ΔS1 φ+ x (1 + r) = φCall
1,u u
⎪ ΔS
⎨ 1 + x (1 + r)
= φ u
Call ⎪ u −φ m
.
1,u
Δ S1 − S1 1,m
= φCall
u − φCall
m Δ = Call Call
⎪
⎩ ⎪
⎪ S11,u −S11,m
1,d
ΔS1 + x (1 + r) = φCall d ⎩ ΔS 1,d + x (1 + r) = φ d
1 Call
⎧ ⎧
⎪
⎨ x (1 + r) = φCall − ΔS1
u 1,u
⎨ x (1 + r) = −8
Δ=1 Δ=1
⎪
⎩ x (1 + r) = φ d − ΔS 1,d ⎩
Call 1 x (1 + r) = −2
Since the last system has no solution, the Call option cannot be replicated only
by means of the underlying and of cash.
2. We display below the dynamics of the two stocks and the payoff of the option
written on them:
S11,u = 16; S12,u = 20 φu = 9

S01 = 8; S02 = 10 → S11,m = 12; S12,m = 8 φm = 1
.
S11,d = 2; S12,d = 6 φd = 0
− − − − − − −− − − − − − − − − −− − − − − − − − − −−
0 1 year Payoff the option
Verifying the existence of a replicating strategy for the option is equivalent to

verifying the existence of a number Δ1 of shares of one stock, of a number Δ2
of shares of the other stock and of an amount x of cash such that
⎧
⎪ 1,u 2,u
⎨ Δ1 S1 + Δ2 S1 + x (1 + r) = φ u
1,m 2,m
. Δ S1 + Δ S1 + x (1 + r) = φ m .
1 2
⎪
⎩ Δ1 S 1,d + Δ2 S 2,d + x (1 + r) = φ d
1 1
In the present case, the previous system (of 3 equations in 3 variables) is

equivalent to
⎧ ⎧
⎨ 16Δ1 + 20Δ2 + x (1 + r) = 9 ⎨ 16Δ1 + 20Δ2 + x (1 + r) = 9
. 12Δ1 + 8Δ2 + x (1 + r) = 1 10Δ1 + 2Δ2 = 1
⎩ ⎩
2Δ1 + 6Δ2 + x (1 + r) = 0 2Δ1 + 6Δ2 + x (1 + r) = 0
⎧ ⎧
⎨ 16Δ1 + 20Δ2 + x (1 + r) = 9 ⎨ 16Δ1 + 20Δ2 + x (1 + r) = 9
. Δ2 = 1/2 − 5Δ1 Δ2 = 1/2 − 5Δ1
⎩ ⎩
14Δ + 14Δ = 9
1 2 −56Δ1 = 2
⎧ ⎧
⎪
⎨ x = 9−16Δ
1 −20Δ2
⎨ x = −3.846
1+r
Δ = 28
2 19 Δ2 = 19
⎪
⎩ Δ1 = − ⎩ 1 28
1
28
Δ = − 281
It is then possible to replicate the option by selling −Δ1 = 28 1

shares of the
first stock, buying Δ = 28 shares of the second stock and borrowing 3.846
2 19
euros. So, the cost of the replicating trading strategy is equal to
1 19
Δ1 S01 + Δ2 S02 + x = −
. ·8+ · 10 − 3.846 = 2.654 euros.
28 28
Exercise 3.7 A stock (indexed by A) is available on the market at the current price
S0A = 8 euros. In 1 year, the price may increase by 25% or decrease by 25%.
Another stock (indexed by B) is also available. Its current price is S0B = 12 euros
that, in 1 year, may increase by 25% (when also the price of stock A is increased)
or decreased by 25% (when also the price of stock A is decreased). The risk-free
interest rate on the market is 4% per year.
1. Consider a European Call option with maturity of 1 year, with strike of 8 euros
and written on stock A. Verify if it is possible to replicate such an option only by
means of stock A and of cash invested or borrowed at risk-free rate.
2. Verify if it is possible to replicate the European Call option above by taking long
positions in stock A, short positions in stock B and cash. If yes, compute the cost
of the replicating strategy.
3. Verify if it is possible to replicate the European Call option above only by means
of stock A, of cash and of one (and only one) short position in stock B. If yes,
compute the cost of the replicating strategy.
Solution
1. The prices of stocks A and B may increase by a factor u = 1.25 (growth factor)
or decrease by a factor d = 0.75. These prices and the payoff of the European
Call option on stock A, with maturity of 1 year and strike of 8 euros, can be
summarized as follows:
S1A,u = S0A u = 10 u
φCall =2
S1B,u = S0B u = 15

S0A = 8
S0B = 12
.

S1A,d = S0A d = 6 d
φCall =0
S1B,d = S0B u = 9
− − − − − − −− −−−−−−−−−−− −−−−−−−−−
0 1 year Payoff of Call on A
In order to verify if the Call option above is attainable by means of stock A

and of cash, we need to check if the following system has solutions or not:

ΔA S1A,u + x (1 + r)T = φCall
u
. ,
ΔA S1A,d + x (1 + r)T = φCall
d
where ΔA represents the number of shares of stock A to buy or sell and x the
amount of cash to be invested or borrowed.
Substituting the price of A, the previous system (of 2 equations in 2 variables)

becomes

10ΔA + x (1 + r) = 2 10ΔA + x (1 + r) = 2
.
6Δ + x (1 + r) = 0
A 4ΔA = 2

x = − 1.04
3
x = −2.88
ΔA = 12 ΔA = 12
Consequently, the Call option can be replicated by buying ΔA = 12 shares of

stock A and by borrowing 2.88 euros at risk-free rate. The replicating strategy
then costs
.ΔA S0A + x = 4 − 2.88 = 1.12 euros.
2. Verifying the existence of a replicating strategy for the Call option above by
means of long positions in stock A, of short positions in stock B and of cash is
equivalent to verifying if there exist a number ΔA ≥ 0 of shares of stock A, a
number ΔB ≤ 0 of shares of stock B and an amount x of cash such that
⎧
⎪ A,u B,u
⎨ ΔA S1 + ΔB S1 + x (1 + r) = φ u
. ΔA S1A,d + ΔB S1B,d + x (1 + r) = φ d .
⎪
⎩ ΔA ≥ 0; ΔB ≤ 0
Substituting our data, the previous system becomes

⎧ ⎧
⎨ 10ΔA + 15ΔB + x (1 + r) = 2 ⎨ 10ΔA + 15ΔB + x (1 + r) = 2
. 6ΔA + 9ΔB + x (1 + r) = 0 4ΔA + 6ΔB = 2
⎩ A ⎩ A
Δ ≥ 0; ΔB ≤ 0 Δ ≥ 0; ΔB ≤ 0
⎧ ⎧
⎨ 10ΔA + 15ΔB + x (1 + r) = 2 ⎨ 5 − 15ΔB + 15ΔB + x (1 + r) = 2
ΔA = 12 − 32 ΔB ΔA = 12 − 32 ΔB
⎩ A ⎩ A
Δ ≥ 0; ΔB ≤ 0 Δ ≥ 0; ΔB ≤ 0
⎧
⎨ x = − 1.04 = −2.88
3
ΔA = 12 − 32 ΔB
⎩ B
Δ ≤0
Consequently, the Call option can be replicated with ΔB ≤ 0 stocks B, ΔA =

2 − 2 Δ stocks A and by borrowing 2.88 euros. The replication cost is then
1 3 B
equal to
1 3 B
ΔA S0A + ΔB S0B + x =
. − Δ · 8 + 12ΔB − 2.88 = 1.12 euros.
2 2
3. In the present case, we should verify if it is possible to replicate the Call option
only by means of stock A, of cash and by selling one (and only one) short position
in stock B. We just need to impose ΔB = −1 and apply the result obtained in the
previous item.
The required replicating strategy of the Call reduces to
⎧ ⎧
⎨ x = − 1.04
3
= −2.88 ⎨ x = − 1.04
3
= −2.88
. Δ = 2 − 32 (−1)
A 1
=2
ΔA
⎩ B ⎩ B
Δ = −1 Δ = −1
In the present case, then, the Call option can be replicated by buying 2 shares of
stock A, selling one share of stock B and borrowing 2.88 euros. It is immediate
to check that such a replicating strategy, as well, costs 1.12 euros.
Exercise 3.8 The current price of a stock is S0 = 8 euros. In 1 year, such price may
move up to 16 euros or to 12 euros, or move down to 8.32 euros. Each event may
happen with probability 1/3.
On the market, the risk-free interest rate is 4% per year and it is possible to sell
(at the price of one euro) a European Call option written on the stock above, with
maturity of 1 year and with strike of 11 euros.
Consider the following strategy:
• buy one share of the stock;
• sell the Call;
• borrow 7 euros at the risk-free rate.
Verify that such a strategy represents an arbitrage opportunity.
Solution Consider strategy (A):
• buy one share of the stock;
• sell the Call;
• borrow 7 euros at the risk-free rate.
The initial value V0 (A) and the final value V1 (A) of strategy A are the following:
t =0 t = 1 year
buy one stock stock price

⇒ −S0 = −8 ST
sell the Call Call payoff

.
⇒ +C0 = 1 − (ST − K)+
borrow 7 euros give back the loan

⇒ +c = 7 −c (1 + r)1 = −7 · 1.04 = −7.28
V0 (A) = C0 − S0 + c = 0 V1 (A) = ST − (ST − K)+ − c (1 + r)
Recall that ST − (ST − K)+ = min{ST ; K}.

Since V0 (A) = 0, in order to verify that the strategy above is an arbitrage
opportunity we just need to check that V1 (A) = ST − (ST − K)+ − c (1 + r) =
min{ST ; K} − c (1 + r) ≥ 0 with P (V1 (A) > 0) > 0.
Since P (ST = 16) = P (ST = 12) = P (ST = 8.32) = 13 , the final value
V1 (A) of the strategy A is given by
⎧
⎨ min{16; 11} − 7.28 = 3.72; with prob. 1/3
.V1 (A) = min{12; 11} − 7.28 = 3.72; with prob. 1/3
⎩
min{8.32; 11} − 7.28 = 1.04; with prob. 1/3
Therefore, V1 (A) ≥ 0 and P (V1 (A) > 0) = 1 > 0.

As V0 (A) = 0, and because of the previous arguments, strategy A is an arbitrage
opportunity.
Exercise 3.9 A stock A is available on the market at the current price of 12 euros.
In 4 months, such price may move up to 16 or down to 10 euros. In each of the next
4-month periods, the stock price may increase or decrease by the same percentage
as in the previous period. The risk-free interest rate on the market is 4% per year.
Consider also a stock B whose price is the square of the price of stock A at any
time and at any node of the tree.
1. Denote by St and S̃t the prices (at time t) of stocks A and B, and by u and d
(respectively, ũ and d̃) the increase factor and the decrease factor on each period.
Verify that the price of stock B follows a binomial model with S̃0 = S02 ,
ũ = u2 and d̃ = d 2 .
2. Compute the current price of a European Put option on stock B, with maturity of
8 months and strike of 144 euros.
3. Find the replicating strategy of the European Put option. Verify if the result found
in item 2. is in line with the replicating strategy cost.
4. Compare the above Put with a similar Put option on stock A with strike of 12
euros. Which one is more expensive?
5. Using the Put-Call Parity, deduce the prices of the corresponding European Call
options.
Exercise 3.10 The current price of a stock is S0 = 8 euros. In 1 year, such price (S1 )
may move up to 16 euros, to 12 euros or down to 8.32 euros, each with probability
1/3.
The risk-free interest rate available on the market is 4% per year.
1. Consider the following strategy: buy one share of the stock and borrow 8 euro at
the risk-free rate.
Verify if the strategy is an arbitrage opportunity or not.
2. Discuss what happens if the stock price (in 1 year) assumes the values 16, 12 and
8.32 euros with probability 1/2, 1/4 and 1/4, respectively.
3. Discuss whether the interest rate can be chosen so that the corresponding market
model is free of arbitrage.
4. Verify if the strategy in item 1. remains an arbitrage opportunity even if S̃1 =
S1 − 4 represents the value of another stock in 1 year and the interest rate is 4%
per year.
Chapter 4
Absence of Arbitrage and Completeness
of Market Models
In the following, we recall the notions of arbitrage, completeness and option pricing
in quite general one-period (or multi-period) market models, but always based on a
finite sample space.
For a detailed treatment and further details on the subject, we refer to Pliska [37],
among others.
Consider a market model consisting in a non-risky asset, called bond, whose
value will be denoted by B, and in n risky assets, called stocks, whose values will
be denoted by .S 1 , S 2 , . . . , S n .
Let us focus, first, on a one-period model. This means that any asset on the market
can be exchanged just at the beginning and at the end of the given interval of time.
By convention, let .t = 0 and .t = 1 be the corresponding dates.
The initial prices of the (non-risky and risky) assets are known: .B0 for the bond,
1 2 n
.S , S , . . . , S for the stocks. The bond price is deterministic, equal to
0 0 0
B1 = B0 (1 + r)
.
at .t = 1, where r is the risk-free interest rate on the period. On the other hand, the
prices of the stocks at .t = 1 will be random, namely
S1i = S1i (ω) , for i = 1, 2, . . . , n,

.
with .ω ∈ Ω = {ω1 , ω2 , . . . , ωm } and .m ∈ N (finite sample space). Any state of the

world will be assumed to be possible, that is .P (ωk ) > 0 for any .k = 1, 2, . . . , m
(under the real probability P on .Ω).

64 4 Absence of Arbitrage and Completeness of Market Models
It is then possible to form different trading portfolios by means of the risky and
non-risky assets above. A trading strategy or portfolio is characterized by a vector

α = α 0 , α 1 , . . . , α n ∈ Rn+1 ,
.
where .α 0 and .α i (.i = 1, 2, . . . , n) stand, respectively, for the number of non-risky

assets and the number of shares of stock .S i in the portfolio. In particular, negative
values of .α j (.j = 0, 1, 2, . . . , n) correspond to short positions, while positive values
correspond to long positions in asset j .
Given a strategy .α, its value at time .t ∈ {0, 1} is given by

n
Vt (α) = α Bt +
.
0
α i Sti . (4.1)
i=1
A strategy .α is said to be an arbitrage opportunity if .V0 (α) = 0 and .V1 (α) ≥ 0

with .P (V1 (α) > 0) > 0. In other words, an arbitrage opportunity is a way to make
a profit at zero cost (hence the name of free lunch).
It is therefore reasonable to exclude arbitrage opportunities or, better, to focus
on market models with no arbitrage opportunities. As already recalled in Chap. 3,
the First Fundamental Theorem of Asset Pricing guarantees that the one-period
market model above has no arbitrage opportunities if and only if there exists (at
least) one equivalent martingale measure (EMM, for short), where with equivalent
martingale measure (or risk-neutral measure) we mean a probability measure Q
such that .Q (ωk ) > 0 for any .k = 1, 2, . . . , m (hence, equivalent) and such that

S1i
EQ
. = S0i (4.2)
B1
for any .i = 1, 2, . . . , n (hence, martingale measure). The set of all equivalent

martingale measures will be denoted by .M .
There remains to establish how to evaluate a derivative. First of all, we will
distinguish between attainable derivatives and non-attainable derivatives. We recall
that a derivative with payoff .Φ is said to be attainable if there exists a trading
strategy .α such that .V1 (α) = Φ. If all the derivatives available on the market are
attainable (that is, they can be replicated), then the market is said to be complete;
otherwise, it is said to be incomplete.
The following result provides a necessary and sufficient condition for a one-
period model to be complete.
Theorem 4.1 Suppose that the market model considered is free of arbitrage. It is
also complete if and only if the matrix
⎡ ⎤
B1 (ω1 ) S11 (ω1 ) · · · S1n (ω1 )
⎢ B1 (ω2 ) S 1 (ω2 ) · · · S1n (ω2 ) ⎥
⎢ 1 ⎥
.⎢ .. .. .. .. ⎥
⎣ . . . . ⎦
B1 (ωm ) S11 (ωm ) · · · S1 (ωm )
n
has rank equal to the cardinality of .Ω, that is m.

The Second Fundamental Theorem of Asset Pricing guarantees that if a one-
period model as above is Free of Arbitrage, then it is complete if and only if there
exists a unique equivalent martingale measure (i.e. card.(M ) = 1).
If the market model is free of arbitrage, it is well known that:

• a derivative (with payoff .Φ) is attainable if and only if .EQ BΦ1 is constant for
any .Q ∈ M ;
• if a derivative (with payoff .Φ) is attainable, then its initial price coincides with
the cost of its replicating strategy:

Φ
EQ
. , with Q ∈ M ; (4.3)
B1
• if a derivative (with payoff .Φ) is not attainable, then its initial price belongs to
the no-arbitrage interval .(V− (Φ) , V+ (Φ)), where

Φ Φ
.V− (Φ) = inf EQ ; V+ (Φ) = sup EQ . (4.4)
Q∈M B1 Q∈M B1
If the infimum and supremum .V− (Φ) and .V+ (Φ) are attained, then the no-
arbitrage interval is closed.
Consider now a multi-period model where transactions may occur at times
0, 1, 2, . . . , T
.
and consisting in a non-risky asset (called bond), whose value will be denoted by B,
and in n risky assets (called stocks), whose values will be denoted by .S 1 , S 2 , . . . , S n .
Assume (as in the one-period model) that the sample space is finite
Ω = {ω1 , ω2 , . . . , ωm }
.
and that any elementary event is possible with respect to a probability P given a
priori, that is .P (ωk ) > 0 for any .k = 1, . . . , m. The information available in time
is represented by a filtration .(Ft )t=0,1,...,T with .F0 = {∅, Ω} and .FT = P (Ω),
where .P (Ω) denotes the power set of .Ω.
The bond price at .t ∈ {1, 2, . . . , T } is assumed to be equal to
Bt = B0 (1 + r)t ,
. (4.5)
i .B0 = 1 and r stands as usual for the risk-free interest rate. The stocks prices
where
.St t=0,1,...,T (for .i = 1, 2, . . . , n) are assumed to be stochastic processes, adapted
to .(Ft )t=0,1,...,T and such that .S0i > 0 for any .i = 1, 2, . . . , n.

In the multi-period model, a trading strategy is a vector .α = αt0 , αt1 , . . . , αtn
t=1,...,T of processes, where .αt and .αt (.i = 1, 2, . . . , n) represent, respectively, the
0 i
number of bonds and stock .S in the portfolio from time .(t − 1) to t.

i
The value of a trading strategy .α is then given by

n
V0 (α) = α10 B0 +
. α1i S0i
i=1

n
Vt (α) = αt0 Bt + αti Sti , for t = 1, . . . , T .
i=1
A strategy .α is said to be self-financing if .Vt (α) = αt+10 B + n α i S i for

t i=1 t+1 t
any .t = 1, . . . , T . In other words, a strategy is self-financing when at any time t
what one obtains by selling the strategy .αt is enough (and not too much) to finance
the purchase of the strategy .αt+1 ; hence nothing has to be added and nothing can be
withdrawn.
Furthermore, a strategy .α is called an arbitrage opportunity if it is self-financing
and if it satisfies .V0 (α) = 0 and .VT (α) ≥ 0, with .P (VT (α) > 0) > 0.
Similarly to the one-period case, the First Fundamental Theorem of Asset
Pricing guarantees that the multi-period model described above is free of arbitrage
opportunities if and only if there exists (at least) one equivalent martingale measure,
that is, a probability measure Q such that .Q (ωk ) > 0 for any .k = 1, 2, . . . , m
(equivalent) and such that

Sti Ssi
EQ F = , (4.6)
Bt
. s
Bs
for any .s ≤ t and .i = 1, 2, . . . , n (a martingale measure). The set of all equivalent

martingale measures will be denoted by .M .
It is also well known that if the multi-period model is free of arbitrage, then all
one-period sub-markets are free of arbitrage.
Furthermore, a derivative with payoff .Φ is said to be attainable if there exists
a self-financing trading strategy .α such that .VT (α) = Φ. If all the derivatives
available on the market are attainable (that is, they can be replicated), then the
market is said to be complete; otherwise, it is called incomplete.
The Second Fundamental Theorem of Asset Pricing guarantees that if a multi-
period model as above is free of arbitrage, then it is complete if and only if there
exists a unique equivalent martingale measure (i.e. card.(M ) = 1).
Moreover, if the multi-period model is free of arbitrage, then it is complete if and
only if all its one-period sub-markets are complete.
As in the one-period case, if the market model is free of arbitrage it is well known
that:

• a derivative (with payoff .Φ) is attainable if and only if .EQ BΦT is constant for
any .Q ∈ M ;
• if a derivative (with payoff .Φ) is attainable, then its price at .t ∈ {0, 1, . . . , T }
coincides with the cost of its replicating strategy, which equals

Φ
Vt (Φ) = Bt · EQ Ft , with Q ∈ M . (4.7)
BT
.
Exercise 4.2 Consider a one-period (annual) market model consisting in a non-

risky asset (paying a risk-free rate of 5% per year) and in two stocks with prices S 1
and S 2 :
⎧
⎨ 12; if ω = ω1
S01 = 10; S11 (ω) = 8; if ω = ω2
⎩
6; if ω = ω3
.
⎧
⎨ 10; if ω = ω1
S02 = 5; S12 (ω) = 4; if ω = ω2 ,
⎩
5; if ω = ω3
where P (ω1 ) = P (ω2 ) = P (ω3 ) = 13 .

Establish if there exist arbitrage opportunities on the market. If yes, give an
example.
Solution In the present example a one-period market model is considered, where
time is measured in years. The bond price evolves as follows:
B0 = 1
.
B1 = 1 + r
with r = 0.05 per year, while the prices of the stocks can be summarized as

S11,u ; S12,u = (12; 10) (on ω1 )

1 2
S0 ; S0 = (10; 5) −→ S11,m ; S12,m = (8; 4) (on ω2 )
.

S11,d ; S12,d = (6; 5) (on ω3 )
− − − − − − −− − − − − − − − − − − −−
t =0 T = 1 year
By the First Fundamental Theorem of Asset Pricing, a market is free of arbitrage

if and only if there exists an equivalent martingale measure. So, we only need to
establish if there exists a probability measure Q satisfying qk = Q (ωk ) > 0 for any
k = 1, 2, 3 and
⎧
⎪
⎪ 1 =E S11
⎨ 0
S Q B1
. .
⎪
⎪ 2 =E S12
⎩ 0
S Q B1
The previous system can be rewritten as

⎧
⎪
⎨ 1+r q1 + 1+r q2 + 1+r q3 = 10
12 8 6
1+r q1 + 1+r q2 + 1+r q3 = 5

.
10 4 5
⎪
⎩q + q + q = 1
1 2 3
⎧
⎨ 12q1 + 8q2 + 6q3 = 2
21
10q1 + 4q2 + 5q3 = 21 ,

⎩ 4
q1 + q2 + q3 = 1
so its unique solution is

⎧
⎨ q1 = 16
5
. q = 21
16 > 1
⎩ 2
q3 = − 58 < 0
Accordingly, there do not exist any equivalent martingale measures. Hence, the
market above is not free of arbitrage.

An example of arbitrage opportunity is given by the strategy α = α 0 , α 1 , α 2 =
(0; −1; 2), consisting in zero positions in the bond, a short position in stock S 1 and
two long positions in stock S 2 . For such a strategy, indeed,
V0 (α) = 0 · B0 − 1 · S01 + 2 · S02 = −10 + 10 = 0,

.
while
V1 (α) = −1 · S11 + 2 · S12

⎧
.
⎨ 8; if ω = ω1
= 0; if ω = ω2 .
⎩
4; if ω = ω3
Consequently, V0 (α) = 0 and V1 (α) ≥ 0 with P (V1 (α) > 0) = 23 .

Exercise 4.3 Consider a one-period (annual) model formed by a bond (paying a
risk-free rate of 5% per year) and by two stocks with prices S 1 and S 2 evolving as
follows:
⎧
⎨ 12; if ω = ω1
S0 = 10; S1 (ω) = 10; if ω = ω2
1 1
⎩
6; if ω = ω3
.

15; if ω = ω1
S02 = 10; S12 (ω) = ,
8; if ω ∈ {ω2 , ω3 }
with P (ω1 ) , P (ω2 ) , P (ω3 ) > 0 and P (ω1 ) + P (ω2 ) + P (ω3 ) = 1.

1. Establish if the market is free of arbitrage and complete.
2. Consider two derivatives (A and B) with maturity T of 1 year and with payoffs
+
ST1 +ST2 +
Φ =
A
2 −8 and Φ B = ST1 − ST2 , respectively. Find their initial
prices and their replicating strategies.
3. Discuss whether the market formed only by the bond and by stock S 1 would
remain free of arbitrage and complete.
4. In the market formed by the bond and by the stock S 1 , find the no-arbitrage
intervals for the prices of derivatives A and B.
Solution
1. The market model we are dealing with is a one-period model with initial time 0
and final time 1 year. The bond price evolves as
.B0 = 1
B1 = 1 + r
with r = 0.05 per year, while the prices of the stocks can be summarized as
follows:

S11,u ; S12,u = (12; 15) (on ω1 )

1 2
S0 ; S0 = (10; 10) −→ S11,m ; S12,m = (10; 8) (on ω2 )
.

S11,d ; S12,d = ( 6; 8) (on ω3 )
− − − − − − − − −− − − − − − − − − − − − − −−
t =0 T = 1 year
In order to establish if the market above is free of arbitrage (or not), we need
to verify if there exists (at least) one equivalent martingale measure, that is a
probability measure Q satisfying qk = Q (ωk ) > 0 for any k = 1, 2, 3 and
⎧
⎪ 1
⎪ S1
⎨ S0 = EQ B11
. .
⎪
⎪ S2
⎩ S02 = EQ B11
The previous system is equivalent to

⎧
⎪
⎨ 1+r q1 + 1+r q2 + 1+r q3 = 10
12 10 6
1+r q1 + 1+r q2 + 1+r q3 = 10

.
15 8 8
⎪
⎩q + q + q = 1
1 2 3
⎧
⎨ 12q1 + 10q2 + 6q3 = 2
21
15q1 + 8q2 + 8q3 = 2 21 .

⎩
q1 + q2 + q3 = 1
The systems above have a unique (q1 , q2 , q3 ) satisfying q1 , q2 , q3 ∈

solution
(0, 1), namely (q1 , q2 , q3 ) = 14 , 56 , 56 . Hence, the market is free of arbitrage
5 33 3
and, by the Second Fundamental Theorem of Asset Pricing, also complete. The
unique equivalent martingale measure Q is given by Q (ω1 ) = 14 5
, Q (ω2 ) = 3356
and Q (ω3 ) = 563
.
An alternative way to deduce the completeness of the market consists in

observing
⎡ ⎤ ⎡ ⎤
B1 (ω1 ) S11 (ω1 ) S12 (ω1 ) 1.05 12 15
.rk ⎣B1 (ω2 ) S (ω2 ) S (ω2 )⎦ = rk ⎣1.05 10 8 ⎦ = 3,
1 2
1 1
B1 (ω3 ) S11 (ω3 ) S12 (ω3 ) 1.05 6 8
i.e. the rank is equal to the cardinality of the sample space.

2. Since the market formed by the bond and stocks S 1 and S 2 is free of arbitrage
and complete, derivatives A and B (with maturity of 1 year and with payoffs Φ A
and Φ B ) are attainable. Their initial prices are given, respectively, by
1 1
V0 Φ A =
. EQ Φ A ; V0 Φ B = EQ Φ B . (4.8)
1+r 1+r
It is immediate to check that

⎧ ⎧
⎨ 5.5; if ω = ω1 ⎨ 0; if ω = ω1
.Φ (ω) = 1; if ω = ω2 ; Φ B (ω) = 2; if ω = ω2 ,
A
⎩ ⎩
0; if ω = ω3 0; if ω = ω3
hence, by (4.8), we deduce that

1 1
V0 Φ A =
. EQ Φ A = [5.5 · Q (ω1 ) + 1 · Q (ω2 ) + 0] = 2.432
1+r 1.05
1 1
V0 Φ B = EQ Φ B = [0 + 2 · Q (ω2 ) + 0] = 1.1224.
1+r 1.05
There remains to find the replicating strategies of the two derivatives above.
A replicating
0 1 trading strategy for the derivative A with payoff Φ A is a vector
αA = αA , αA , αA satisfying V1 (α) = Φ A ; equivalently, a vector solving the
2
following system of equations

⎧
⎪ 1,u 2,u
⎨ αA · B1 + αA · S1 + αA · S1 = 5.5
0 1 2
. αA · B1 + αA · S1 + αA · S12,m = 1
0 1 1,m 2
⎪
⎩ α 0 · B + α 1 · S 1,d + α 2 · S 2,d = 0
A 1 A 1 A 1
⎧ 0
⎨ αA · 1.05 + αA1 · 12 + α 2 · 15 = 5.5
A
αA · 1.05 + αA
0 1 · 10 + α 2 · 8 = 1
⎩ 0 A
αA · 1.05 + αA1 · 6 + α2 · 8 = 0
A
The unique solution of the systems is αA = (−5.7823, 0.25, 0.5714), giving the
required replicating strategy of A. Its cost V0 (αA ) = −5.7823 + 0.25 · 10 +
0.5714 · 10 = 2.432 coincides with the initial price of the derivative A (as we
should expect, and indeed happens).
0Similarly, one obtains that the replicating strategy of option B is αB =
αB , αB1 , αB2 = (0.40816, 0.5, −0.42857). Consequently, its cost V0 (αB ) =
0.40816 + 5 − 4.2857 = 1.1225 coincides with the initial price of derivative
B (again, as it should be).
3. Proceeding as above, it is easy to check that the one-period model formed only
by the bond and by stock S 1 is free of arbitrage but incomplete. There exist,
indeed, infinitely many equivalent martingale measures, i.e. probability
measures

S11
Q satisfying qk = Q (ωk ) > 0 (for any k = 1, 2, 3) and S01 = EQ B1 . In the
present case, the set of equivalent martingale measures is given by

1 3 9 3 1 1

.M = Q Q (ω1 ) = q ∈ ; ; Q (ω2 ) = − q; Q (ω3 ) = q − .
4 4 8 2 2 8
4. According to the previous item, the no-arbitrage interval for the price of the
option with payoff Φ A is given by

1 1
. inf EQ Φ ;
A
sup EQ Φ A
.
1 + r Q∈M 1 + r Q∈M

Since EQ Φ A = 5.5 · Q (ω1 ) + 1 · Q (ω2 ) + 0 = 5.5q + 9
8 − 1.5q = 4q + 9
8
for Q ∈ M , we get

1 1 1 9 9
. inf EQ Φ A = inf 4q +
1+ = = 2.0238
1 + r Q∈M 1.05 1 3
4 <q< 4
1.05 8 8

1 1 9 1 9
sup EQ Φ =
A
sup 4q + = 3+ = 3.9286.
1 + r Q∈M 1.05 1 <q< 3 8 1.05 8
4 4
It follows that the no-arbitrage interval for the price of option A is

(2.0238; 3.9286).
Similarly, one can obtain that the no-arbitrage price of option B belongs to the
interval (0; 1.4286).
Exercise 4.4 Consider a bond (paying a risk-free rate of 3% per year) and two
stocks with prices S 1 and S 2 evolving as follows:
S21,C = 14; S22,C = 8

S21,D = 12; S22,D = 4
S11,A = 12; S12,A = 6 → S21,E = 9; S22,E = 2
S21,F = 7; S22,F = 9
S01 = 10; S02 = 5
. S21,G = 8; S22,G = 4
S11,B = 6; S12,B =3

S21,H = 4; S22,H = 5
− − − − − − − − −− − − − − − − − − − − −− − − − − − − − − − − −−
0 1 year 2 years
Establish if the two-period market model formed by the bond and two stocks
above is free of arbitrage or not.
Solution Denote by ω1 the state of the world corresponding to node C where at
T = 2 the price of the first stock has reached 14 and the price of the second stock
has reached 8, by ω2 the state of the world corresponding to node D, and so on.
By the two-period model described above, we deduce that
.Ω = {ω1 , ω2 , . . . , ω6 }
and the filtration corresponding to the information available in time is the following
one:
F0 = {∅, Ω}
.
F1 = {∅, {ω1 , ω2 , ω3 , ω4 }, {ω5 , ω6 }, Ω}

F2 = P (Ω) .
We shall write U = {ω1 , ω2 , ω3 , ω4 } and D = {ω5 , ω6 }.

Recall that if a multi-period model is free of arbitrage then all its one-period
sub-markets are free of arbitrage.
Let us consider the first one-period sub-market, written in bold:
S21,C = 14; S22,C = 8

S21,D = 12; S22,D = 4
S1,A
1 = 12; S2,A
1 =6 → S21,E = 9; S22,E = 2
S21,F = 7; S22,F = 9
S10 = 10; S20 = 5
. S21,G = 8; S22,G = 4
S1,B
1 = 6; S2,B
1 = 3

S21,H = 4; S22,H = 5
− − − − − − − − −− − − − − − − − − − − −− − − − − − − − − − − −−
0 1 year 2 years
The sub-market above is free of arbitrage if and only if there exists at

least an equivalent martingale measure, i.e. a probability measure Q such that
Q (U ) , Q (D) ∈ (0, 1) and
⎧
⎪
⎪ S1
⎨ EQ B11 = S01
. .
⎪
⎪ S2
⎩ EQ B11 = S02
Calling q = Q (U ) and observing that Q (D) = 1 − Q (U ), the previous condition

becomes
⎧ 1,A
⎨ S1 · Q (U ) + S11,B
· Q (D) = S01
. 1+r 1+r
⎩ S12,A · Q (U ) + S12,B
· Q (D) = S02
1+r 1+r
!
12
1+r ·q + 6
1+r · (1 − q) = 10
.
6
1+r ·q + 3
1+r · (1 − q) = 5
Trivially, the systems above reduce to 12q + 6 (1 − q) = 10 (1 + r). Hence
2 5 ∼
.q = + r = 0.72
3 3
and
.Q (U ) = q = 0.72; Q (D) = 1 − q = 0.28.
It follows that the first sub-market is free of arbitrage and, because of the uniqueness
of the equivalent martingale measure, complete.
We consider now the one-period sub-market in bold:
S1,C
2 = 14; S2,C
2 =8
S1,D
2 = 12; S2,D
2 =4
S1,A
1 = 12; S2,A
1 =6 → S1,E
2 = 9; S2,E
2 =2
1,F
S2 = 7; S2,F
2 =9
S01 = 10; S02 = 5
. S21,G = 8; S22,G = 4
S11,B = 6; S12,B =3

S21,H = 4; S22,H = 5
− − − − − − − − −− − − − − − − − − − − −− − − − − − − − − − − −−
0 1 year 2 years
As previously, the sub-market is free of arbitrage if and only if there exists at least
one probability measure Q ( ·| U ) satisfying qk = Q ( ωk | U ) > 0 (for k = 1, 2, 3, 4)
and
⎧
⎪
⎪ S1 S11,A
⎨ EQ B22 U = B1
.
S2
.
⎪
⎪ S12,A
⎩ EQ B22 U = B1
The condition above becomes

⎧
⎪
⎨ (1+r)2 · q1 + (1+r)2 · q2 + (1+r)2 · q3 + · q4 =
14 12 9 7 12
⎪ (1+r)2 1+r
.
8
· q1 + 4 2 · q2 + 2 2 · q3 + 9
· q4 = 6
⎪
⎪ (1+r)2 (1+r) (1+r) (1+r)2 1+r
⎩
q1 + q2 + q3 + q4 = 1
⎧
⎪
⎨ 7q1 + 5q2 + 2q3 = 5 + 12r
−q1 − 5q2 − 7q3 = −3 + 6r
⎪
⎩
q 4 = 1 − q1 − q2 − q3
Consequently,
⎧
⎪
⎪ q1 = 56 q3 + 13 + 3r
⎪
⎨
q2 = 8−27r
15 − 30 q3
47
. .
⎪
⎪ q3 ∈ (0, 1)
⎪
⎩
q4 = 15
2
− 65 r − 15
4
q3
By imposing qk > 0, we obtain

⎧
⎪ q1 = 6 q3 + 3 + 3r
5 1
⎪
⎪
⎨
q2 = 15 − 47
8−27r
30 q3
. .
⎪
⎪ 0 < q < min{ 4 ; 5 ; 47 } = 0.306
2−18r 4−18r 16−54r
⎪
⎩
3
q4 = 152
− 65 r − 154
q3
It follows that also the second sub-market is free of arbitrage (but incomplete).
Furthermore,
.Q (ωk ) = Q ( ωk | U ) · Q (U ) , for k = 1, 2, 3, 4.
Finally, we consider the last one-period sub-market:
S21,C = 14; S22,C = 8

S21,D = 12; S22,D = 4
S11,A = 12; S12,A = 6 → S21,E = 9; S22,E = 2
S21,F = 7; S22,F = 9
S01 = 10; S02 =5
. S1,G
2 = 8; S2,G
2 =4
S1,B
1 = 6; S2,B
1 = 3

S1,H
2 = 4; S2,H
2 =5
− − − − − − − − −− − − − − − − − − − − −− − − − − − − − − − − −−
0 1 year 2 years
As previously, such a sub-market is free of arbitrage if and only if there exists at

least one probability measure Q ( ·| D) satisfying qk = Q ( ωk | D) > 0 (for k = 5, 6)
and
⎧
⎪
⎪ S1 S11,B
⎨ Q B22 D =
E B1
.
S2
.
⎪
⎪ S12,B
⎩ EQ B22 D = B1
The condition above can be rewritten as

⎧
⎪
⎨ (1+r)2 · q5 + · q6 =
8 4 6
⎪ (1+r)2 1+r
.
4
· q5 + 5
· q6 = 3
⎪
⎪ (1+r)2 (1+r)2 1+r
⎩
q5 + q6 = 1
⎧
⎪
⎨ 8q5 + 4q6 = 6 + 6r
4q5 + 5q6 = 3 + 3r .
⎪
⎩
q5 + q6 = 1
It is easy to check that the system above is incompatible, hence there does not
exist any equivalent martingale measure for the sub-market taken into account. This
implies that such sub-market is not free of arbitrage and, consequently, the same
holds for the overall two-period market model.
Exercise 4.5 Consider a two-period model (with annual periods) formed by a non-
risky asset (B, paying a risk-free rate of 2% per year) and by a stock (S) whose price
evolves as follows:
S2C = 18
S1A = 15 → S2D = 15
S2E = 10
S0 = 10
S2F = 10
.
S1B =5

S2G = 4
− − − − −− −−−−−−− −−−−−−−−−
0 1 year T = 2 years
Denote by ω1 (respectively, ω2 , . . . , ω5 ) the state of the world corresponding

to the node C (respectively, D, E, F, G). The probability measure P given a priori
satisfies P (ωi ) = 15 for any i = 1, 2, . . . , 5.
1. Check whether the two-period model above is free of arbitrage and complete.
2. Consider a European Call option with maturity of 2 years and with strike of 10,
written on stock S. Establish if such option is attainable and compute its price of
no arbitrage.
3. Find the no-arbitrage interval for the price of a derivative with payoff
Φ C = max{(ST − 10)+ ; CT },
.
where CT (ω1 ) = CT (ω3 ) = 4 and CT (ω2 ) = CT (ω4 ) = CT (ω5 ) = 0.

4. Compute the price of the derivative with payoff Φ C by means of the martingale
measure Q∗ ∈ M1 = { Q ∈ M | 0.1 ≤ Q (ω1 ) ≤ 0.2} minimizing the relative
entropy between all martingale measures in M1 and the probability measure P .
Solution By the two-period model specified above, we deduce that
. Ω = {ω1 , ω2 , . . . , ω5 }
and the filtration corresponding to the information available is
F0 = {∅, Ω}
.
F1 = {∅, {ω1 , ω2 , ω3 }, {ω4 , ω5 }, Ω}

F2 = P (Ω) .
Denote by U = {ω1 , ω2 , ω3 } and by D = {ω4 , ω5 }.

1. Recall that if a multi-period model is free of arbitrage then all its one-period
sub-markets are free of arbitrage. Furthermore, it is complete if and only if all its
one-period sub-markets are complete.
Consider now the first one-period sub-market model in bold:
S2C = 18
1 = 15
SA → S2D = 15
S2E = 10
S0 = 10
S2F = 10
.
1=5
SB

S2G = 4
−−−−− −−−−−−− − − − − − − −−
Such a sub-market is free of arbitrage if and only if there exists at least one
probability measure Q satisfying Q (U ) , Q (D) ∈ (0, 1) and

S1
EQ
. = S0 .
B1
Denoting by q = Q (U ) and observing that Q (D) = 1 − Q (U ), the condition

above becomes
15 5
. q+ (1 − q) = 10.
1+r 1+r
Hence
1
.Q (U ) = q = + r = 0.52
2
1
Q (D) = 1 − q = − r = 0.48.
2
By the arguments above, we conclude that the first one-period sub-market is
free of arbitrage and complete (because of the existence and uniqueness of the
martingale measure).
We focus now on the one-period sub-market in bold:
2 = 18
SC
1 = 15
SA → S2 = 15
D
2 = 10
SE
S0 = 10
S2F = 10
.
S1B = 5

S2G = 4
−−−−− −−−−−−− −−−−−−−−−
As previously, such a sub-market is free of arbitrage if and only if there exists

at least one probability measure Q ( ·| U ) (conditional probability, given the event
U ) satisfying qk = Q ( ωk | U ) > 0 (for k = 1, 2, 3), q1 + q2 + q3 = 1 and

S2 S1A
EQ U = .
B2
.
B1
The condition above becomes

!
2 · q1 + · q2 + · q3 =
18 15 10 15
. (1+r) (1+r)2 (1+r)2 1+r
q1 + q2 + q3 = 1

8q1 + 5q2 = 5 + 15r
q3 = 1 − q1 − q2
Hence
⎧
⎨ q1 ∈ (0, 1)
. q = 1 + 3r − 85 q1 .
⎩ 2
q3 = 35 q1 − 3r
Nevertheless, we have to restrict our attention only to those solutions that

correspond to equivalent martingale measures. We need, therefore, to impose
qk > 0 for any k = 1, 2, 3, obtaining that
⎧
⎨ 0.1 = 5r < q1 <
5
8 (1 + 3r) ∼
= 0.66
. q = 1.06 − 85 q1 .
⎩ 2
q3 = 35 q1 − 0.06
By the arguments above, the sub-market taken into account is free of arbitrage
but incomplete.
Furthermore, Q (ωk ) = Q ( ωk | U ) · Q (U ) for k = 1, 2, 3, so

1
.Q (ω1 ) = + r q1
2

1 8
Q (ω2 ) = +r 1.06 − q1
2 5

1 3
Q (ω3 ) = +r q1 − 0.06 .
2 5
We finally focus on the last one-period sub-market in bold:
S2C = 18
S1A = 15 → S2D = 15
S2E = 10
S0 = 10
SF2 = 10
.
1=5
SB

2=4
SG
−−−−− −−−−−−− −−−−−−−−−
As previously, such a sub-market is free of arbitrage if and only if there exists

at least one probability measure Q ( ·| D) (conditional probability, given the event
D) satisfying qk = Q ( ωk | D) > 0 (for k = 4, 5), q4 + q5 = 1 and

S2 SB
EQ
.
D = 1 .
B2 B1
The previous condition becomes

!
10
q + 4
q = 5
. (1+r)2 4 (1+r)2 5 1+r
,
q4 + q5 = 1
that admits as unique solution

q4 = 1
6 (1 + 5r) ∈ (0, 1)
. .
q5 = 5
6 (1 − r) ∈ (0, 1)
Accordingly, such a sub-market is free of arbitrage and complete. Furthermore,

Q (ωk ) = Q ( ωk | D) · Q (D) for k = 4, 5, so

1 1
. Q (ω4 ) = − r (1 + 5r)
2 6

5 1
Q (ω5 ) = − r (1 − r) .
6 2
To conclude, the two-period model taken into account is free of arbitrage and
incomplete. The set of all equivalent martingale measures for the two-period
model is
⎧ ⎫
⎪ ⎪
⎪ Q (ω1 ) = 12 + r q1
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ Q (ω2 ) = 1
+ r 1 + 3r − 8
q1 ⎪
⎪
⎨ 2 5 ⎬
5
.M = Q Q (ω3 ) = 2 + r 5 q1 − 3r ; for 5r < q1 < (1 + 3r) .
1 3
⎪
⎪ 8 ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ Q (ω4 ) = 1 1
− r (1 + 5r) ⎪
⎪
⎪
⎪
6 2 ⎪
⎪
⎪
⎩ Q (ω5 ) = 5 1 − r (1 − r) ⎪
⎭
6 2
(4.9)
2. Due to the incompleteness of the above two-period model, there is no guarantee

that the European Call option with maturity of 2 years, with strike of 10 and
written on S is attainable. The payoff of the Call option is given by
⎧
⎨ 8; ω = ω1
+
.Φ (ω) = (ST (ω) − 10) = 5; ω = ω2 .
⎩
0; ω ∈ {ω3 , ω4 , ω5 }
Looking for a self-financing replicating strategy for the option with payoff
Φ is the
same as looking for a self-financing replicating strategy (αt )t=1,2 with
αt = αtB , αtS (where α B , resp. α S , represents the number of bonds, resp. of
stocks, to be held in portfolio).
At maturity (T = 2), such a strategy (if it exists) has to satisfy V2 (α) = Φ,

i.e.
⎧ B,u
⎪
⎪ α2 · B2 + α2S,u · 18 = 8
⎪
⎪
⎪ B,u S,u
⎨ α2 · B2 + α2 · 15 = 5
. α2 · B2 + α2S,u · 10 = 0
B,u
⎪
⎪
⎪
⎪ α B,d · B2 + α2S,d · 10 = 0
⎪
⎩ 2B,d
α2 · B2 + α2S,d · 4 = 0
where α2·,u and α2·,d denote, respectively, the number of shares to be held in
portfolio from time 1 to time 2 in the node “up” or “down” at t = 1. The unique
α2 satisfying the system above is given by
⎧ B,u
⎪
⎪ α2 = − 10 2
⎪
⎨ S,u (1+r)
.
α2 = 1 .
⎪
⎪ α2B,d = 0
⎪
⎩ S,d
α2 = 0
It follows that V1u (α) = α2B,u · B1 + α2S,u · S1A = 5+15r

1+r and V1d (α) =
α2B,d · B1 + α2S,d · S1B = 0.
At time t = 1, a replicating self-financing strategy has to satisfy

α1B · B1 + α1S · S1A = V1u (α)
. ,
α1B · B1 + α1S · S1B = V1d (α)
hence
!
α1B (1 + r) + α1S · 15 = 5+15r
1+r
.
α1B (1 + r) + α1S · 5 = 0
!
α1B = − 5(1+3r)2
2(1+r) .
α1S = 1+3r
2(1+r)
The European Call option of concern is then attainable, so the price of no

arbitrage has to coincide with the cost of the self-financing replicating strategy,
that is
5 (1 + 3r) 1 + 3r
.V0 = α1B · B0 + α1S · S0 = − + · 10 = 2.6490. (4.10)
2 (1 + r) 2 2 (1 + r)
An alternative way to compute the option price is to find the no-arbitrage

interval and observe that it reduces to a point, namely the value V0 in (4.10).
3. The payoff of the claim C is equal to

⎧
⎪
⎪ 8; ω = ω1
⎨
+ 5; ω = ω2
.Φ (ω) = max{(ST (ω) − 10) ; CT (ω)} =
C
.
⎪
⎪ 4; ω = ω3
⎩
0; ω ∈ {ω4 , ω5 }
It is easy to check that such a claim is not attainable in the market we are dealing
with. We will then look for the no-arbitrage interval for its price.
Let Q be an equivalent martingale measure, i.e. Q ∈ M . From (4.9) it follows
that
C

EQ ΦB2
= 1
[8 · Q (ω1 ) + 5 · Q (ω2 ) + 4 · Q (ω3 )]
.
(1+r)2
= 1
8 1
+ r q1 + 5 1
+ r 1 + 3r − 8
q1 + 4 1
+r 3
5 q1 − 3r
(1+r)2 2 2 5 2
= 1+2r
5 + 3r + 5 q1 .
12
2(1+r)2
Hence

ΦC 1 + 2r 12
. inf EQ = inf 5 + 3r + q1
Q∈M B2 5r<q1 < 58 (1+3r) 2 (1 + r)2 5
5 (1 + 2r) (1 + 3r)
= = 2.65
2 (1 + r)2
and

ΦC 1 + 2r 12
. sup EQ = sup 5 + 3r + q1
Q∈M B2 5r<q1 < 58 (1+3r) 2 (1 + r)2 5
(13 + 15r) (1 + 2r)

= = 3.32.
4 (1 + r)2
The price of option C belongs then to the no-arbitrage interval (2.65; 3.32).
4. As verified in the previous items, the market considered is free of arbitrage and
incomplete (there exist indeed infinitely many equivalent martingale measures)
and the derivative with payoff Φ C is not attainable. A priori, therefore, we are
only able to find the no-arbitrage interval for the price of such a claim.
A criterion to “choose” one among the infinitely many equivalent martingale
measures (using which we can evaluate the derivative C) is the Minimal
Relative Entropy Criterion. More precisely, such a criterion suggests to select
the equivalent martingale measure Q that minimizes the relative entropy of Q

with respect to P :

dQ dQ
H (Q, P ) EP
. ln .
dP dP
In the present case, H (Q, P ) (with Q ∈ M1 ) becomes

H (Q, P ) = EP dQ ln dQ
5
dP
dP
Q(ωi ) Q(ωi )
= i=1 P (ωi ) ln P (ωi ) P (ωi )

= 5i=1 Q (ωi ) [ln (Q (ωi )) − ln (P (ωi ))]

= 5i=1 Q
(ωi ) ln (Q (ωi ))+ ln 5
.
= 2 + r · q1 ln (q1 ) + 1 + 3r − 85 q1 ln 1 + 3r − 85 q1
1

+ 35 q1 − 3r ln 35 q1 − 3r

+ 12 − r 16 (1 + 5r) ln 16 (1 + 5r) + 56 (1 − r) ln 56 (1 − r)

+ 12 + r ln 12 + r + 12 − r ln 12 − r + ln 5.
We deduce that
. arg min H (Q, P )

Q∈M1
⎡ ⎤
q1 ln (q1 ) + 1 + 3r − 85 q1 ln 1 + 3r − 85 q1
= arg min ⎣ 3 ⎦. (4.11)
Q∈M1 + q − 3r ln 3
q − 3r
5 1 5 1
Since Q ∈ M1 if and only if Q ∈ M and 0.1 ≤ Q (ω1 ) ≤ 0.2 (or,

equivalently, 0.1 ≤ 12 + r q1 ≤ 0.2), we deduce that Q ∈ M1 if and only
if
! !
0.1 ≤ 12 + r q1 ≤ 0.2 5+10r ≤ q1 ≤ 5+10r
1 2
.
5r < q1 < 58 (1 + 3r) 5r < q1 < 58 (1 + 3r)

0.199 ≤ q1 ≤ 0.398
,
0.1 < q1 < 0.6625
hence q1 ∈ [0.199; 0.398]. It follows that (4.11) can be rewritten as
. arg min H (Q, P ) = arg min g (q1 ) ,

Q∈M1 q1 ∈[0.199;0.398]

where g (q1 ) = q1 ln (q1 ) + 1 + 3r − 85 q1 ln 1 + 3r − 85 q1 + 35 q1 − 3r ·

· ln 35 q1 − 3r . Since g (q1 ) = ln (q1 )− 85 ln 1 + 3r − 85 q1 + 35 ln 35 q1 − 3r
≤ 0 on the interval [0.199; 0.398], minQ∈M1 H (Q, P ) is attained at q1∗ =
5+10r = 0.398. Accordingly, the equivalent martingale measure chosen by the
2
criterion is given by

Q∗ (ω1 ) = 1
+ r q1∗ = 0.2
2
8 ∗
Q∗ (ω2 ) = 1
+ r 1 + 3r − q = 0.22
2 5 1
3 ∗
. Q∗ (ω3 ) = 2 + r 5 q1 − 3r = 0.09
1
.
Q∗ (ω4 ) = 6 2 − r (1 + 5r) = 0.09
1 1
Q∗ (ω5 ) = 6 2 − r (1 − r) = 0.4
5 1
It follows that the price of the derivative C (evaluated by means of Q∗ ) is

equal to

ΦC 1
E
. Q∗ = 8 · Q∗ (ω1 ) + 5 · Q∗ (ω2 ) + 4 · Q∗ (ω3 )
B2 (1 + r) 2
1
= [8 · 0.2 + 5 · 0.22 + 4 · 0.09] = 2.94
(1.02)2
which, as expected, belongs to the no-arbitrage interval of the price of C found

previously.
Exercise 4.6 Consider a one-period (annual) model formed by a non-risky asset

(B, paying a risk-free interest rate of 4% per year) and by a stock (S) whose price is
⎧
⎨ 8; if ω = ω1
S0 = 5;
. S1 (ω) = 5; if ω = ω2 ,
⎩
3; if ω = ω3
with P (ω1 ) , P (ω2 ) , P (ω3 ) > 0.

1. Establish if the market above is free of arbitrage and complete.

2. Find the no-arbitrage price (or the no-arbitrage interval for the price) of a
European Put option on the underlying S, with strike of 5 euros and with maturity
of 1 year.
3. Suppose now that an additional asset is introduced on the market. Such an asset
is a derivative (denoted by C) with initial price of 1 euro and with payoff
⎧
⎨ 0; if ω = ω1
.Φ (ω) =
C
2; if ω = ω2 .
⎩
1; if ω = ω3
Discuss if the new market (formed by B, S and the derivative C) is free of

arbitrage and complete.
Exercise 4.7 Consider a two-period model (with annual periods) formed by a non-
risky asset (B, paying a risk-free interest rate of 4% per year) and by a stock (S)
whose price evolves as follows
S2D = S1A · u
S1A =8 → S2E = S1A · d

S0 = 5 −→ S1B = 5 −→ S2F = S1B · u
S2G = S1B · d
.
S1C = 3 −→ S2H = S1C · u
S2L = S1C · d
− − −− − − − − −− − − − − − − −−
with u = 1.5 and d = 0.5.

1. Establish if the two-period market above is free of arbitrage and complete.
2. Find the no-arbitrage price (or the no-arbitrage interval for the price) of a
European Call option on the underlying S, with strike of 7 euros and with
maturity of 2 years.
Exercise 4.8 Consider a one-period (annual) model formed by a non-risky asset

(B, paying a risk-free interest rate of 4% per year) and by two stocks (S 1 and S 2 )
whose prices evolve as follows
S11,A = 12; S12,A = 15

S11,B = 10; S12,B = 8
S01 = 10; S02 = 10
S11,C = 8; S12,C = 4
.
S11,D = 4; S12,D = 8
− − − − − − − − −− − − − − − − − − − − −−
0 1 year
Establish if the one-period market above is free of arbitrage and complete.

Chapter 5
Itô’s Formula and Stochastic Differential
Equations
Given a stochastic process .(Vt )t≥0 having trajectories with bounded variation and
a sufficiently regular function f , it is possible to define the integral of .Zt = f (Vt )
with respect to .dVt as follows
t t
. Zs dVs = f (Vs )dVs , (5.1)
0 0
i.e. in the sense of a Riemann-Stieltjes integral.

On the other hand, if the trajectories of the process .(Vt )t≥0 are not of bounded
variation and if the function f is not regular enough, the integral above may not
make sense as a Riemann-Stieltjes integral. This is the case of the Brownian motion,
for which (in general) one cannot define the integral

. f (Ws )dWs
as a Riemann-Stieltjes integral. t
A way out is to introduce Itô’s stochastic integral . 0 Hs dWs for processes
.(Ht )t≥0 that satisfy a “suitable measurability assumption” (for instance, progressive

t
measurability) together with .P 0 Hs2 ds < +∞ = 1. Under these hypotheses,
one can write the following equation
t t
Xt = X0 +
. g(Xs , s)ds + σ (Xs , s)dWs , (5.2)
0 0
where (under some suitable regularity hypothesis on g and .σ ) the first integral can
be understood as a Riemann-Stieltjes integral, while the second as an Itô integral.

90 5 Itô’s Formula and Stochastic Differential Equations
The previous expression can also be written as:

dXt = g(Xt , t)dt + σ (Xt , t)dWt
. ,
X0 = x
which is referred to as a Stochastic Differential Equation (SDE). Note that the two
formulations are equivalent: the latter is just a different way to write the former.
A fundamental result in stochastic analysis is the so-called Itô’s Lemma.
In the following, we will denote with .f (x, t) ∈ C 2,1 (R, R+ ) a function that is
continuously differentiable twice in x and once in t. Similarly for .f (x1 , . . . , xn , t) ∈
C 2,1 (Rn , R+ ).
Lemma 5.1 (Itô’s Lemma for Functions of One Variable Plus Time) Let
f (x, t) ∈ C 2,1 (R, R+ ) be a given function.
.
If the process .(Xt )t≥0 satisfies the following equation:
dXt = μ (t) dt + σ (t) dWt ,

. (5.3)
then the process .(Zt )t≥0 , defined by .Zt f (Xt , t), satisfies
∂f ∂f 1 ∂ 2f
dZt =
. (Xt , t) dt + (Xt , t) dXt + (Xt , t) σ 2 (t) dt
∂t ∂x 2 ∂x 2

∂f ∂f σ 2 ∂ 2f ∂f
= +μ + 2
(Xt , t) dt + σ (t) (Xt , t) dWt . (5.4)
∂t ∂x 2 ∂x ∂x
For functions of two or more variables plus time, Itô’s Lemma can be formulated
as follows.
Lemma 5.2 (Itô’s Lemma for Functions of Two or More Variables Plus Time)
Let .f (x1 , x2 , . . . , xn , t) be a function in .C 2,1 (Rn , R+ ) depending on n variables
.x1 , . . . , xn and on t.
If, for any .i = 1, 2, . . . , n, .Xi = Xti t≥0 satisfies the following equation:
dXti = μi (t) dt + σ i (t) dWti ,

. (5.5)
then the process .(Zt )t≥0 , defined by .Zt f Xt1 , Xt2 , . . . , Xtn , t , satisfies
∂f 1 ∂f 1
n
dZt =
. Xt , . . . , Xtn , t dt + Xt , . . . , Xtn , t dXti
∂t ∂xi
i=1
∂ 2f 1
n
1
+ σ i σ j ρij Xt , . . . , Xtn , t dt
2 ∂xi ∂xj
i,j =1
⎡ ⎤
∂f
n
∂f 1
n
∂ 2f
=⎣ + μi + σ i σ j ρij ⎦ Xt1 , . . . , Xtn , t dt
∂t ∂xi 2 ∂xi ∂xj
i=1 i,j =1
∂f 1
n
+ σi Xt , . . . , Xtn , t dWti , (5.6)
∂xi
i=1

where .ρij = E dW i dW j /dt is the correlation between the (standard) Brownian
motions .W i and .W j .
When f does not depend on time t, Eq. (5.4) simplifies to

σ 2
dZt = μf (Xt ) +
. f (Xt ) dt + σf (Xt ) dWt , (5.7)
2
while Eq. (5.6) reduces to

⎡ ⎤
∂ 2f ⎦ 1
n n
∂f 1
dZt = ⎣
. μ i
+ i j
σ σ ρij Xt , . . . , Xtn dt
∂xi 2 ∂xi ∂xj
i=1 i,j =1
∂f 1
n
+ σi Xt , . . . , Xtn dWti . (5.8)
∂xi
i=1
We finally recall the notion of Brownian local time and the Tanaka formula. The
occupation time of a set A by time t of a Brownian motion is defined by .YtA
t
0 1A (Wu ) du, where .1A (x) is the indicator function of the set A taking value 1 if
.x ∈ A, taking value 0 if .x ∈
/ A. The Brownian local time .Lt is the Radon-Nikodym
derivative of .Yt with respect to the Lebesgue measure on the real line (they can be
proved to be equivalent), i.e.

YtA =
. Lt (u) du. (5.9)
A
The Tanaka formula provides a representation of the function of a Brownian

motion .f (x) = (x − K)+ = max {x − K; 0} in terms of the Brownian local time
of K. Namely,
t
+ + 1
. (Wt − K) = (W0 − K) + 1(K,+∞) (Wu )dWu + Lt (K), (5.10)
0 2
where .Lt (K) can be proved to be given by

t
Lt (K) =
. δK (Wu )du (5.11)
0
and .δK (x) is the Dirac Delta distribution defined by:

t t
. δK (Wu )du = lim n 1[K− 1 ,K+ 1 ) (Wu )du . (5.12)
0 n→+∞ 0 2n 2n
For an exhaustive treatment we recommend the texts by Björk [6], Mikosch [32],
Øksendal [34] and Pascucci [35] among many others.
Exercise 5.3 Let (Wt )t≥0 be a standard Brownian motion, and assume St1 t≥0
and
St2 t≥0 satisfy the following SDEs:
dSt1 = μ1 St1 dt + σ1 St1 dWt ;

. S01 = s01 > 0
dSt2 = μ2 St2 dt + σ2 St2 dWt ; S02 = s02 > 0
with μ1 , μ2 ∈ R and σ2 > σ1 > 0.

1. For f (x) = ln x, find the stochastic differential equation satisfied by the process
f St1 t≥0 and the SDE satisfied by f St2 t≥0 .
1
S
2. Find the stochastic differential equation satisfied by Yt = g St1 , St2 = ln t2
St
when μ = μ1 = μ2 . Establish whether (Yt )t≥0 can be a geometric Brownian
motion and/or a Brownian motion and/or a Brownian motion with drift.
3. Do the same as above but this time for Zt = h St1 , St2 = ln St1 · St2 .
Solution
1. Since f (x) = 1
x and f (x) = − x12 , Itô’s formula implies that
1 2
df St1 = d ln St1 = f St1 dSt1 + f St1 σ1 St1 dt
.
2
1 1 1 2
= 1 dSt1 − σ1 St1 dt
St 2 S1 2
t
1 1
= 1 μ1 St1 dt + σ1 St1 dWt − σ12 dt
St 2

1
= μ1 − σ12 dt + σ1 dWt .
2
It follows that f St1 t≥0

satisfies the following stochastic differential equation:

df St1 = μ1 − 12 σ12 dt + σ1 dWt
. .
f S01 = ln S01
Note that the previous SDE implies

1 2

1
. ln St = ln S0 + μ1 − σ1 t + σ1 Wt .
1
2
Hence

1 2
1
.St = S01 · exp μ1 − σ1 t + σ1 Wt . (5.13)
2
Proceeding as above, we get

df St2 = μ2 − 12 σ22 dt + σ2 dWt
. .
f S02 = ln S02
2. In order to find the stochastic differential equation satisfied by g St1 , St2 , we can
proceed (at least) in two ways.
One (longer) is to apply Itô’s formula to the function g.
S1
Another is based on the fact that Yt = g St1 , St2 = ln t2 = ln St1 −
St
ln St2 (using the properties of logarithms, as S01 , S02 > 0). From item 1. and
from μ = μ1 = μ2 it follows that

dYt = d ln St1 − d ln St2
.

= df St1 − df St2
1 2
= σ2 − σ12 dt + (σ2 − σ1 ) dWt .
2
Hence (Yt )t≥0 satisfies the following SDE:
⎧
⎨ dYt = 1
2
2
σ1 dt + (σ2 − σ1 ) dWt
σ2 −
2
. S01 .
⎩ Y0 = ln
S02
We deduce that (Yt )t≥0 is neither a Brownian motion, nor a Brownian motion
with drift, nor a geometric Brownian motion, yet it is a Brownian motion with
drift μ∗ = 12 σ22 − σ12 and diffusion σ ∗ = σ2 − σ1 .
3. Proceeding as above, we obtain

.dZt = d ln St1 + d ln St2

= df St1 + df St2

1 2
= 2μ − σ + σ22
dt + (σ1 + σ2 ) dWt .
2 1
Hence, (Zt )t≥0 is a Brownian motion with drift μ̂ = 2μ − 1

2 σ12 + σ22 and
diffusion σ̂ = σ1 + σ2 .
Exercise 5.4 Assume that the price dynamics of two stocks is represented by the
processes St1 t≥0 and St2 t≥0 , solutions of the following SDEs:
dSt1 = μ1 St1 dt + σ 1 St1 dWt1 ,

. S01 = s1 ,
dSt2 = μ2 St2 dt + σ 2 St2 dWt2 , S02 = s2 ,
where Wt1 t≥0 , Wt2 t≥0

are independent standard Brownian motions, μ1 , μ2 ∈ R
and σ 1 , σ 2 > 0.
1. Find the SDE satisfied by
2
.f St1 , St2 = St1 − St1 St2 − K,
with K > 0, representing the value of a derivative on S 1 and S 2 .

2. Find the SDE satisfied by
2
g t, St1 , St2 = St1 − St1 St2 − Kt.
.
Solution
1. By Itô’s formula in several variables (see (5.8)),
∂f ∂f
df St1 , St2 = S 1 , S 2 dSt1 + ∂S 1 2 2
2 St , St dSt
∂S 1 t t
∂2f 2 σ 1 S 1 2 dt + ∂ f
2 2
+ 12 1
2 St , St t t
1 2
2 St , St σt2 St2 dt
∂ (S 1 ) ∂ (S 2 )
2f

.
+ 12 2 ∂S∂1 ∂S 1 2 1 1 2 2
2 St , St σ St σ St ρ12 dt
= 2St1 − St2 dSt1 − St1 dSt2

2
+ 12 2 σt1 St1 dt + 0 − 2σ 1 St1 σ 2 St2 ρ12 dt
2
= 2St1 − St2 dSt1 − St1 dSt2 + σt1 St1 dt,
∂f ∂f ∂2f
because, in our case, ∂x1 (x1 , x2 ) = 2x1 −x2 , ∂x2 (x1 , x2 ) = −x1 , ∂x12
(x1 , x2 ) =
∂2f ∂2f
2,
∂x22
(x1 , x2 ) = 0, ∂x1 ∂x2 (x1 , x2 ) = −1 and ρ12 = 0 (by the independence of
the two Brownian motions).
2
Furthermore, f S01 , S02 = S01 − S01 S02 − K.
2. Using Itô’s formula in several variables (see (5.6)) and proceeding as above, we
obtain
∂g ∂g ∂g
dg t, St1 , St2 = t, S 1 , S 2 dt + ∂S 1 t, St , St dSt +
1 2 1 t, St1 , St2 dSt2
∂t t t ∂S 2
∂2g 2 σ 1 S 1 2 dt
+ 12 1
2 t, St , St t t
∂ (S 1 )
∂2g 2 σ 2 S 2 2 dt
. + 1
2 t, St , St t t
∂ (S 2 )
2

+ 12 2 ∂S∂1 ∂S g 1 2 1 1 2 2
2 t, St , St σ St σ St ρ12 dt
2
= −Kdt + 2St1 − St2 dSt1 − St1 dSt2 + σt1 St1 dt,
2
with g 0, S01 , S02 = S01 − S01 S02 .
Exercise 5.5 Consider a stock price evolving as the following geometric Brownian
motion
dSt = μSt dt + σ St dWt ,

. (5.14)
with current stock price S0 = 40 euros, drift μ = 0.1 and volatility σ = 0.4 (per
year).
1. Establish whether the probability to exercise a European Call option is greater
than the probability to exercise a European Put option, both with strike of 24
euros, maturity of 2 years and written on the stock above.
2. Suppose now that the price S̃T of another stock at time T = 2 (years) is a multiple
of the price of the first stock, i.e. S̃T = cST for some c > 0. Establish if there
exist c > 0 such that the probability to exercise the European Put option having
the second stock as underlying is at least twice the probability to exercise the
corresponding Call option.
3. Find the greatest strike K ∗ making the Call payoff φ = (ST − K ∗ )+ greater than
6 euros (or equal) with probability at least of 20%.
Solution As already pointed out in Exercise 5.3 (see Eq. (5.13)), the stock price is
given by:

1
.St = S0 exp μ − σ 2 t + σ Wt . (5.15)
2
1. In order to exercise the European Call option, the stock price at maturity should
be greater than or equal to the strike. Hence,
P ({ Call is exercised}) = P (ST ≥ K)

.
= P (ln(ST ) ≥ ln(K))

1
= P ln (S0 ) + μ − σ 2 T + σ WT ≥ ln (K)
2
⎛ ⎞
ln(K) − ln(S0 ) − μ − 12 σ 2 T
= P ⎝WT ≥ ⎠.
σ
WT
Since WT ∼ N (0; T ) and, consequently, √ ∼ N (0; 1), we get
T
⎛ ⎞
WT ln(K) − ln(S0 ) − μ − 1 2
2 σ T
.P ({Call is exercised}) = P ⎝ √ ≥ √ ⎠
T σ T
⎛ ⎞
ln(K) − ln(S0 ) − μ − 12 σ 2 T
= 1−N⎝ √ ⎠
σ T
⎛ ⎞
ln(24) − ln(40) − 0.1 − 12 (0.4)2 · 2
= 1−N⎝ √ ⎠
0.4 · 2
= 1 − N (−0.974) = N (0.974) = 0.835,
where, as usual, N(·) denotes the cumulative distribution function of a standard

normal.
In order to exercise the European Put option, at maturity the stock price should
not exceed the strike. Since St is a continuous random variable for any fixed t > 0
(by Eq. (5.15)), we deduce that P (ST ≤ K) = P (ST < K), so
P ({Put is exercised}) = P (ST ≤ K) = P (ST < K)

.
= 1 − P (ST ≥ K)
= 1 − P ({Call is exercised}) = 0.165.
The probability of exercising the Call option is therefore greater than the one of
exercising the Put option.
2. Suppose now that S̃T = cST for some c > 0. We have to establish if there exist
some suitable c > 0 so that
P ({new Put is exercised}) ≥ 2P ({new Call is exercised}) .

. (5.16)
In order to solve this problem, we start with computing the probabilities above.
On the one hand,

.P ({new Call is exercised}) = P S̃T ≥ K

K
= P (cST ≥ K) = P ST ≥ ,
c
which corresponds to the probability to exercise a Call with strike K/c and with
the first stock as underlying. By the arguments of item 1., we get
P ({new Call is exercised}) = P ST

≥ K
c
ln(K/c)−ln(S0 )− μ− 12 σ 2 T
=1−N √
σ T
.

ln(K)−ln(c)−ln(S0 )− μ− 12 σ 2 T
=1−N √ .
σ T
On the other hand,

P ({new Put is exercised}) = 1 − P S̃T ≥ K

. ln(K)−ln(c)−ln(S0 )− μ− 12 σ 2 T
=N √ .
σ T
Equation (5.16) becomes, therefore,

⎛ ⎞
ln(K) − ln (c) − ln(S0 ) − μ − 12 σ 2 T
.3N ⎝ √ ⎠≥2
σ T

1 2 √
ln(K) − ln(S0 ) − μ − σ T − q2/3 · σ T ≥ ln (c) ,
2
where q2/3 denotes the quantile at level α = 2/3 of the standard Normal, defined
as the solution of N q2/3 = 2/3. Since q2/3 = 0.4307, it follows 0 < c ≤
0.452.
3. First, notice that
P (ST − K)+ ≥ 6 = P (ST ≥ K + 6)

=P (ln ST ≥ ln (K + 6))
ln K+6
S0 − μ− 12 σ 2 T
=P WT ≥ σ
.

1 2
WT ln K+6
S0 − μ− 2 σ T
= P √ ≥ √
T σ T

K+6
ln S − 0.1− 21 (0.4)2 ·2
= 1−N 0 √ .
0.4 2
Since such a probability has to be greater than (or equal to) 20%, we deduce that

1
ln K+6
S0 − 0.1− 2 (0.4) ·2
2
1−N √ ≥ 0.2
0.4· 2

1
ln K+6
S0 − 0.1− 2 (0.4) ·2
2
N √ ≤ 0.8
0.4· 2

1
ln K+6
S0 − 0.1− 2 (0.4) ·2
2
.
√ ≤ 0.8416
0.4· 2

ln K+6
40 ≤ 0.516
K ≤ 61.01.
The largest ‘admissible’ strike is, then, K ∗ = 61.01 euros.

Exercise 5.6 Consider a standard Brownian motion (Wt )t≥0 and a stochastic
process (St )t≥0 satisfying the following stochastic differential equation:
dSt = μt dt + σt dWt ,
. (5.17)
with μt and σt not constant everywhere but time-dependent.

1. When σt = 0.1 for t ∈ [0, 4] and

0.04t; for 0 ≤ t ≤ 2
μt =
. ,
0.02 (10 − t) ; for 2 < t ≤ 4
compute: (a) the probability of having (at time t = 4) a profit greater than (or
equal to) 0.4; (b) the average profit at time t = 4.
2. When μt is the same as above, while

0.3; for 0 ≤ t ≤ 2
σt =
. ,
0.4; for 2 < t ≤ 4
compute: (a) the probability of having a profit at least equal to 0.4 at time t = 4;
(b) the average profit at time t = 4.
3. Set Xt St − S0 for any t ≥ 0. Discuss whether there exist (μt )t≥0 and (σt )t≥0
such that (Xt )t≥0 is a martingale (with respect to the filtered space considered at
the beginning).
Solution
1. By integrating Eq. (5.17), we obtain
t t
St = S0 +
. μs ds + σs dWs .
0 0
Hence
4 4
S4 − S0 =
. μs ds + σs dWs
0 0
2 4 4
= 0.04 s ds + 0.02 (10 − s) ds + 0.1 dWs
0 2 0
!2 !4
s2 !
!
= 0.02 s 2 ! + 0.02(10s − ) !! + 0.1 · (W4 − W0 )
0 2 2
= 0.36 + 0.1 · (W4 − W0 )
= 0.36 + 0.1 W4 .
(b) It is straightforward to obtain that the average profit at time t = 4 is equal to

E [S4 − S0 ] = 0.36 + 0.1 · E [W4 ] = 0.36, where the last equality is due to a
well-known property of standard Brownian motions.
For item (a), we deduce that
P ({profit at time t = 4 is at least equal to 0.4})

= P (0.36 + 0.1 W4 ≥ 0.4)
. W4
= P (0.1 W4 ≥ 0.04) = P √ ≥ 0.04
√
4 0.1 4
= 1 − N (0.2) = 0.42.
2. Under the hypothesis on μt and on σt , we get

4 4
S4 − S0 =
. μs ds + σs dWs
0 0
2 4 2 4
= 0.04 s ds + 0.02 (10 − s) ds + 0.3 dWs + 0.4 dWs
0 2 0 2
. = 0.36 + 0.3 (W2 − W0 ) + 0.4 (W4 − W2 )

= 0.36 + 0.3 W2 + 0.4 (W4 − W2 ) .
By the properties of standard Brownian motions we know, indeed, that W2

and (W4 − W2 ) are independent Normal random variables. We remind also that:
X ∼ N m X ; sX 2 and Y ∼ N m ; s 2 are independent
Y Y
⇒ aX + bY ∼ N amX + bmY ; a 2 sX 2 + b2 s 2 for real numbers a and b.
Y
Consequently, we deduce that S4 − S0 ∼ N (0.36; 0.09 · 2 + 0.16 · 2), so
S4 − S0 ∼ N (0.36; 0.5). The average profit at time t = 4 is then equal to
E [S4 − S0 ] = 0.36, while
P (profit at time t = 4 is at least equal to 0.4)

4 − S0 ≥ 0.4)
= P (S
S4 −S
√ 0 −E[S4 −S0 ] ≥ 0.4−E[S 4 −S0 ]
.
=P √
V (S4 −S
0) V (S4 −S0 )
= 1−N √ 0.04
= 0.48.
0.5
3. Taking μt = 0 and σt = σ (with σ ∈ R) for any t ≥ 0, we get

t
Xt = St − S0 =
. σ dWs = σ Wt
0
for any t ≥ 0. By the properties of Brownian motions and by the arguments

above, it follows that (Xt )t≥0 is a martingale (when μt ≡ 0, σt ≡ σ ).
Exercise 5.7 By applying the Tanaka formula and the definition of Brownian local
time, prove that the payoff of a European Call option, on a stock whose dynamics is
described by a geometric Brownian motion with null drift and diffusion coefficient
σ , admits the following representation:
T T
1
. (ST − K)+ = (S0 − K)+ + 1(K,+∞) dSt + σ 2 St2 δK (St )dt.
0 2 0
Solution A naive application of Itô’s formula would provide:

T t
1
f (ST ) = f (S0 ) +
. f (St )dSt + σ 2 f (St )dt,
0 2 0
but in order to apply correctly the Itô’s formula we must verify the condition that
both the first and the second derivative of the function f exist and are continuous.
Actually, Itô’s formula holds also if there is a finite number of points where f
is not defined, provided that f is continuous on the whole domain of f (and
|f | is bounded wherever it is defined). This is not the case of the function under
consideration. We can nevertheless apply an alternative procedure which could
provide the desired result. Define the following sequence of functions:
⎧
⎨ 0; x ≤ K − 2n
1
fn (x) =
.
n
(x − K) +
2 1
(x − K) + 8n ;
1
K − 2n < x < K +
1 1 .
⎩ 2 2 2n
x − K; x ≥ K + 2n
1
Hence
⎧
⎨ 0; x ≤ K − 2n
1

.fn (x) = n (x − K) + 12 ; K − 2n < x < K +
1 1
⎩ 2n
1; x ≥ K + 2n
1
and it can be immediately verified that all the functions fn are continuous on the
whole domain of f . Moreover, their second derivatives
⎧
⎨ 0; x < K − 2n
1
,

.fn (x) = K − 2n < x < K +
1 1
n;
⎩ 2n
0; x > K + 2n
1
are not defined at the points x = K ± 2n

1
(even if their absolute values are bounded
everywhere else). Itô’s formula can therefore be applied to the functions fn , so that
T T
1
fn (ST ) = fn (S0 ) +
. fn (St )dSt + σ 2 fn (St ) St2 dt (5.18)
0 2 0
since the stock price St is assumed to solve dSt = σ St dWt .

We can also remark that the sequence (fn )n∈N converges to the function f , and
the sequence of the first derivatives converges to
⎧
⎨ 0; x<K

.f (x) = lim fn (x) = ; x=K .
1
n→∞ ⎩2
1; x>K
By passing to the limit in the first integral appearing in Eq. (5.18), we notice that
the value of f at the single point x = K will not affect its value, hence that integral
can be expressed as follows:
T
. 1(K,+∞) (St )dSt . (5.19)
0
By collecting the results and passing to the limit n → +∞ (this is possible by

the dominated convergence theorem), we finally obtain:
T T
1
f (ST ) = f (S0 ) +
. 1(K,+∞) (St )dSt + σ 2 lim n St2 1 1 1
(S )dt
t
2 n→∞ K− 2n ,K+ 2n
0 0
which, by (5.12), can be written as:

T T
+ + 1
. (ST − K) = (S0 − K) + 1(K,∞) (St )dSt + σ 2 St2 δK (St )dt.
0 2 0
Exercise 5.8 Consider a stock price (St )t≥0 evolving as the following geometric
Brownian motion
dSt = μSt dt + σ St dWt .

.
Find the stochastic differential equation satisfied by f (St , t) = ert · ln St4 − K

and compare it with the one satisfied by g (St ) = ln St4 − K (where K is a given
positive real number).
Exercise 5.9 The financial position Xt (at time t) evolves as
dXt = μdt + σt dWt ,

.
with X0 = 80,000 euros, μ = 400 (per year) and with σt (per year) given by
"8
k=0 100 (k + 1) 1[k;k+1) (t) ; 0≤t <9
σt =
. (5.20)
σ ∗; t ≥9
where σ ∗ > 0. (We remind that the indicator function 1A (t) equals 1 when t ∈ A,
and 0 when t ∈
/ A.)
1. Find the distribution of Xt (as a function of t).
2. Make Xt explicit as function of X0 , μ, σt , t and Wt .
3. Find σ ∗ such that P (X10 < 10000) < 100 1

. Is such a σ ∗ reasonable from a
financial point of view?
4. Compute E [X100 − X10 ], V (X100 − X10 ), E [X200 − X10 ] and V (X200 − X10 )
for (σt )t≥0 as in (5.20).
Due to the properties of Brownian motions, the quantities above are easy to
compute. Is the same true for E [X200 /X10 ] and V (X200 /X10 )?
Exercise 5.10 Let St1 , St2 and St3 be the prices of three different stocks at time t.
Suppose that St1 t≥0 , St2 t≥0 , St3 t≥0 satisfy
.dSt1 = μ1 St1 dt + σ 1 St1 dWt1

dSt2 = μ2 St2 dt + σ 2 St2 dWt2
dSt3 = μ3 St3 dt + σ 3 St3 dWt3 ,
respectively, where W 1 , W 2 and W 3 are standard Brownian motions.

1. Find the dynamics of the value f St1 , St2 , St3 of the portfolio formed by four
shares of the first stock, two shares of the second and one share of the third.
2. Recall that cash invested (or borrowed) in a bank account at a risk-free rate r
evolves as follows:
.Xt = X0 ert .
Applying the fact above, find the dynamics of the value g St1 , St2 , St3 , t of the
portfolio formed by four shares of the first stock, two shares of the second and
one of the third and with a bank loan equal to the quantity of cash necessary (at
the initial time) to buy the stocks.
Chapter 6
Partial Differential Equations in Finance
Let u be a function of several real variables .(t, x1 , x2 , . . . , xn ):
u : R+ × Rn →R
.
(t, x1 , x2 , . . . , xn ) → u (t, x1 , x2 , . . . , xn )
An equation involving one or more partial derivatives of the function u with

respect to its variables (sometimes also u itself) is called a Partial Differential
Equation (PDE). Every function u satisfying the equation is called a solution.
The notion of PDE generalizes in a natural way the notion of Ordinary
Differential Equation (ODE) to the case where the unknown map depends on more
than one variable. While the most general solution of an ODE depends in general
on some arbitrary constants, the set of solutions satisfying a PDE is parameterized
by arbitrary functions. Sometimes it is necessary to determine a solution satisfying
extra conditions, for example some initial and/or boundary conditions. These
conditions are also necessary in order to have existence, uniqueness and stability
results for the corresponding solutions.
PDEs are one of the richest and most fruitful subjects of Mathematical Analysis.
A detailed discussion, even of the most basic results of the theory, would require a
much deeper treatment than that allowed in a summary like the present one. Below
we shall just recall in a very concise way the concepts necessary to understand and
solve the exercises we are going to propose, and we prompt the interested reader
to consult one of the several textbooks written on the subject. We recommend, for
instance, Salsa [40] and, concerning the most important applications of derivatives
valuation, Wilmott et al. [43].
In our applications we shall focus on PDEs for functions of two real variables of
semilinear type, where the partial derivatives of highest order appear always with
power 1.
106 6 Partial Differential Equations in Finance
Semilinear PDEs of second order can be classified according to some basic

properties. In the particular framework we have chosen to work, i.e. real functions
of two real variables, the classification can be described in the following way.
Once mixed partial derivatives of second order have been eliminated (possibly
by a linear transformation of the independent variables), one faces the following
possibilities:
• if both second-order derivatives appear with the same sign, the equation is said
to be of elliptic type;
• if both second-order derivatives appear with opposite sign, the equation is said to
be of hyperbolic type;
• if the equation contains only one second-order derivative, and the first-order
derivative in the other variable, the equation is of parabolic type.
For parabolic equations a further distinction is introduced: if the two derivatives
with highest order with respect to the two variables have the same sign, the equation
is called backward parabolic, otherwise forward parabolic.
The classification of a PDE provides some useful information about the kind of
conditions to assign in order to ensure existence of solutions. It is possible to prove,
for example, that the initial-value problem for elliptic equations is ill-posed: in this
case in fact, the existence of a solution is not guaranteed and, if a solution exists, it
can be not unique and/or it can depend on the initial data in a discontinuous manner.
The Black-Scholes equation
∂u 1 2 2 ∂ 2 u ∂u
. + σ S + rS − ru = 0 (6.1)
∂t 2 ∂S 2 ∂S
(see Chap. 7) is the most popular PDE in financial applications. The unknown
function depends on two independent variables .S, t and the equation involves
first derivatives in both, but only the second derivative in S. Moreover, the highest
derivatives in S and t (the first derivative with respect to time t and the second with
respect to the underlying S) appear with the same sign. The Black-Scholes equation
is thus a backward parabolic PDE.
For a backward parabolic equation it is possible to prove that, under quite general
regularity conditions on both data and coefficients, given a final condition (in t) and
(possibly) two boundary conditions (in S), a solution exists, is unique and depends
continuously on the data.
For forward parabolic equations existence and uniqueness of a solution is
guaranteed once an initial (instead of final) datum is assigned.
The interest in parabolic equations, in particular in the backward parabolic ones,
for financial applications is motivated by a deep relationship between this type of
equations and diffusion stochastic processes, i.e. processes described by stochastic
differential equations driven by a standard Brownian motion. This relationship is
clarified by the following very important result, known in the literature as the
Feynman-Kac Representation Theorem.
Theorem 6.1 Let .u(x, t) be a solution of the PDE:
∂u 1 2 ∂ 2u ∂u
. + σ (x, t) 2 + μ(x, t) − ru = 0 (6.2)
∂t 2 ∂x ∂x
with final condition:
u(x, T ) = Φ(x).
. (6.3)
Then this solution can be represented as follows:
u(x, t) = e−r(T −t) Et,x [Φ(XT )],

. (6.4)
where .(Xs )s≥t satisfies the stochastic differential equation:

dXs = μ(Xs , s)ds + σ (Xs , s)dWs
. (6.5)
Xt = x
and .Et,x denotes the dependence on the variables t and x. This dependence is
inherited via the condition (6.5).
∂u
The result holds under the further hypothesis that the process .σ (s, Xs ) ∂x (s, Xs )
t 2
is square-integrable, i.e. . 0 σ (s, Xs ) ∂x (s, Xs ) ds < +∞, P -almost surely (a.s.).
∂u
The relevance of this result is primary: it allows to represent the solution of a

backward parabolic PDE as the expectation of its final condition with respect to a
probability measure determined by a suitable stochastic differential equation. The
reader eager to study in depth the relationship existing between backward parabolic
PDEs and stochastic differential equations of diffusion type can find a more detailed
treatment in the textbook by Björk [6].
Since only for a very limited number of PDEs explicit solutions are available,
numerical procedures are needed to obtain approximate solutions. Several fast and
accurate numerical methods can be used to compute solutions when these are not
available in explicit form.
Now let us concentrate on one specific class of methods allowing to get solutions
of PDEs in explicit form: the so-called similarity methods, sometimes successful in
providing parabolic PDE solutions, in particular for diffusion equations:
∂u ∂ 2u
. (x, t) = 2 (x, t) (6.6)
∂t ∂x
and some variations thereof.
The similarity method is based on the invariance properties of the PDE and its
initial (final) and boundary data under a class of transformations of the independent
variables, and it can be applied if this invariance property holds and can be identified.
It is easy to see, for example, that the classical diffusion Eq. (6.6) does not change if t
is replaced by .λ2 t and at the same time x is replaced by .λx. This invariance property,
known as scaling invariance (in this particular case the scaling is called parabolic for
obvious reasons), suggests to look for a solution of a special form, where the two
independent
√ variables appear always in the same combination, depending on the
ratio .x/ t. It is necessary that also the data exhibit the same invariance property
and, as we shall see immediately in the applications, sometimes the data themselves
suggest the particular form to look for.
Once this special combination has been detected, both the unknown and the data
of the PDE can be expressed as functions of this new variable, which becomes the
only independent variable for the problem, and the PDE is reduced to an ordinary
differential equation.
Here is a rule of thumb to find similarity solutions for second order parabolic
PDEs with two independent variables. The rule consists in finding, by trials, a
solution of the form .u = t α f (x/t β ) by looking for .α and .β such that the equation
admits a similarity reduction, i.e. it becomes an ordinary differential equation for
the unknown .U (y) in the new independent variable .y = x/t β . The applications
we are going to provide should explain in a more satisfactory way the concepts just
exposed.
Exercise 6.2 Show that, for functions a, b : R → R of class C 1 ,
V (St , t) = a(t)St + b(t)

.
satisfies the Black-Scholes equation:
∂V 1 ∂ 2V ∂V
. + σ 2 St2 2 + rSt − rV = 0 (6.7)
∂t 2 ∂S ∂S
if and only if a(t) = a and b(t) = bert with a and b real constants.
Solution First of all, we should remark that in the problem considered no bound-
ary condition has been assigned. If (6.7) is furnished with the final condition
V (ST , T ) = option payoff, then the problem admits one and only one solution:
the one provided by the Black-Scholes formula for Call or Put options.
Let us consider V (St , t) = a(t)St + b(t) and verify that if V satisfies Eq. (6.7)
then a and b are constant. We get
∂V
. = a (t)St + b (t)
∂t
∂V
= a(t)
∂S
∂ 2V
= 0,
∂S 2
and, by substituting these partial derivatives in (6.7), we can verify that, if V satisfies
Eq. (6.7), then
∂V 1 2 2 ∂ 2 V ∂V
. + σ St 2
+rSt −rV = a (t)St +b (t)+rSt a(t)−ra(t)St −rb(t) = 0
∂t 2 ∂S ∂S
or, equivalently,
a (t) St + b (t) − rb (t) = 0.

.
From the previous equation and by imposing the coefficients of both the first-
and the second-degree terms in S to be equal to zero, we get
a (t) = 0
.
b (t) − rb(t) = 0.
Consequently, if V satisfies (6.7) then a(t) = a and b(t) = bert with a, b real
constants.
The inverse implication is immediate to prove: if a and b are constant, then
V (St , t) = aSt + b satisfies Eq. (6.7).
Exercise 6.3 Suppose that the real function u (x, t) satisfies the following problem
on the real, positive half-line:
∂u ∂ 2u
. = 2, for x > 0, t > 0
∂t ∂x
with
u (x, 0) = u0 (x) , for x > 0,

.
u (0, t) = 0, for t > 0.
Find a function h(x, s, t) allowing to write the solution of the problem in the form:
+∞
u (x, t) =
. u0 (s) h(x, s, t)ds. (6.8)
0
Solution Let us start by pointing out that the Green function for the problem
considered allows to write the solution of the problem in the following form:
+∞
u (x, t) =
. u0 (s) g(x − s, t)ds, (6.9)
−∞
i.e. as a convolution integral extended to the whole real line. The function g playing
the role of the integral kernel in the previous expression is the fundamental solution
of the PDE under examination, i.e. the solution with Dirac’s function δ(x) as initial
datum. It is well known that the Green function g(x, s, t) for the present problem is
the following:

1 (x − s)2
g(x, s, t) = √ exp −
. . (6.10)
2 πt 4t
Notice that the function we are looking for is not the Green function, which
is already known, since the integral providing the solution is not necessarily of
convolution type and the integration domain is just the positive real half-line.
With this in mind, let us consider the following auxiliary problem (AP). Let the
function v (x, t) be defined by reflection with respect to x = 0, i.e.

−u (−x, t) ; for x < 0
v (x, t) =
.
u (x, t) ; for x > 0
with v0 (x) = v (x, 0) for x ∈ R.

We will verify that v (0, t) = 0 and prove that the solution of the problem:
∂v ∂ 2v
. = 2, x ∈ R, t > 0
∂t ∂x
with
v (x, 0) = v0 (x) , for x > 0

.
v (0, t) = 0, for t > 0
is
+∞
1 (x−s)2 (x+s)2
.v (x, t) = √ u0 (s) e− 4t − e− 4t ds.
2 πt 0
By definition of v and by the simple remark that u (0, t) = 0, one obtains

simultaneously the following conditions:
v 0+ , t = lim u (x, t) = 0
.
x→0+
v 0− , t = lim [−u (−x, t)] = 0.

x→0−
Since v must be continuous, then v (0, t) = 0.

From the preliminary remark, if v(x, t) is regular enough and if limx→±∞
v(x, t) = 0, we know that the following equality holds:
+∞
1 (x−s)2
v (x, t) = √
. v0 (s) e− 4t ds. (6.11)
2 πt −∞
By definition, however,

−u (−x, 0) = −u0 (−x) ; for x < 0
v0 (x) = v (x, 0) =
. (6.12)
u (x, 0) = u0 (x) ; for x > 0
By plugging (6.12) into (6.11), we obtain:

+∞
1 (x−s)2
.v (x, t) = √ v0 (s) e− 4t ds
2 πt −∞
0 +∞
1 − (x−s)
2 (x−s)2
= √ −u0 (−s) e 4t ds + u0 (s) e− 4t ds .
2 πt −∞ 0
(6.13)
0 +∞
1 (x+y)2 (x−s)2
= √ u0 (y) e− 4t dy + u0 (s) e− 4t ds (6.14)
2 πt +∞ 0
+∞ 2
+∞
1 − (x+y) (x−s)2
= √ − u0 (y) e 4t dy + u0 (s) e− 4t ds
2 πt 0 0
+∞ +∞
1 (x+z) 2 (x−z)2
= √ − u0 (z) e− 4t dz + u0 (z) e− 4t dz
2 πt 0 0
+∞ 2
1 (x−z) (x+z)2
= √ u0 (z) e− 4t − e− 4t dz,
2 πt 0
where equality (6.14) follows from the change of variable y = −s (dy = −ds ).
From the last equality and the relationship between the initial problem and
problem (AP), one obtains that the function h(x, s, t) we are looking for is
(x−s)2 (x+s)2
h (x, s, t) = √1
2 πt
e− 4t − e− 4t .
Remark The result just obtained can be applied to barrier option valuation and
consists in a slightly modified version of the “image method”.
Exercise 6.4 Find the similarity solution of the following mixed problem:
∂u ∂ 2u
. = 2 + 3x 2 , for x > 0, t > 0 (6.15)
∂t ∂x
with
u (0, t) = 0,
. for t > 0. (6.16)
u(x, t)
lim = 0, for t > 0. (6.17)
x→∞ x4
u (x, 0) = 0, for x > 0. (6.18)
Solution Our goal is to find a transformation of x and t of the kind:

x
ξ=
. ,
tα
generating a solution u (x, t) of the following type:
x
u (x, t) = t β U
. = t β U (ξ ) ,
tα
and to write (6.15) in terms of ξ and U . The resulting second-order ODE in U will
be much simpler to solve.
Let us begin by determining suitable coefficients α and β. First of all, in terms of
U and ξ , we get
∂u x x x
. = βt β−1 U α + t β U α −α
∂t t t t α+1
= βt β−1 U (ξ ) − αxt β−α−1 U (ξ ) = βt β−1 U (ξ ) − αξ t β−1 U (ξ ) ,
∂u x 1 x
= tβU α α
= t β−α U α = t β−α U (ξ ) ,
∂x t t t
∂ 2u x 1 x
= t β−α U α = t β−2α U α = t β−2α U (ξ ) .
∂x 2 t tα t
Equation (6.15) becomes then:
βt β−1 U (ξ ) − αξ t β−1 U (ξ ) = t β−2α U (ξ ) + 3x 2

.
βt β−1 U (ξ ) − αξ t β−1 U (ξ ) = t β−2α U (ξ ) + 3ξ 2 t 2α

and, dividing both sides by t 2α ,

t β−2α−1 βU (ξ ) − αξ U (ξ ) = t β−4α U (ξ ) + 3ξ 2 .
. (6.19)
In order to eliminate the explicit dependence on t in the previous equation, let us

choose α and β such that:

β − 2α − 1 = 0
. .
β − 4α = 0
This implies α = 1/2 and β = 2.

The transformation induced on x and t is then given by
x
ξ=√
.
t
and we are looking for a solution u (x, t) of the following type:
u (x, t) = t 2 U (ξ ) .
.
By writing again (6.15) in terms of ξ and U or, in other words, Eq. (6.19) with
α = 1/2 and β = 2, one obtains
. βU (ξ ) − αξ U (ξ ) = U (ξ ) + 3ξ 2
1
U (ξ ) + ξ U (ξ ) − 2U (ξ ) = −3ξ 2 . (6.20)
2
Condition (6.16) on u becomes U (0) = 0, while condition (6.17) becomes
limξ →+∞ Uξ(ξ4 ) = 0. That is: as t → 0 we have ξ → +∞ and
u (x, t) t 2 U (ξ ) U (ξ )
0←
.
4
= = .
x x4 ξ4
We must then find the general solution of Eq. (6.20). Since this is a non-
homogeneous ordinary differential equation, its general solution will be provided
by the general solution of the associated homogeneous equation:
1
U (ξ ) + ξ U (ξ ) − 2U (ξ ) = 0
. (6.21)
2
plus a particular solution of (6.20).
Let us start from the particular solution. We can try with a polynomial, and we
immediately observe that Up (ξ ) = − 14 ξ 4 is a particular solution of (6.20).
Now we look for the general solution of Eq. (6.21). If we knew two “indepen-
dent” solutions U1H and U2H of Eq. (6.21), the general solution would be provided
by a linear combination of U1H and U2H .
By looking again for a polynomial solution, we find that U1H (ξ ) = ξ 4 +12ξ 2 +12
satisfies Eq. (6.21).
Another solution of Eq. (6.21) can be obtained using the method of variation of
constants:
U2H (ξ ) = a (ξ ) U1H (ξ ) ,
.
with a being a suitable function, to be determined. To this aim we require that U2H
satisfies Eq. (6.21).
Since

. U2H (ξ ) = a (ξ ) U1H (ξ ) + a (ξ ) U1H (ξ )

U2H (ξ ) = a (ξ ) U1H (ξ ) + 2a (ξ ) U1H (ξ ) + a (ξ ) U1H (ξ ) ,
the function a must satisfy
1 1
a U1 + 2a U1 + aU1 + ξ aU1 + ξ a U1 − 2aU1 = 0,
. (6.22)
2 2
where for notational simplicity the dependence on ξ has been omitted and U1
denotes U1H . Since U1H satisfies (6.21), Eq. (6.22) can be written as:
1
a U1 + 2a U1 + ξ a U1 = 0,
.
2
i.e.
a ξ U1
. = − − 2 . (6.23)
a 2 U1
By denoting by h the function a , the differential Eq. (6.23) can be written

as a first-order ordinary differential equation whose solution can be obtained via
separation of variables.
Excluding h (ξ ) ≡ 0, we obtain:
ξ2
. ln (h (ξ )) = ln(h(0)) − − 2 ln(ξ 4 + 12ξ 2 + 12), (6.24)
4
hence
1
h (ξ ) = a (ξ ) = a (0) e−ξ
2 /4
.
2
.
ξ4 + 12ξ 2 + 12
Now, in order to get U2H we have two possibilities. First, we can integrate a to
get a and consequently U2H . Second (more simply), looking at a we notice that a
good candidate for U2H is
ξ 1 2
−ξ 2 /4
f (ξ ) e
. + g (ξ ) e− 4 s ds,
−∞
where f (ξ ) and g(ξ ) are polynomials in ξ , the former of degree not higher than
3 and the latter of degree not higher than 4. The coefficients of these polynomials
can be easily computed by substituting f (ξ ) and g(ξ ) in the equation that U2H must
satisfy.
Proceeding as described above we obtain another solution of Eq. (6.21), indepen-
dent of U1H :
ξ
1 2 1 2
U2H (ξ ) = 2ξ ξ 2 + 10 e− 4 ξ + ξ 4 + 12ξ 2 + 12
. e− 4 s ds.
0
The general solution of the homogeneous Eq. (6.21) is then

H
UG
. (ξ ) = aU1H (ξ ) + bU2H (ξ )
and, consequently, the general solution of the non-homogeneous (6.20) is
UG (ξ ) = Up (ξ ) + aU1H (ξ ) + bU2H (ξ ) .
.
U (ξ )
By imposing U (0) = 0 and limξ →+∞ ξ4
= 0, it is possible to determine the
constants a and b. By the first condition we get a = 0. Since limξ →∞ Uξ(ξ4 ) =
√
− 14 + b π , from the second condition we get b = 4√1 π .
The similarity solution required is then:
1 1 1 2
U (ξ ) = − ξ 4 + √ (ξ 3 + 10ξ )e− 4 ξ +
.
4 2 π
ξ
1 s2
+ √ (ξ 4 + 12ξ 2 + 12ξ ) e− 4 ds.
4 π 0
Finally, the solution of the initial problem (6.15)–(6.18) is

x
u (x, t) = t 2 U
. √
t
1 1 √ 3 1 2
= − x4 + √ tx + 10xt 3/2 e− 4 ξ +
4 2 π
x/√t
1 s2
+ √ x + 12x t + 6xt
4 2 3/2
e− 4 ds.
4 π 0
It is easy to check that this expression solves our problem by direct substitution
into (6.15).
Exercise 6.5 By applying the Feynman-Kac representation formula, find the solu-
tion of the following problem on the domain [0, T ] × R:
∂V ∂V 1 ∂ 2V
. + μx + σ 2x2 2 = 0 (6.25)
∂t ∂x 2 ∂x
V (T , x) = ln x 4 + k
with given constants μ, σ and k.

Solution By the Feynman-Kac representation formula, we know that the solution
of the problem:
∂V ∂V 1 ∂ 2V
. + m (t, x) + s 2 (t, x) =0
∂t ∂x 2 ∂x 2
V (T , x) = Φ (x)
is
V (t, x) = E [Φ (XT )] ,
.
where, for u ≥ t

dXu = m (u, Xu ) du + s (u, Xu ) dWu
. .
Xt = x
In the present case, m (t, x) = μx, s (t, x) = σ x and Φ (x) = ln x 4 + k. The

solution of the problem proposed is then given by:
V (t, x) = E [Φ (XT )] ,
.
where for u ≥ t

dXu = μXu du + σ Xu dWu
. (6.26)
Xt = x
Since the solution of (6.26) is given by

1
Xu = Xt exp
. μ − σ 2 (u − t) + σ (Wu − Wt )
2

1 2
= x exp μ − σ (u − t) + σ (Wu − Wt ) ,
2
we obtain that

1
XT = x exp
. μ − σ 2 (T − t) + σ (WT − Wt )
2
and

1
. ln XT4 + k = ln x 4 exp 4 μ − σ 2 (T − t) + 4σ (WT − Wt ) +k
2

1 2
= ln x + 4 μ − σ (T − t) + 4σ (WT − Wt ) + k.
4
2
We finally get the solution of the problem proposed:
V (t, x) = E [Φ (XT )]
.

1
= E ln x 4 + 4 μ − σ 2 (T − t) + 4σ (WT − Wt ) + k
2

1 2
= ln x + 4 μ − σ (T − t) + k,
4
2
once we observe E [WT − Wt ] = 0 (W is a standard Brownian motion).

Exercise 6.6 In the Black-Scholes setting, consider an option written on an under-
lying with the following dynamics:
dSt = μSt dt + σ St dWt

.
with S0 = 20 euro, μ = 0.16 and σ = 0.36 (per year). The risk-free interest rate is
0.04 per year.
By applying the Feynman-Kac formula, compute the initial value of the option
with underlying S and payoff
+
Φ (ST ) = ln ST2 − K
.
with strike K = 6 euros and with maturity T = 1 year.

Solution In the Black-Scholes model the value of the option above at time t is
assumed to be a function of St and t (denoted by F (St , t)) satisfying the following
PDE:
∂F ∂F 1 ∂ 2F
. + rx + σ 2 x 2 2 − rF = 0
∂t ∂x 2 ∂x
+
F (T , ST ) = ln ST2 − K = Φ (ST ) .
By the Feynman-Kac formula, we know that the solution of
∂F ∂F 1 ∂ 2F
. + m (t, x) + s 2 (t, x) 2 − rF = 0
∂t ∂x 2 ∂x
F (T , x) = Φ (x)
is given by
F (t, x) = e−r(T −t) E [Φ (XT )] ,

.
where, for u ≥ t

dXu = m (u, Xu ) du + s (u, Xu ) dWu
. .
Xt = x
+
Since, in our case, m (t, x) = rx, s (t, x) = σ x and Φ (x) = ln x 2 − K , the
solution of the initial problem is given by
F (t, x) = e−r(T −t) E [Φ (XT )] ,

.
where, for u ≥ t

dXu = rXu du + σ Xu dWu
. . (6.27)
Xt = x
Since the solution of Eq. (6.27) is

1
Xu = x exp
. r − σ 2 (u − t) + σ (Wu − Wt ) ,
2
we obtain that

1
. XT = x exp r − σ 2 (T − t) + σ (WT − Wt )
2
and

1 2
. ln XT2 = ln x exp 2 r − σ (T − t) + 2σ (WT − Wt )
2
2

1
= ln x 2 + 2 r − σ 2 (T − t) + 2σ (WT − Wt ) .
2
We finally get the initial value of the option considered:
.F (S0 , 0) = e−rT E [Φ (XT )]

+
−rT 1 2
=e E ln X0 + 2 r − σ T + 2σ WT − K
2
2

+
= e−rT E ln S02 + 2r − σ 2 T + 2σ WT − K .
There remains then to compute

+
E
. ln S02 + 2r − σ 2 T + 2σ WT − K .
Since (Wt )t≥0 is a standard Brownian motion and T = 1,
WT = W1 ∼ N (0; 1) .
.
Accordingly, if we denote by Z = W1 ∼ N (0; 1), we get:

+
−r
F (S0 , 0) = e
. E 2 ln (S0 ) + 2r − σ 2 + 2σ Z − K

+
= e−r 2 ln (S0 ) + 2r − σ 2 + 2σ z − K fZ (z) dz
R

+ 1 z2
−r
=e 2 ln (S0 ) + 2r − σ 2 + 2σ z − K √ e− 2 dz
R 2π
+∞
−r 1 z2
=e K−2 ln(S0 )−2r+σ 2
2 ln (S0 ) + 2r − σ 2 + 2σ z − K √ e− 2 dz
2σ
2π
+∞ 1 z2
= e−r K−2 ln(S0 )−2r+σ 2
2 ln (S0 ) + 2r − σ 2 − K √ e− 2 dz
2σ
2π
+∞
−r 1 z2
+e K−2 ln(S0 )−2r+σ 2
2σ z √ e− 2 dz
2σ
2π

K − 2 ln (S0 ) − 2r + σ 2
. = e−r 2 ln (S0 ) + 2r − σ 2 − K 1−N
2σ
+∞
1 z2
+e−r −2σ √ e− 2
2π K−2 ln(S0 )−2r+σ 2
2σ

K − 2 ln (S0 ) − 2r + σ 2
= e−r 2 ln (S0 ) + 2r − σ 2 − K 1−N
2σ
2
K−2 ln(S0 )−2r+σ 2
1 −
+e−r 2σ √ e 8σ 2
2π
= 0.25,
where N (·) denotes the cumulative function of a standard normal random variable.
The initial value of the option is then 0.25 euros.
Exercise 6.7 Compute the solution φ(x, t) of the diffusion equation:
∂φ ∂ 2φ
. = , x ∈ R, t > 0, (6.28)
∂t ∂x 2
satisfying the following conditions:
φ(x, 0) = δ(x), limx→±∞ φ(x, t) = 0,

. +∞
limx→±∞ ∂φ∂x = 0, −∞ φ(x, t)dx = 1.
Solution The solution to be determined is called the fundamental solution of

the diffusion equation; it can be used to find solutions of the diffusion equation
satisfying more general data, as we shall see later.
A first method to solve this initial/boundary-value problem is based on the
Fourier transform. The Fourier transform of a function f (x, t) is defined as
+∞
ϕ(k, t)
. f (x, t)eikx dx. (6.29)
−∞
Notice that, by integrating twice by parts, the following relations hold for the
Fourier transforms of the derivatives of f (under the hypothesis that both f and its
first derivative in x vanish at infinity):
+∞ +∞ +∞
∂f ikx
. e dx = f (x, t)eikx − ik f (x, t)eikx dx = −ikϕ,
−∞ ∂x −∞ −∞
+∞ 2 +∞ +∞
∂ f ikx ∂f ikx ∂f ikx
2
e dx = e − ik e dx = −k 2 ϕ.
−∞ ∂x ∂x −∞ −∞ ∂x
It is then immediate to verify that the function φ must satisfy
∂ϕ
. = −k 2 ϕ, (6.30)
∂t
with the initial condition
ϕ(k, 0) = 1.
. (6.31)
The solution satisfying the required initial condition of the previous ODE can be
found immediately
ϕ(k, t) = e−k t .
2
. (6.32)
Now, in order to find the solution φ(x, t) of the original problem we have to
invert the Fourier transform:
+∞
1
.φ(x, t) = ϕ(k, t)e−ikx dx. (6.33)
2π −∞
The explicit computation of the integral (easily performed by completing the square
appearing in the argument of the exponential function) provides the following result
2
1 x
φ(x, t) = √ exp −
. . (6.34)
2 πt 4t
Another method for solving the problem is based on a similarity technique. We

look for a solution f (x, t) of the following form: f (x, t) = t α U (ξ ), with ξ = x/t β .
The form of the PDE suggests for β the value β = 1/2, while the condition on the
integral of the unknown extended to the entire real line suggests for α the value
α = 1/2. The similarity solution we are looking for is then of the following kind:
1
f (x, t) = √ U (ξ ).
. (6.35)
t
The similarity reduction provides the following ODE for the new unknown U (ξ ):
ξ 1
U (ξ ) + U (ξ ) + U (ξ ) = 0,
. (6.36)
2 2
together with boundary conditions:
. lim U (ξ ) = 0
ξ →±∞
+∞
U (ξ )dξ = 1.
−∞
A straightforward computation provides the solution U (ξ ):
ξ2
U (ξ ) = Ae−
. 4 + B. (6.37)
√ the first boundary condition implies B = 0, the second implies A =

While
1/(2 π ). Hence the solution of our original problem φ(x, t) becomes:
2
1 x
.φ(x, t) = √ exp − . (6.38)
2 πt 4t
Exercise 6.8 Let φ be the fundamental solution of the diffusion equation obtained
in the previous exercise. Verify that the function
+∞
f (x, t)
. φ(ξ − x, t)u(ξ )dξ (6.39)
−∞
is a solution of the diffusion equation
∂f ∂ 2f
. = 2, x ∈ R, t > 0, (6.40)
∂t ∂x
satisfying the initial condition f (x, 0) = u(x).

Solution We already know that the function φ(x, t) is a solution of the diffusion
equation satisfying the initial condition φ(x, 0) = δ(x). It is immediate to verify
that the diffusion equation is invariant under the change of variable x → ξ − x,
i.e. if φ(x, t) is a solution, φ(ξ − x, t) is also a solution, but satisfying the initial
datum φ(ξ − x, 0) = δ(ξ − x). Moreover the diffusion equation is linear, so that
multiplying the equation:
∂φ ∂ 2φ
. = (6.41)
∂t ∂x 2
by u(ξ ) and integrating in ξ , the function

+∞
f (x, t)
. φ(x − ξ, t)u(ξ )dξ (6.42)
−∞
satisfies the diffusion equation with initial datum

+∞ +∞
f (x, 0) =
. φ(x − ξ, 0)u(ξ )dξ = δ(x − ξ )u(ξ )dξ = u(x). (6.43)
−∞ −∞
Exercise 6.9 Find a suitable change of variables turning the Black-Scholes equa-
tion into a PDE with constant coefficients.
Solution The Black-Scholes PDE, holding for the value F (St , t) of every deriva-
tive, written on an underlying asset whose price dynamics is described by a
geometric Brownian motion, and consistent with the no-arbitrage requirement, is
the following:
∂F 1 ∂ 2F ∂F
. + σ 2 S 2 2 + rS − rF = 0. (6.44)
∂t 2 ∂S ∂S
This PDE exhibits an explicit dependence on the variable S, denoting the underlying
value, in both the coefficients of the terms involving the first and the second
derivatives in S.
As a geometric Brownian motion is the exponential of a Brownian motion (with
drift), an “educated guess” suggests the following change of variable:
. x = ln(S),
S = ex ,
F (S, t) = h(x(S), t),
h(x, t) = F (S(x), t).
By computing the partial derivatives of the new unknown with respect to the new
independent variables, we get
∂F ∂h ∂x 1 ∂h
. = = ,
∂S ∂x ∂S S ∂x

∂ 2F ∂ 2 h ∂x 2 ∂h ∂ 2 x 1 ∂ 2h 1 ∂h
2
= 2
+ 2
= 2 2
− 2 .
∂S ∂x ∂S ∂x ∂S S ∂x S ∂x
By plugging these expressions into the original equation, one obtains

∂h 1 2 ∂ 2 h σ 2 ∂h
. + σ + r− − rh = 0, (6.45)
∂t 2 ∂x 2 2 ∂x
where the coefficients’ dependence on the independent variables has been dropped.
Exercise 6.10 Find a suitable change of variables turning the Black-Scholes
equation for a European Call option into the diffusion equation with proper initial-
boundary conditions.
Solution Solving the problem requires two steps. In the first, we shall remove the
dependence on the independent variable S in the PDE coefficients. To this end,
inspired both by Exercise 6.9 and by the functional form of the European Call option
payoff, we exploit a change of variables of the following kind:
. x = ln(S/K),
S = Kex ,
σ2
τ= (T − t),
2
2
t = T − 2τ
σ
F (S, t) = Kh(x(S), τ (t)),
h(x, t) = K −1 F (S(x), t (τ )),
where K is the strike of the European option considered. We introduce the new time
variable τ , which is the dimensionless version of the “time to maturity” variable
(T − t); this change of sign in the time variable allows to turn the parabolic type
of the Black-Scholes equation (parabolic backward) into the parabolic type of the
diffusion equation (parabolic forward). Moreover, we introduce the new parameter
q = 2r/σ 2 to simplify the notation. After these changes, the Black-Scholes equation
becomes
∂h ∂ 2 h ∂h
. − 2 − (q − 1) + qh = 0. (6.46)
∂τ ∂x ∂x
The final condition of the Black-Scholes equation for the European Call option is its
payoff F (S, T ) = max(S − K; 0); expressed in the new variables, it can be written
as h(x, 0) = max(ex − 1; 0).
As far as the second step is concerned, we try to remove the terms involving h
and its first derivative in x. To this end we introduce a further change of variables of
the following kind:
h(x, τ ) = exp{αx + βτ }g(x, τ ),

. (6.47)
where α, β have to be determined so that the required terms can be removed, and we
look for the PDE that the new function g must satisfy. After a direct computation of
all the derivatives of the old function in terms of the new variables, we obtain

∂g ∂ 2g ∂g ∂g
βg +
. = 2 + (q − 1) αg + + 2α + (α 2 − q)g. (6.48)
∂τ ∂x ∂x ∂x
To remove the terms involving g and its first derivative in x, the coefficient α, β
must fulfill the following conditions:

2α + (q − 1) = 0
. .
β = α 2 + (q − 1)α − q
The required values are: α = −(q − 1)/2, β = −(q + 1)2 /4. By choosing then α
and β as above, the function g appearing in

q −1 (q + 1)2
h(x, τ ) = exp −
. x− τ g(x, τ ), (6.49)
2 4
must satisfy the diffusion equation for x ∈ (−∞, +∞) and τ > 0:
∂g ∂ 2g
. = 2, (6.50)
∂τ ∂x
with
q+1 q−1

g(x, 0) = max e 2 x − e 2 x ; 0 .
. (6.51)
Exercise 6.11 Find the explicit solution of the Black-Scholes equation for a
European Call option by using the results obtained in the previous exercises.
Solution By the previous exercise we know that the Black-Scholes equation for a
European Call option can be reduced, by a suitable change of variables (which we
performed in two separate steps), to the diffusion equation:
∂g ∂ 2 g,
. = (6.52)
∂τ ∂x 2
with initial datum:
q+1 q−1

g(x, 0) = max e 2 x − e 2 x ; 0 .
. (6.53)
Moreover, we know from Exercise 6.8 that the solution of the diffusion equation
satisfying a general initial datum can be expressed, via an integral representation
formula, as a convolution of the fundamental solution:
+∞
. g(x, τ ) φ(ξ − x, τ )g(ξ, 0)dξ, (6.54)
−∞
where

1 (ξ − x)2
.φ(ξ − x, τ ) = √ exp − . (6.55)
2 πτ 4τ
Hence, in order to find explicitly this solution, we must compute the following
integral:

+∞ 1 (ξ − x)2 q+1 q−1
g(x, τ )
. √ exp − max e 2 ξ − e 2 ξ ; 0 dξ. (6.56)
−∞ 2 πτ 4τ
We remark that the support of the integrand is simply ξ > 0. Due to monotonicity
of the exponential function ( q > 1), we have indeed
q+1 q−1
e
. 2 ξ −e 2 ξ > 0 ⇐⇒ ξ > 0. (6.57)
So, by changing the integration domain, the integral can be written in the following
way:

+∞ 1 (ξ − x)2 q+1 ξ q−1

g(x, τ )
. √ exp − e 2 − e 2 ξ dξ. (6.58)
0 2 πτ 4τ
Let us consider separately the two contributions:

+∞ 1 (ξ − x)2 q+1
J1 =
. √ exp − e 2 ξ dξ
0 2 πτ 4τ

+∞ 1 (ξ − x)2 q−1
J2 = √ exp − e 2 ξ dξ.
0 2 πτ 4τ
√ √
By introducing the new variable z = (ξ − x)/ 2τ (dz = dξ/ 2τ ), the integral J1
can be written as follows:
+∞ 2 q+1 √
1 z x+z 2τ
.J1 = √ − 2
√ exp e dz, (6.59)
2π −x/ 2τ 2
by completing the square in the argument of the exponential function, it can be

computed in the following way:
⎧ ⎫
q+1 √
2
+∞
⎪
⎨ z− 2τ ⎪
⎬
1 q+1 (q+1)2 2
J1 = √ e 2 x
. √ e 4 τ
exp − dz
2π −x/ 2τ ⎪
⎩ 2 ⎪
⎭
+∞ 2
1 q+1 (q+1)2 y
= √ x
e 2 e 4 τ
√ exp − dy,
2π x q+1
−√ − 2 2τ 2
2τ
√
where we introduced a further integration variable y = z − (q + 1) 2τ /2 (dy =
dz). The last integral can be immediately expressed in terms of the standard normal
distribution function, simply observing that:
+∞ 2 x 2
u u
. exp − du = exp − du = N(x). (6.60)
−x 2 −∞ 2
Hence we finally get:

q +1 (q + 1)2 x q + 1√
.J1 = exp x+ τ N √ + 2τ . (6.61)
2 4 2τ 2
The integral J2 can be computed similarly:

q −1 (q − 1)2 x q − 1√
.J2 = exp x+ τ N √ + 2τ . (6.62)
2 4 2τ 2
In order to find the solution to our original problem (the Black-Scholes equation
for a European Call option) we need to write our solution in terms of the original
variables. From
σ2
. x = ln(S/K), τ= (T − t),
2
C(S, t) = K · h(x(S), τ (t)),

q −1 (q + 1)2
h(x, τ ) = exp − x− τ g(x, τ ),
2 4
we obtain the well-known formula:
C(S, t) = S · N(d1 ) − Ke−r(T −t) N(d2 ),

. (6.63)
where d1 , d2 are defined, as usual, by

2
ln S
K + r + σ2 (T − t)
d1 =
. √ ,
T −t
2
ln S
K + r − σ2 (T − t)
d2 = √ .
T −t
Exercise 6.12 By using a similarity reduction method, derive an explicit valuation

formula for Exchange options, i.e. for options written on two different underlying
assets with values S1 (t), S2 (t) and with payoff given by CEX (S1 (T ), S2 (T )) :=
max S1 (T ) − S2 (T ); 0, by assuming that the dynamics of the two underlying assets
is described by two geometric Brownian motions driven by two Wiener processes
W1 (t), W2 (t) correlated with correlation coefficient ρ, i.e.
. dS1 (t) = μ1 S1 (t)dt + σ1 dW1 (t)

dS2 (t) = μ2 S2 (t)dt + σ2 dW1 (t),
with dW1 (t)dW2 (t) = ρdt.

Solution We must solve the following two-dimensional Black-Scholes equation:
∂f ∂f ∂f ∂ 2 f, ∂ 2 f, ∂ 2 f,
. + rS1 + rS2 + σ12 S12 2 + σ22 S22 2 + ρσ1 σ2 − rf = 0.
∂t ∂S1 ∂S2 ∂S1 ∂S2 ∂S1 ∂S2
(6.64)
with final datum:
f (S1 , S2 , T ) = max {S1 − S2 ; 0} .

. (6.65)
By introducing the new independent variable z := S1 /S2 and the new unknown
f (S1 , S2 , t) := S2 g(S1 /S2 , t), and observing that ∂f ∂g ∂f 1 ∂g ∂g
∂t = S2 ∂t , ∂S1 = S2 S2 ∂z = ∂z ,
∂f ∂ 2 f, 1 ∂ 2 g, ∂ 2 f, 2 ∂ 2 f, 2
= g − S2 S12 ∂g ∂g
∂z = g − z ∂z , = S2 ∂z2 , ∂S 2 = z Sz2 ∂∂zg,
2 , = − Sz2 ∂∂zg,
2 ,
∂S2 S2 ∂S12 2 ∂S12
the previous PDE for g becomes:
∂g 1 ∂ 2g
. + σ 2 z2 2 = 0, (6.66)
∂t 2 ∂z
where
.σ 2 = σ12 + σ22 − 2ρσ1 σ2 , (6.67)

and the final datum:
g(z, T ) = max (z(T ) − 1; 0).

. (6.68)
The PDE just written is of Black-Scholes type, with r = 0, K = 1, so we can

write immediately its solution as follows:
g(z, t) = zN(d1 ) − N(d2 ),

. (6.69)
where:
1
d1 =
. √ [ln z + σ 2 T /2]
σ T −t
1
d2 = √ [ln z − σ 2 T /2].
σ T −t
By introducing the old variables f, S1 , S2 ) the previous formula can be written

as follows:
f (S1 , S2 , t) = S1 N(d1 ) − S2 N(d2 )

.
1 S1
d1 = √ [ln ( ) + σ 2 T /2]
σ T −t S2
1 S1
d2 = √ [ln ( ) − σ 2 T /2],
σ T −t S2
and this is the solution to our problem.
Exercise 6.13 In a Black-Scholes setting, consider an option with underlying

evolving according to the following dynamics:
dSt = μSt dt + σ St dWt ,

.
with S0 = 16 euros, μ = 0.12 and σ = 0.3. The risk-free interest rate r available
on the market is 0.04. The time unit is 1 year.
1. If the payoff of the option considered is
+ +
Φ (ST ) = ln (ST /K)3
. = ln ST3 − 3 ln K
with K = 20 euros and maturity T = 1, compute the price of the option today
(t = 0) by using the Feynman-Kac representation formula.
2. What is the event with higher probability: to have a strictly positive payoff for
the option of item 1. or for a European Call option with the same K and T ? Or
are their probabilities the same?
3. Is the probability to get a payoff at least of 10 euros for the option of item 1.
higher than for a European Call option with the same K and T ? Or are their
probabilities the same?
Exercise 6.14 By solving the Black-Scholes equation, equipped with the proper
final condition, find the formula for the European Put option price.
Chapter 7
Black-Scholes Model for Option Pricing
and Hedging Strategies
As in the previous chapters, we consider a market model consisting in two assets:

one non-risky (bond), the other risky (stock). While before we focused on discrete-
time market models, here we introduce the so-called Black-Scholes model: a well-
known example of continuous-time market model.
The key assumption behind the Black-Scholes model is that the stock price S
evolves as a geometric Brownian motion, i.e.

1 2
.St = S0 exp μ − σ t + σ Wt , (7.1)
2
where .S0 is the initial stock price, .μ the drift, .σ the volatility (or diffusion parameter)
and .(Wt )t≥0 is a standard Brownian motion. By Itô’s Lemma, the price of S is
described by the following Stochastic Differential Equation:

dSt = μSt dt + σ St dWt
. .
S0 = s0
The bond price B is evaluated by continuous compounding with a risk-free rate

r. Namely, if .B0 is the initial price of the bond:
Bt = B0 exp(rt) ,
. (7.2)
or, equivalently,

dBt = rBt dt
. .
B0 = b0
132 7 Black-Scholes Model for Option Pricing and Hedging Strategies
Furthermore, the model is based on the following assumptions: perfect liquidity

(one can buy or sell any real number of an asset), short selling is allowed, transaction
costs are never applied, bid and ask prices of any asset coincide. Moreover, the
market model is also assumed to be free of arbitrage.
In the Black-Scholes model, it has been proved (see [6, 21, 33] for details) that
there exists an equivalent martingale measure useful for the pricing of a derivative
and consistent with the no-arbitrage assumption. The equivalent martingale measure
is also called “risk-neutral”; roughly speaking, in fact, under such a measure the
underlying evolves (more or less) as it would with a non-risky asset. Under the
risk-neutral measure, the underlying price is described by the following Stochastic
Differential Equation:

dSt = rSt dt + σ St dWt
. .
S0 = s0
It was also shown that the Black-Scholes model, as well as the binomial model,
is complete. The Second Fundamental Theorem of Asset Pricing implies therefore
the uniqueness of the risk-neutral measure for such a market model. For complete
market models, the price of any derivative is uniquely determined by the no-
arbitrage assumption.
Denote by .F (St , t) the price (at time t) of a derivative having S as underlying.
Such a price can be obtained by solving the following Partial Differential Equation
(called Black-Scholes equation):
∂F σ 2 ∂ 2F ∂F
. + + rS − rF = 0, (7.3)
∂t 2 ∂S 2 ∂S
endowed with some suitable final and boundary conditions. Equivalently, .F (St , t)
can be obtained by the discounted expected value of the payoff of the derivative,
under the equivalent martingale measure and conditioned by .Ft (where .(Ft )t≥0 is
the filtration generated by .(St )t≥0 ). In other words,
.F (St , t) = e−r(T −t) EQ [F (ST , T )|Ft ]. (7.4)
Black-Scholes Formula for European Call/Put Options on Stocks Without

Dividends
Consider a European Call option, whose payoff is given by
C(ST , T ) = max {ST − K; 0} ,

.
where K stands for the strike of the option and .ST for the underlying price at
maturity. The solution of the Black-Scholes PDE (as well as the discounted expected
value of the payoff under the equivalent martingale measure) gives the following
price:
C(St , t) = St N(d1 ) − Ke−r(T −t) N(d2 ),

. 0≤t <T (7.5)
where .N (·) denotes the cumulative distribution function of a standard normal and
d1 , d2 are defined as:
.
2
ln(St /K) + (r + σ2 )(T − t)
.d1 √ . (7.6)
σ T −t
2
ln(St /K) + (r − σ2 )(T − t) √
d2 √ = d1 − σ T − t (7.7)
σ T −t
For the pricing of a Put option one can proceed similarly by taking into account that
the final value (equal to the payoff) of a Put is given by
P (ST , T ) = max {K − ST ; 0} .
.
An alternative way to evaluate the Put option is based on the relation between the
prices of Call and a Put options written on the same underlying, with same strike and
maturity. The prices of Call and Put options as above, indeed, satisfy the so-called
Put-Call parity:
C(St , t) − P (St , t) = St − Ke−r(T −t) .

. (7.8)
Proceeding as explained above, the price (at time t) of a European Put option with
maturity T , with strike K and written on the underlying S is given by
P (St , t) = Ke−r(T −t) N(−d2 ) − St N(−d1 ),

. 0 ≤ t < T. (7.9)
Black-Scholes Formula for European Call/Put Options on Stocks Paying

“Continuous” Dividends
The stock S is said to pay “continuous” dividends when the dividend rate .D0 is
constant in time, that is, the sum .D0 Sdt is distributed in the infinitesimal interval dt
to the owners of the stock.
The Black-Scholes formula for a European Call option written on an underlying
paying continuous dividends can be obtained by (7.5), by replacing r with the risk-
free rate .(r − D0 ) in the expressions of .d1 and .d2 (.d1 , d2 will denote the new values)
and by replacing .St with .St e−D0 (T −t) in the first term of the formula. This gives
C(St , t) = St e−D0 (T −t) N(d1 ) − Ke−r(T −t) N(d2 ),

. (7.10)
where
σ2
ln(St /K) + (r − D0 + 2 )(T − t)
.d1 √ . (7.11)
σ T −t
σ2
ln(St /K) + (r − D0 − 2 )(T − t)
d2 √ (7.12)
σ T −t
A similar formula holds for European Put options.

Black-Scholes Formula for European Call/Put Options on Stocks Paying
“Discrete” Dividends
When dividends are not paid “continuously” (as in the previous case where a
dividend rate was applied) but just at fixed times (known a priori), dividends are
said to be “discrete”.
The Black-Scholes formula for a European Call option written on a stock paying
discrete dividends is the following:

C (S0 , 0) = (S0 − D) N d̂1 − Ke−rT N d̂2 ,
. (7.13)
where D denotes the discounted value of all dividends paid until the option maturity
T and

ln ((S0 − D)/K) + r + 12 σ 2 T
.d̂1 √
σ T

ln ((S0 − D)/K) + r − 12 σ 2 T √
d̂2 √ = d̂1 − σ T
σ T
A similar formula holds for European Put options written on stocks paying
discrete dividends.
As for options on stocks without dividends, a Put-Call Parity holds for European
Call/Put options with the same maturity and strike and written on the same stock
paying discrete dividends:
C (S0 , 0) − P (S0 , 0) = S0 − D − Ke−rT .

.
In order to simplify notations, in the following we will often write .Ct instead of
C (St , t) and .Pt instead of .P (St , t).
.
The Black-Scholes model allows not only to compute explicitly the price of a
European Call/Put option on a given underlying S, but also to find the so-called
hedging strategy that the seller of a derivative should have to be hedged against the
risk assumed. An hedging strategy can be static (sometimes called “buy and hold”
strategy) or dynamic. In the first case, the portfolio composition, once built, will be
unchanged for the whole duration of the strategy; in the dynamic case, continuous
(or periodic) changes in the composition of the portfolio would be needed.
The simplest dynamic hedging strategy is the one called Delta-hedging, where
an “instantaneously riskless” portfolio is composed by one derivative and by .−Δ
(Delta) shares of the underlying (the negative sign denotes a “short” position). The
‘Delta’ of a derivative of value .F (St , t) is defined as:
∂F
. ΔF , (7.14)
∂S
and, obviously, it changes with time. By the Black-Scholes formula, one can deduce
that the Delta of a European Call option on a stock without dividends is given by:
ΔCall = N(d1 ),
.
while the Delta of the corresponding Put option is given by
ΔP ut = N (d1 ) − 1.
.
Notice that .ΔCall > 0, while .ΔP ut < 0.

From a theoretical point of view, in order to keep the portfolio riskless one
should be able to modify its composition at any instant of time. Obviously, this
is unreasonable and too expensive in practice. Nevertheless, the portfolio should
be re-balanced at intervals of time close enough to keep the riskiness under control.
When the value of the derivative varies so quickly that it becomes difficult to control
the risk associated to such a variation, Gamma-hedging strategies are used. The
‘Gamma’ .Γ of a derivative is defined as follows:
∂ 2F
ΓF
. . (7.15)
∂S 2
In the Black-Scholes model, the Gamma of a European Call/Put option written on a
stock without dividends is given by:
N (d1 )
ΓCall = ΓP ut =
. √ , (7.16)
σ St T − t
where .N (·) denotes the density function of a standard normal.

The Gamma-hedging strategy consists in composing a portfolio whose Gamma
is equal to zero. In order to cancel both Delta and Gamma of a portfolio (in such a
case the portfolio is said to be Delta and Gamma-neutral), the portfolio is required
to be composed by more than two assets (differently from the Delta-neutral case).
We will explain how to apply this idea in the exercises.
Besides Delta and Gamma, that quantify the riskiness of a derivative associated
to changes in the underlying price, the Greek letters .ρ (Rho), .ν (Vega) and .θ
(Theta) are used to index the risk associated to changes in the interest rate r, in the
volatility .σ and in time t, respectively. Because of their symbols, these quantities
are collectively known as “Greeks”. They are used to introduce portfolios that are
either Rho-neutral, Vega-neutral or Theta-neutral.
For a European Call option on a stock without dividends, .ρ, .ν and .θ are given by
∂F
ρ
. = K(T − t)e−r(T −t) N(d2 ). (7.17)
∂r
∂F √
ν = St N (d1 ) T − t (7.18)
∂σ
∂F σ St N (d1 )
.θ =− √ − rKe−r(T −t) N(d2 ). (7.19)
∂t 2 T −t
For hedging strategies we will adopt the convention (and approximation) not
to discount their future cash-flows. This approximation is supported by the short
duration of such strategies.
By composing portfolios formed by suitable shares of the underlying assets and
of the derivatives one can achieve different risk profiles, which fit different goals and
risk attitudes of investors. Some of these tools will be considered in the exercises
below.
The most popular combinations of European options in use in order to obtain
some of the risk profiles mentioned above are the following:
• Spreads: Bull Spread (two calls written on the same underlying, with the same
maturity, but .K1 < K2 , long position on .C1 , short position on .C2 ), Bear Spread
(two calls written on the same underlying, with the same maturity, but .K1 <
K2 , short position on .C1 , long position on .C2 ), Butterfly Spread (4 call options
written on the same underlying, with the same maturity, but .K1 < K2 < K3 ,
.K2 = (K1 + K3 )/2, long position on .C1 and .C3 , two short position on .C2 ).
• Straddles: a long position on one call and one put written on the same underlying,
with the same maturity and the same strike.
• Strangles: a long position on one call and one put written on the same underlying,
with the same maturity, but different strikes, the strike of the call greater than the
strike of the put.
• Strips (long position on one call and two puts written on the same underlying,
with the same maturity and the same strike) and Straps (long position on two
calls and one put written on the same underlying, with the same maturity and the
same strike).
For a detailed treatment and further details on the subject, we send the interested
reader to Björk [6], Hull [25] and Wilmott et al. [43, 44], among many others.
Options can be written also on exchange rates. If the exchange rate dynamics
is described by a geometric Brownian motion with diffusion coefficient .σX , it is
possible to prove that a European call option written on an exchange rate .X(t) :=
units of domestic currency/units of foreign currency, can be evaluated like a usual
call option written on a dividend distributing asset, where the (constant) dividend
rate is given by the foreign risk-free interest rate:
C(X(t), t) = xe−rf (T −t) N(d1 ) − Ke−rd (T −t) N(d2 ),

. (7.20)
ln (x/K) + (rd − rf + σX2 /2)

. d1 = √ . (7.21)
σ T −t
ln (x/K) + (rd − rf − σX2 /2)
d2 = √ , (7.22)
σ T −t
and .rd , rf are the risk-free interest rates for the domestic and the foreign currency
respectively.
Exercise 7.1 It is well known that, in market models that are free of arbitrage, the
prices of a European Call and a European Put option written on the same stock
(without dividends) satisfy
C0 ≤ S0
. and P0 ≤ Ke−rT , (7.23)
where K stands for the strike of the options, T for the maturity of the options and r
for the risk-free rate per year.
1. Verify that
. − Ke−rT ≤ C0 − P0 ≤ S0
and find an arbitrage opportunity in the following cases:

(a) C0 − P0 > S0 ;
(b) C0 − P0 < −Ke−rT .
2. It is well known that, in market models that are free of arbitrage, the following
Put-Call Parity holds for European options written on the same underlying paying
discrete dividends:
C0 − P0 = S0 − D − Ke−rT ,
.
where D stands for the discounted value of all the dividends paid until the
maturity T of the options.
(a) Find an arbitrage opportunity when C0 = P0 + S0 − Ke−rT and the
discounted value of dividends is strictly positive (D > 0).
(b) Show that, under the no-arbitrage assumption, C0 > S0 − D can never occur.
Solution
1. From inequalities (7.23) it follows immediately that
.C0 − Ke−rT ≤ C0 − P0 ≤ S0 − P0 .
Since C0 , P0 ≥ 0 (see below), we deduce
. − Ke−rT ≤ C0 − Ke−rT ≤ C0 − P0 ≤ S0 − P0 ≤ S0 ,
that is, the claim.

Let us verify that C0 , P0 ≥ 0. Suppose by contradiction that C0 < 0 and
consider the following strategy:
t =0 t =T
Buy the call

⇒ −C0 > 0 (ST − K)+
.
Invest (−C0 ) at rate r

⇒ +C0 −C0 erT > 0
−C0 + C0 = 0 (ST − K)+ − C0 erT > 0
Consequently, if C0 were strictly negative then there would exist arbitrage

opportunities (for instance, the one built above). This would contradict the no-
arbitrage assumption, hence it follows that C0 ≥ 0.
Similarly, one can check P0 ≥ 0.
(a) Assume now that C0 > P0 + S0 .

As shown below, the following strategy represents an arbitrage opportu-
nity.
t =0 t =T
Sell the call

⇒ +C0 − (ST − K)+
Buy one stock and one put

.
⇒ −S0 − P0 ST + (K − ST )+
Invest (C0 − S0 − P0 )
⇒ − (C0 − S0 − P0 ) (C0 − S0 − P0 ) erT
C0 − S0 − P0 − (C0 − S0 − P0 ) = 0 K + (C0 − S0 − P0 ) erT > 0
(b) Suppose now that C0 − P0 < −Ke−rT , from which C0 − P0 < 0.

As checked below, the following strategy represents an arbitrage oppor-
tunity.
t =0 t =T
Sell the put

⇒ +P0 − (K − ST )+
Buy the call

.
⇒ −C0 (ST − K)+
Invest (P0 − C0 )
⇒ − (P0 − C0 ) (P0 − C0 ) erT
P0 − C0 − (P0 − C0 ) = 0 ST − K + (P0 − C0 ) erT > ST > 0
2. Note, first, that the Put-Call Parity (in the presence of dividends) can be rewritten
as
C0 + D = S0 + P0 − Ke−rT .
.
(a) It is clear that if we had C0 = P0 + S0 − Ke−rT for D > 0, then there would
exist arbitrage opportunities.
Suppose indeed that C0 = P0 + S0 − Ke−rT and consider the following
portfolio:
• one long position in the stock;
• one long position in the Put;
• one short position in the Call;
• bank loan of Ke−rT .
The value V0 of such a portfolio at t = 0 is then equal to
.V0 = −S0 − P0 + C0 + Ke−rT = 0,
while its value VT at maturity T of the two options is equal to
VT = ST + DerT + (K − ST )+ − (ST − K)+ − K = DerT > 0.

.
The portfolio above represents then an arbitrage opportunity.

(b) Suppose now that C0 > S0 − D.
The following strategy is an arbitrage opportunity.
t =0 t =T
Sell the call

⇒ +C0 − (ST − K)+
Buy one share of the underlying

⇒ −S0 ST + DerT
.
Invest/borrow (C0 − S0 )
⇒ − (C0 − S0 ) (C0 − S0 ) erT
C0 − S0 − (C0 − S0 ) = 0 − (ST − K)+ + ST

+DerT + (C0 − S0 ) erT
In fact, the initial value of such a strategy is zero, while its final value
equals
VT = − (ST − K)+ + ST + DerT + (C0 − S0 ) erT

.
= min{ST ; K} + (C0 − S0 + D) erT

≥ (C0 − S0 + D) erT > 0,
where the last inequality is due to the initial assumption C0 > S0 − D.

Exercise 7.2 Consider a stock (without dividends) whose price evolves as in the
Black-Scholes model with drift μ of 10% per year, volatility σ of 40% per year and
current price of S0 = 16 euros. The risk-free interest rate r on the market is 4% per
year.
1.
(a) Compute the initial price of a European Call option on the stock above, with
strike of 18 euros and maturity T of 1 year.
(b) Deduce the price of the corresponding Put option.
2. Suppose now that in 6 months’ time the stock will cost 16.4 euros. Discuss
whether it is convenient to wait 6 months before buying the Call option above
and investing (at risk-free rate) what we would have paid for buying the Call at
the initial time.
What if the stock costed 19.2 euros in 6 months from now?
3. Discuss whether it is possible to establish a priori some bounds on the initial
prices of the Call and the Put option. If yes, verify that such bounds hold in the
present case.
4. Consider the same Call option as above but written on a stock with S0 = 16
euros, paying a dividend of 2 euros in 4 months and of 4 euros in 8 months.
(a) Compute the initial price of such an option and compare it with the price
of the Call option (of item 1.) on the stock without dividends. Determine
whether the option price on the stock paying dividends is greater than the
one not paying dividends.
(b) Discuss whether, in general, the price of a Call option on a stock with
dividends could be greater than the same without dividends. Justify the result
by means of no-arbitrage arguments.
Solution
1.
(a) Recall that the Black-Scholes formula for European Call options on stock
without dividends is the following:
.C0 = S0 N (d1 ) − Ke−rT N (d2 ) ,

where N denotes the cumulative distribution function of a standard normal

and

ln (S0 /K) + r + 12 σ 2 T
.d1 = √ ;
σ T
√
d2 = d1 − σ T .

ln (16/18) + 0.04 + 1
2 (0.4)2
d1 =
. = 0.0055
0.4
d2 = 0.0055 − 0.4 = −0.3945,
since T = 1 year.
The initial price of the European Call option above is then equal to
C0 = 16 · N (0.0055) − 18 · e−0.04 · N (−0.3945) = 2.04 euros.

.
(b) Applying the Put-Call Parity (for European options with same maturity and
strike, written on the same stock without dividends), we deduce that
P0 = C0 − S0 + Ke−rT = 2.04 − 16 + 18 · e−0.04 = 3.34 euros.

.
We would obtain the same result by applying again the Black-Scholes

formula in the case of European Put options on stocks without dividends.
In such a case, indeed,
P0 = Ke−rT N (−d2 ) − S0 N (−d1 ) .

.
2. We compute now the price of the Call option at t = 6 months when we assume
(1)
that S6m = 16.4 euros. Under this assumption,

(1)
ln S6m /K + r + 12 σ 2 (T − t)
(1)
d1
. = √
σ T −t

ln (16.4/18) + 0.04 + 12 (0.4)2 12
= = −0.117
0.4 · 12
(1) (1) 1
d2 = d1 − 0.4 · = −0.40,
2
hence

C6m = 16.4 · N d1 − 18 · e−0.04·0.5 N d2
(1) (1) (1)
. = 1.356 euros.
It would be convenient to wait 6 months before buying the Call option of item
1. and investing (at risk-free rate) what we would have paid for buying the Call
at the initial time depending on whether
(1)
.C0 · er/2 > C6m .
(1)
Since C0 · er/2 = 2.04 · e0.04·0.5 = 2.08 > C6m = 1.356, we conclude that it
would be convenient to wait before buying the Call option.
(2)
Proceeding as above, we obtain that for S6m = 19.2 euros

ln (19.2/18) + 0.04 + 1
2 (0.4)2 1
2
(2)
d1
. = = 0.440
0.4 · 12
(2) (2) 1
d2 = d1 − 0.4 · = 0.157
2
and

C6m = 19.2 · N d1 − 18 · e−0.04·0.5 N d2
(2) (2) (2)
. = 2.94 euros.
(2)
Since C0 · er/2 = 2.04 · e0.04·0.5 = 2.08 < C6m = 2.94, we conclude that in
such a case it would not be convenient to wait 6 months to buy the option.
3. As well known, prices of European options written on stocks without dividends
fulfill the following inequalities:
S0 − Ke−rT ≤ C0 ≤ S0
.
Ke−rT − S0 ≤ P0 ≤ Ke−rT
In the present case, we deduce immediately that
0 = max{−1.29; 0} = max{S0 − Ke−rT ; 0} ≤ C0 ≤ S0 = 16

.
1.29 = max{1.29; 0} = max{Ke−rT − S0 ; 0} ≤ P0 ≤ Ke−rT = 17.29.
The previous bounds are then fulfilled by the initial price of the Call and the Put.
4. Consider now a stock paying a dividend of 2 euros in 4 months and of 4 euros in

8 months.
(a) We recall that the Black-Scholes formula for European options written on
stocks paying discrete dividends is the following:

Ĉ0 = (S0 − D) N d̂1 − Ke−rT N d̂2 ,
.
where N denotes as usual the cumulative distribution function of a standard

normal and

ln ((S0 − D)/K) + r + 12 σ 2 T
.d̂1 = √
σ T
√
d̂2 = d̂1 − σ T .
Since the discounted dividends amount to

4 8
D = 2 · e−0.04· 12 + 4 · e−0.04· 12 = 5.87
.
and

ln ((16 − 5.87)/18) + 0.04 + 1
2 (0.4)2
d̂1 =
. = −1.137
0.4
d̂2 = −1.137 − 0.4 = −1.537,
it follows that the initial price of the new Call option equals
Ĉ0 = (16 − 5.87) · N (−1.137) − 18 · e−0.04 N (−1.537) = 0.22 euros.

.
Consequently, the price of the Call “without dividends” is greater than the
other one. More precisely, the difference between the two prices amounts to
C0 − Ĉ0 = 2.04 − 0.22 = 1.82 euros.
(b) In general, we cannot have Ĉ0 > C0 . Suppose by contradiction that Ĉ0 > C0
and consider the following strategy, where ST stands for the price (at time
T ) of the stock that does not pay any dividend:
t =0 t =T
Buy the call without dividends

⇒ −C0 (ST − K)+
Sell the call paying dividends

+
. ⇒ +Ĉ0 − ST − DerT − K

Invest Ĉ0 − C0

⇒ − Ĉ0 − C0 Ĉ0 − C0 erT

Ĉ0 − C0 − Ĉ0 − C0 = 0 VT
+
Since ST − DerT − K ≤ (ST − K)+ , we deduce that the final value
VT of the strategy above is
+
VT = − ST − DerT − K + (ST − K)+ + Ĉ0 − C0 erT
.

≥ Ĉ0 − C0 erT > 0.
If Ĉ0 > C0 at any time, there would exist an arbitrage opportunity (e.g. the
one just built). Under the no-arbitrage assumption, then, necessarily Ĉ0 ≤
C0 .
Exercise 7.3 Consider the same stock A—paying discrete dividends—and the
same European Call option of Exercise 7.2, item 4.
The risk-free interest rate r available on the market is of 4% per year.
1. Consider a Call option with strike of 18 euros, maturity of 1 year and written on a
stock B with the same features as stock A but paying continuous dividends (with
dividend rate d).
Find the dividend rate d such that the price of the Call option written on the
stock A paying discrete dividends coincides with the price of the Call option
above written on stock B.
2. Compute the discounted value of the dividends paid by stock B.
Solution
1. From Exercise 7.2 we already know that the price of the Call option written on
stock A (paying discrete dividends), with strike of 18 euros and maturity of 1 year
is equal to Ĉ0 = 0.22 euros.
Furthermore, we remind that the price of a Call with strike of 18 euros,
maturity of 1 year and written on stock B (paying continuous dividends) is given
by
C0d = S0 e−dT N(d1 ) − Ke−rT N(d2 ),

.
where
σ2
ln(S0 /K) + (r − d + 2 )T
.d1 = √
σ T
σ2
ln(S0 /K) + (r − d − 2 )T
d2 = √ .
σ T
Hence

ln(16/18) + 0.12 − d
C0d = 16 · e−d · N
.
0.4

−0.04 ln(16/18) − 0.04 − d
− 18 · e ·N .
0.4
We look then for a dividend rate d so that
C0d = Ĉ0 = 0.22 euros.

.
Numerically, we get
C0d
d = 0.1 1.3505
d = 0.2 0.8586
d = 0.3 0.5232
d = 0.4 0.3050
d = 0.45 0.2289
d = 0.454 0.2236
d= 0.4567 0.2200
d = 0.457 0.2196
d = 0.46 0.2158
d = 0.5 0.1697
We can conclude that the “implicit” dividend rate we are looking for is d ∼
=
0.4567.
2. Since the price of stock B evolves as

1
StB = S0B · exp
. r − d − σ 2 t + σ Wt
2
with S0B = S0 = 16, we deduce that the discounted value of all dividends paid
by stock B is equal to

D ∗ = S0 1 − e−dT = 16 · 1 − e−0.4567 = 5.866 euros.
.
As expected, D ∗ is approximately equal to the discounted value D of all

dividends paid by stock A (where D = 5.87, see Exercise 7.2).
Exercise 7.4 Consider a stock whose price evolves as in the Black-Scholes model,
with volatility σ = 0.32 (per year) and initial price S0 = 30 euros. Such a stock
pays a dividend of one euro in 3 months and of one euro in 9 months. The risk-free
interest rate available on the market is of 4% per year.
1. Compute the price of a European Call option with maturity of 1 year, strike of 25
euros and written on the stock above.
2. What is the price of the corresponding Put option?
3. Discuss whether the price would change if the risk-free interest rate decreased
from 4% to 2% per year. What would such a change affect?
4. Suppose now that the risk-free interest rate is equal to 4% per year. Compute the
new price of the stock of item 1. “adjusted” by the discounted value of dividends,
i.e. obtained by detracting the discounted value of dividends.
5. Compare the price of the Call of item 1. to the price of the Call with strike of
25 euros, maturity of 1 year and written on the stock without dividends, with
volatility σ = 0.32 (per year) and with the current price found in item 4.
Solution
1. The discounted value D of all dividends of the stock amounts to
3 9
D = 1 · e−0.04· 12 + 1 · e−0.04· 12 = 1.96 euros.
.
Recall that, by the Black-Scholes formula, the price of a European Call option
written on a stock paying discrete dividends is given by

C0 = Ŝ0 · N d̂1 − Ke−rT · N d̂2 ,
.
where Ŝ0 = S0 − D and

ln Ŝ0 /K + r + 12 σ 2 T
.d̂1 = √
σ T
√
d̂2 = d̂1 − σ T .
0.206
d̂1 =
. = 0.644
0.32
d̂2 = d̂1 − 0.32 = 0.324
and, accordingly,
C0 = 28.04 · N (0.644) − 25 · e−0.04 · N (0.324)

.
= 28.04 · 0.74 − 24.02 · 0.627 = 5.69 euros.
2. By the Put-Call parity on stocks paying dividends, we obtain immediately the

Put price:
P0 = C0 − S0 + D + Ke−rT
.
= 5.69 − 30 + 1.96 + 24.02 = 1.67 euros.
3. Changes in the interest rate affect the discounted value of the dividends paid and
the strike, as well as d1 and d2 in the Black-Scholes formula.
If the risk-free interest rate was r ∗ = 0.02, the new discounted value D ∗ of
the dividends paid by the stock would amount to
3 9
D ∗ = 1 · e−0.02· 12 + 1 · e−0.02· 12 = 1.98.
.
Hence, we would get

ln ((S0 − D ∗ ) /K) + r ∗ + 12 σ 2 T
∗
.d1 = √ = 0.579
σ T
√
d2∗ = d1∗ − σ T = 0.259
and, accordingly,
∗T
C0∗ = S0 − D ∗ N d1∗ − Ke−r
. N d2∗
= 28.02 · N (0.579) − 25 · e−0.02 · N (0.259)
= 5.38 euros.
4. In case r ∗ = 0.04, the current stock price adjusted by the discounted value of the
dividends would be
Ŝ0 = S0 − D = 28.04 euros.

.
5. The price of the Call written on the stock paying the dividends of item 1. is equal
to 5.69 euros.
If we computed the price of a Call option with maturity of 1 year, strike of 25
euros and written on a stock with volatility σ = 0.32 (per year) and current price
Ŝ0 = 28.04 euros, we would find exactly the same price as the one of the Call of
item 1.
Exercise 7.5 Consider a market model where it is possible to trade on a stock (with
current price S0 = 20 euros) and on different European Call and Put options written
on such a stock.
stock follows a geometric Brownian motion, namely St =
The price of the
S0 exp{ μ − 12 σ 2 t + σ Wt } with S0 = 20, μ ∈ R and σ > 0.
Our goal is an investment with 1 year as horizon of time.
1. Consider two European Call options (both with maturity T = 1 year and written
on the stock above), with strikes of
K1 = 20 euros
.
K2 = 40 euros
and prices
C01 = 6 euros
.
C02 = 2 euros,
respectively.
Establish under which conditions on μ and σ > 0 the profit (in 1 year) due to
a bear spread1 formed by the previous options is non-negative with probability
at least of 50%.
1A spread is a trading strategy composed by two or more options of the same kind, that is all
options are either European Calls or European Puts written on the same underlying.
2. Compare the profit due to the bear spread with profit arising from the following
strategy:
– short position in a Call with strike K1 (the lower strike)
– short selling a share of the underlying at the beginning and giving it back at
maturity.
Establish which strategy is most convenient as the price of the underlying
decreases to 16 euros. What if the price increases to 32 euros?
3. Suppose that on the market it is also possible to buy/sell Call options on the
same underlying as before but with different strikes (in particular, we could think
that options of any strike between K1 and K2 are available). Assume that on
this market (where arbitrage opportunities might exist) the prices of the options
above decrease linearly from the price of the option with strike K1 = 20 to the
price C ∗ = 4 euros for the option with strike K ∗ = 28. Furthermore, the prices
decrease linearly from the price of the option with strike K ∗ to that of the option
with strike K2 .
(a) Build a “butterfly spread”2 using the options above, which necessarily relies
on the Call options with strikes K1 and K2 .
What is the profit when we use a butterfly spread instead of the previous
strategy (see item 2.)?
(b) Suppose now that the stock price in 1 year is distributed as in item 1. with
μ = 0.24 and σ = 0.48. Compute the probability with which it is more
convenient to use the butterfly spread instead of the bear spread.
Solution
1. By assumption, the price (at time T ) of the underlying is given by

1 2
.ST = S0 exp μ − σ T + σ WT
2
with S0 = 20 and T = 1 year.

We need to find under which conditions on μ and σ the profit (in 1 year) due to
the bear spread is non-negative with probability at least of 50%. In other words,
μ and σ have to satisfy the following inequality
P ({profit of the bear spread ≥ 0}) ≥ 0.5.

. (7.24)
Among the different spreads, the so-called “bear spread” is used when a decrease in the stock
price is expected. For Call options, a bear spread is obtained by selling the Call option with lower
strike and by buying the Call with higher strike.
2 The so-called “butterfly spread” is used when we expect that the stock price remains more or less
stable. A butterfly spread for Call options can be obtained by buying one Call option with lower
strike, buying one Call option with higher strike and selling two Call options with intermediate
strike.
Then we have to analyze the random variable representing the profit due to the
bear spread.
Recall that a bear spread can be built—by means of Call options—as
follows:
(A) sell the Call option with lower strike (K1 , in the present case)
(B) buy the Call options with higher strike (K2 )
The payoff of a bear spread is then given by:
(A) (B) Total payoff
if ST ≤ K1 0 0 0
.
if K1 < ST ≤ K2 − (ST − K1 ) 0 K1 − ST
if ST > K2 − (ST − K1 ) ST − K2 K1 − K2
since the payoff of a Call option is (ST − K)+ for the buyer and − (ST − K)+
for the seller. Consequently, the profit (due to the payoff and to the Call prices)
is given by:
Total profit
if ST ≤ 20 C01 − C02 = 4
.
if 20 < ST ≤ 40 K1 − ST + C01 − C02 = 24 − ST
if ST > 40 K1 − K2 + C01 − C02 = −16
Figure 7.1 shows the profit due to the bear spread on varying the underlying
price.
Condition (7.24) reduces then to
0.5 ≤ P ({profit ≥ 0}) = P ({0 ≤ ST ≤ 20} ∪ {20 < ST ≤ 40; 24 − ST ≥ 0}) .

.
By the arguments above and by the assumptions on ST , we deduce that
P ({profit ≥ 0}) = P (0 ≤ ST ≤ 24)

.
1
μ− 2 σ 2 T +σ WT
= P 20e ≤ 24

1 24
=P μ − σ 2 + σ W1 ≤ ln
2 20
0
0 10 20 30 40 50 60 70 80
profit of the bear spread
-5
-10
-15
-20
underlying price
Fig. 7.1 Profit of the bear spread
⎛ ⎞
ln (1.2) − μ − 12 σ 2
. = P ⎝W1 ≤ ⎠
σ
⎛ ⎞
ln (1.2) − μ − 21 σ 2
=N⎝ ⎠,
σ
because W1 is distributed as a standard

normal.

ln(1.2)− μ− 12 σ 2
Since μ and σ have to verify N σ ≥ 0.5, we obtain

ln (1.2) − μ − 12 σ 2
. ≥ 0.
σ
The condition on μ and σ (> 0) becomes then
1 2
μ≤
. σ + ln (1.2) .
2
Let us compare the profit due to the bear spread with the profit due to the
following strategy:
– short position in the Call with strike K1 (the lower strike)
– short selling a share of the underlying at the beginning and giving it back at
maturity.
The profit (at maturity) of the strategy above is then given by:
Profit due to the call Profit due to the stock Total profit
. if ST ≤ K1 C01 S0 − ST 26 − ST
if ST > K1 C01 − (ST − K1 ) S0 − ST 46 − 2ST
The comparison between the strategy just considered and the bear spread is
manifest from Fig. 7.2.
Notice that “our strategy” is very convenient when the stock price decreases,
while it is very risky and could cause big losses when the stock price increases
steeply.
40
20
0
0 10 20 30 40 50 60 70 80
-20
-40
profit
bear spread
-60 alternative strategy
-80
-100
-120
-140
underlying price
Fig. 7.2 Comparison between the profit of a bear spread and the profit of our strategy
If the stock price decreased to 16 euros, we would have a total profit of 10

euros with our strategy, while a profit of 4 euros with the bear spread, making
our strategy more convenient.
If the stock price increased to 32 euros, we would have a total loss of 18 euros
with our strategy, while a loss of 8 euros with the bear spread. The bear spread
would be thus more favorable.
2. First, remind that a butterfly spread can be built—by means of Call options—as
follows:
(A) buy the Call with lower strike (K1 );
(B) buy the Call with higher strike (K2 );
K1 +K2
(C) sell 2 Calls with intermediate strike (namely, with strike K3 = 2 ).
(a) By the initial assumptions, we obtain that the prices of the Call options with
strike between K1 and K2 are given by
⎧ Ki −20
⎨6 − 4 ; if 20 ≤ Ki ≤ 28
i
.C0 = .
⎩ Ki −28
4− 6 ; if 28 ≤ Ki ≤ 40
In the present case, then, the butterfly spread would be built as explained
before where the Call options in (C) have strike K3 = K1 +K2
2
= 30 and price
C0 = 4 − 6 = 3 .
3 30−28 11
The payoff of the butterfly spread is then given by
(A) (B) (C) Total payoff
if ST ≤ K1 0 0 0 0
. if K1 < ST ≤ K3 ST − K1 0 0 ST − K1
if K3 < ST ≤ K2 ST − K1 0 −2 (ST − K3 ) K2 − ST
if ST > K2 ST − K1 ST − K2 −2 (ST − K3 ) 0
Consequently, the profit (due to the payoff and the Call prices) is given
by:
Total Profit
if ST ≤ 20 2C03 − C01 − C02 = − 23
. if 20 < ST ≤ 30 ST − K1 + 2C03 − C01 − C02 = ST − 62

3
if 30 < ST ≤ 40 K2 − ST + 2C03 − C01 − C02 = 118

3 − ST
if ST > 40 2C03 − C01 − C02 = − 23
By comparing the profit of the strategy of item 2. to the profit of the

butterfly spread, we deduce that the latter is more advantageous when the
stock price is close to the intermediate strike, while it is less advantageous
when the stock price goes under 20 euros. When the underlying price
increases, however, the butterfly spread becomes more advantageous than
“our strategy”, since loss is bounded with the butterfly spread.
(b) In order to compute the probability for which the butterfly spread is more
convenient than the bear spread, we need to compare the butterfly profit (see
Fig. 7.3) with the bear profit.
10
8
profit of the butterfly spread
0
0 10 20 30 40 50 60 70 80
-2
underlying price
Fig. 7.3 Profit of the butterfly spread

It is evident that for ST ≤ 20 the profit due to the bear is greater than the
profit of the butterfly; vice versa for ST ≥ 40. For 20 < ST < 40 the profit
of the bear is equal to 24 − ST while the profit of the butterfly is equal to
ST − 62 3 for 20 < ST < 30 and 3 − ST for 30 ≤ ST < 40.
118
So, we deduce that
for ST ≤ 20: The bear is the more advantageous

for 20 < ST < 67
3 : The bear is the more advantageous
. for ST = 3 :
67
Indifferent
for 67
3 < ST < 40: The butterfly is the more advantageous
for ST ≥ 40: The butterfly is the more advantageous
Consequently, the probability for which it is more advantageous to use a

butterfly spread than a bear spread is equal to

67 1 2 67
.P ST > = P S0 exp μ − σ T + σ WT >
3 2 3

1 67
=P μ − σ 2 T + σ WT > ln
2 3S0
⎛ ⎞
ln (67/60) − 0.24 − 12 (0.48)2
= P ⎝WT > ⎠
0.48
⎛ ⎞
ln (67/60) − 0.24 − 1
2 (0.48)2
= 1−N⎝ ⎠
0.48
= 0.512.
Notice that μ = 0.24 and σ = 0.48 fulfill the condition found in item 1.,
hence the profit due to the bear spread is non-negative with probability at
least of 50%.
Exercise 7.6 We take a short position in a European Call option on a stock “Smart-
and-Fast” with strike K1 = 8 euros and maturity T = 1 year. The price of the
underlying follows a geometric Brownian motion with S0 = 8 euros, μ = 20% and
σ = 40% per year. The risk-free interest rate available on the market is r = 4% per
year.
1.
(a) Compute the Delta (Δ1 ) of the Call option and, accordingly, establish how
many shares of stock “Smart-and-Fast” we need to buy/sell in order to make
our short position Delta-neutral.
(b) What if our short position was in a Put option (instead of a Call option)?
(c) Find the probability for which the seller of the Call has a (net) loss by the
Delta-neutral portfolio built in item (a), given that the Call is exercised.
2. Discuss whether we can make a Delta-neutral short position in a Call option (as
the one above) but with maturity T ∗ = 2 years instead of T = 1 year.
3. Suppose that another Call option on the same underlying “Smart-and-Fast” but
with strike K2 = 12 euros is available on the market.
(a) Compute Delta and Gamma (Δ2 and Γ2 ) of the new option.
(b) Establish how one could make Delta- and Gamma-neutral the portfolio with
a short position in the first Call option. Is it possible for the Delta- and
Gamma-neutral portfolio to contain a short position in the second Call?
(c) We take a short position in the Call option of item 1. Explain how to make
such a portfolio Gamma-neutral and with a total Delta smaller or equal (in
absolute value) to 1. How many shares of the underlying could we buy at
most?
4. Consider the portfolio composed by a short position in the Call (of item 1.) and
by two long positions in the Put (of item 1.). Verify whether it is possible to make
it Delta-, Gamma- and Vega-neutral if we dispose of a third Call option on the
same underlying but with strike K3 = 6 euros. If yes, explain how.
Solution Remind that:
• once a time t is fixed, the Delta of an option is defined by Δ = Δt ∂F ∂S (St , t),
where F (St , t) stands for the price at time t of the option with underlying S;
• the Delta of the underlying (ΔS ) is equal to 1;
• a portfolio is said to be Delta-neutral if the Delta of the portfolio is zero;
• for European options written on a stock without dividends:
ΔCall = N (d1 ) > 0

.
ΔP ut = N (d1 ) − 1 < 0.
1.
(a) By the arguments above, we obtain immediately that
⎛ ⎞
ln (S0 /K1 ) + r + 12 σ 2 T
.Δ1 = N (d1 ) = N ⎝ √ ⎠
σ T
⎛ ⎞
ln (1) + 0.04 + 12 (0.4)2
=N⎝ ⎠ = N (0.3) = 0.618.
0.4
Since our goal is to make a short position in the Call option Delta-neutral,
we need to find the number x of shares of the underlying to be bought (if
x > 0) or sold (if x < 0) in order to make the portfolio of stocks-option
Delta-neutral. We are then looking for x verifying
. − 1 · Δ1 + x · ΔS = 0,
where the sign “-” is due to the short position in the option.
Since ΔS = 1, x has to solve
. − 0.618 + x = 0.
So, x = 0.618. We conclude that to make a short position in the Call option
Delta-neutral we need to buy 0.618 shares of the underlying.
(b) Suppose now that the short position to be “hedged” is in a Put option instead
of a Call option. In order to make it Delta-neutral we look for a number x of
shares of the underlying to be bought (if x > 0) or sold (if x < 0) satisfying
. − 1 · Δ1,P ut + x · ΔS = 0
−1 · Δ1,P ut + x = 0.
Since Δ1,P ut = Δ1,Call − 1 = 0.618 − 1 = −0.382, to make the portfolio

above Delta-neutral we need to sell 0.382 shares of the underlying.
(c) We have to compute the probability that the seller of the Call has a net loss
by the Delta-neutral portfolio built in item (a), given that the Call option is
exercised. This reduces to compute the following conditional probability

P C0 − (ST − K1 )+ + x (ST − S0 ) ≤ 0 ST ≥ K1 .
.
First, let us compute the initial price C0 of the Call

√ option. By item (a), we
already know that d1 = 0.3, hence d2 = d1 − σ T = −0.1 and
C0 = S0 · N (d1 ) − K1 e−rT · N (d2 )

.
= 8 · N (0.3) − 8 · e−0.04 · N (−0.1) = 1.407 euros.

It follows that the probability we are looking for equals

P C0 − (ST − K1 )+ + x (ST − S0 ) ≤ 0 ST ≥ K1

C +K1 −xS0
P ST ≥K1 ;ST ≥ 0 1−x
= P (ST ≥K1 )

C +K1 −xS0
P ST ≥ 0 1−x
= P (ST ≥K1 )

. C +K1 −xS0
P WT ≥ σ1 ln 0(1−x)S − μ− 12 σ 2 T
= 0
K

P WT ≥ σ1 ln S 1 − μ− 12 σ 2 T
0

1 C0 +K1 −xS0
1−N σ ln (1−x)S0 − μ− 12 σ 2
=
K

1−N σ1 ln S 1 − μ− 12 σ 2
0
= 0.419.
2. For a Call option as in item 1.(a) but with maturity T ∗ = 2 years (instead of
T = 1) the Delta would be equal to
⎛ 2
⎞
ln (S0 /K1 ) + r + σ2 T ∗
(T ∗ =2)
Δ1
. =N⎝ √ ⎠
σ T∗
⎛ ⎞
ln (1) + 0.04 + 12 (0.4)2 · 2
=N⎝ √ ⎠ = 0.664.
0.4 2
It is then easy to check that for Delta-neutrality we need to buy 0.664 shares
of the underlying.
3. Remember that:
• the Gamma of the underlying (ΓS ) is zero;
• a portfolio is said to be Gamma-neutral if the Gamma of the portfolio is zero.
Since the Gamma of the underlying is zero, to make the initial portfolio
Gamma-neutral we should buy or sell other options.
(a) We start computing Δ2 and Γ2 (Delta and Gamma of the new option) as well
as Γ1 (Gamma of the Call option we are selling).
Proceeding as in item 1., we get

⎛ ⎞
ln (S0 /K2 ) + r + 12 σ 2 T
.Δ2 = N (d1 ) = N ⎝ √ ⎠
σ T
⎛ ⎞
ln (8/12) + 0.04 + 12 (0.4)2
=N⎝ ⎠ = N (−0.71) = 0.238.
0.4
To compute Gamma, we recall that for European Call or Put options in

the Black-Scholes model we have
N (d1 ) 1 d12
ΓCall = ΓP ut =
. √ =√ √ e− 2 .
S0 σ T 2π S0 σ T
Hence
1 (0.3)2
Γ1 = = √
. e−
= 0.12 2
2π · 8 · 0.4
1 (−0.71)2
Γ2 = √ e− 2 = 0.097
2π · 8 · 0.4
(b) First of all, let us make our portfolio Gamma-neutral. We will subsequently
make it Delta-neutral by buying or selling a suitable number of shares of the
underlying. Notice that it is necessary to respect this order. Indeed, making
Gamma-neutral a portfolio that is already Delta-neutral might generate a
portfolio that, in the end, is no more Delta-neutral.
The idea is then to find a number y of options of the second kind to
be bought/sold, that makes the portfolio Gamma-neutral. Namely, we are
looking for y solving
. − 1 · Γ1 + y · Γ2 = 0, (7.25)
where the sign “−” is due to the short position in the Call option. The
equation above becomes then
. − 0.12 + y · 0.097 = 0.
Hence y = 1.237. In order to make the portfolio Gamma-neutral we have to

buy 1.237 options with strike K2 .
Since Γ1 , Γ2 > 0 and y has to solve (7.25), it is evident that to make our
portfolio Gamma-neutral it is necessary to take a long position in the second
Call.
In order to make the portfolio Delta-neutral as well, we need to find a
number x of shares of the underlying to be bought or sold that makes the
new stocks-options portfolio Delta-neutral, i.e. such that
. − 1 · Δ1 + 1.237 · Δ2 + x = 0
−0.618 + 1.237 · 0.238 + x = 0.
It follows that x = 0.324. To summarize: in order to make our portfolio

Delta- and Gamma-neutral we have to buy 1.237 Call options with strike K2
and 0.324 shares of the underlying.
(c) Our aim is to make our portfolio Gamma-neutral and with Delta smaller
or equal (in absolute value) to 1. We have therefore to look for a number
y of Call options of the second kind and for a number x of shares of the
underlying satisfying the following conditions

−1 · Γ1 + y · Γ2 = 0
.
−1 ≤ −1 · Δ1 + y · Δ2 + x ≤ 1
Consequently,

y = 1.237
.
−0.676 ≤ x ≤ 1.324
As we would expect, the number of shares of the underlying to be bought to

make the portfolio Delta-neutral (equal to x = 0.618, see item 1.(a)) belongs
to the interval above. Furthermore, the number of shares that we are allowed
to buy so to respect the constraints above is 1 (and no more).
4. In the following we will make our portfolio Delta-, Gamma- and Vega-neutral.
Remember that:
• the Vega of the underlying is zero;
• a portfolio is said to be Vega-neutral if the Vega of the portfolio is null.
First of all, let us make our portfolio Gamma- and Vega-neutral. Only
afterwards we will make it Delta-neutral by buying or selling a suitable number
of shares of the underlying.
The idea is then to find a number y of options of the second kind and z of the
third kind to be bought or sold to make the portfolio Gamma- and Vega-neutral.
Let us compute Δ, Γ and ν of the three options available:

ln(S0 /K3 )+ r+ 12 σ 2 T
Δ3 = N (d1 ) = N √
σ T

ln(8/6)+ 0.04+ 21 (0.4)2
. =N 0.4 = N (1.02) = 0.846
(1.02) 2
Γ3 = √ 1
e− 2 = 0.07
2π ·8·0.4
It is well known that the Vega of European Call or Put options in the Black-
Scholes model is given by
√
νCall = νP ut = S0 T N (d1 ) = S02 σ Γ T .
.
Hence, in our setting we get
ν1 = 82 · 0.4 · Γ1 = 3.072
.
ν2 = 82 · 0.4 · Γ2 = 2.483
ν3 = 82 · 0.4 · Γ3 = 1.792
In order to make our portfolio Gamma-neutral and Vega-neutral y and z have

to solve the following system of linear equations

−Γ1,Call + 2Γ1,P ut + y · Γ2 + z · Γ3 = 0
.
−ν1,Call + 2ν1,P ut + y · ν2 + z · ν3 = 0

Γ1 + y · Γ 2 + z · Γ 3 = 0
ν1 + y · ν2 + z · ν3 = 0

y = −1.24
,
z = −0.000069
because Γ1,Call = Γ1,P ut = Γ1 and ν1,Call = ν1,P ut = ν1 .

Finally, to make such a portfolio Delta-neutral we have to find a number x of
shares of the underlying to be bought or sold so that the new portfolio of stocks-
options becomes Delta-neutral. This means that x, y and z have to solve
. − Δ1,Call + 2Δ1,P ut + y · Δ2 + z · Δ3 + x = 0.
It follows that x = 1.68.

To conclude: in order to make our position Delta-, Gamma- and Vega-neutral

we have to: sell 1.24 options with strike K2 ; sell 0.000069 options with strike
K3 ; and buy 1.68 shares of the underlying.
Exercise 7.7 There are three European call options traded on the market. They
are written on the same underlying, with initial value S(0) = 30 euros, and the
dynamics of its price is described by a geometric Brownian motion with volatility
σ = 0.6. The risk-free interest rate is r = 0.06. The options have the same strike
K = 25 and maturity T = 1 year. Two of them are call options, one of them is a put
option. Define a portfolio of strap type composed by the European options above.
Compute the value at time t = 0 and the Delta, the Vega and the Gamma of the
strap.
By using the same European options above (two calls and one put, written on the
same underlying following the dynamics specified before, with the same maturity
and strike) and one unit of the underlying asset, compose a new portfolio which is
both Δ and Γ -neutral.
Solution Let’s compute the values of the three options included in the portfolio.
C1 = C2 = 30N(d1 ) − 25e−0.06×1 N(d2 ) = 30N(0.7038)

.
−25 × 0.9418 × N(0.1038) = 23.094 − 13.065 = 10.09euros,
P1 = C1 − S(0) + Ke−0.06×1 = 10.09 − 30 + 25 × 0.9418 = 3.635euros,

.
where we computed the put option value by using the Put-Call parity.
The value at time t = 0 of the strap is then the following:
2C1 + P1 = 10.09 + 3.635 = 13.725euros.

.
Let’s compute now the Delta, the Gamma and the Vega of the options considered.
ΔC = N(d1 ) = 0.7698
.
N (d1 ) 0.3114
ΓC =
. √ = = 0.0173
σ S(0) T 18
√
νC = S(0) ×
. T N (d1 ) = 30 × 0.3114 = 9.342
ΔP = N(d1 ) − 1 = −0.2302
.
ΓP = ΓC = 0.0173
.
νP = νC = 9.342.
.
By linearity, the Δ, Γ, ν of the strap π are then given by:
Δπ = 2ΔC + ΔP = 1.309
.
. Γπ = 2ΓC + ΓP = 0.0519
νπ = 2νC + νP = 28.026.
.
Now, we forget about the strap and compose a new portfolio, which is both Δ and
Γ -neutral, by using the same options introduced before and one unit (short position)
of the underlying asset. The conditions that must be satisfied are then:
2αΔC + βΔP − 1 = 0
.
2αΓC + βΓP = 0.
By remembering that ΓC = ΓP , and that ΔP = ΔC − 1, we get β = −2α,

α = 1/2.
Exercise 7.8 Compute the value of a European call option written on the exchange
rate X between between euros and Swedish Crowns. Suppose euro is the domestic
currency and Swedish Crown is the foreign currency, and the domestic and foreign
risk-free rates are rd = 0.08 and rf = 0.04 respectively. Assume moreover that the
exchange rate dynamics is described by a geometric Brownian motion with diffusion
coefficient σX = 0.8 and the initial value X(0) = 9.8. The option strike is K = 9
and the maturity T = 1 year.
Solution We simply need to apply the Black-Scholes formula for a European option
written on an exchange rate:
C(X(t), t) = xe−rf (T −t) N(d1 ) − Ke−rd (T −t) N(d2 ),

. (7.26)
ln (x/K) + (rd − rf + σX2 /2)

d1 =
. √ . (7.27)
σ T −t
ln (x/K) + (rd − rf − σX2 /2)
d2 = √ , (7.28)
σ T −t
and rd , rf are the risk-free interest rates for the domestic and the foreign currency
respectively. In the present case we have:
ln (9.8/9) + (0.08 − 0.04 + 0.64/2)

d1 =
. = 0.5565. (7.29)
0.8
ln (9.8/9) + (0.08 − 0.04 − 0.64/2)
d2 = = −0.2435, (7.30)
0.8
C(X(0), 0) = 9.8e−0.04 N(d1 ) − 9e−0.08 N(d2 ) = 3.3414euros.

. (7.31)
Exercise 7.9 Suppose we want to hedge a European Call option written on an

underlying stock with dynamics described by a geometric Brownian motion with
null drift by means of the following strategy (known as stop-loss strategy): whenever
the stock price is greater than the strike K we buy one unit of the stock, and we sell
it whenever the stock price becomes smaller than the strike. By applying the Tanaka
formula verify that this strategy cannot replicate the European Call option payoff.
Solution In order to replicate the Call option, we should determine an adapted
process (ht )t∈[0,T ] such that:
T
. (ST − K)+ = (S0 − K)+ + ht dSt . (7.32)
0
The stop-loss strategy can be expressed as follows:
ht 1(K,+∞) (St ).
.
By applying the Tanaka formula to the Call option payoff, as in Chap. 5, we get:
T T
1
. (ST − K)+ = (S0 − K)+ + 1(K,+∞) (St )dSt + σ 2 St2 δK (St )dt.
0 2 0
The last term in this expression does not vanish, unless the stock never assumes
the value K (in which case the strategy becomes trivial) and, although the process
(ht )t∈[0,T ] is adapted to the natural filtration generated by the Brownian motion,
it does not allow to write the option payoff in the desired form (7.32). We can
conclude that the stop-loss strategy cannot be a replicating strategy for a European
Call option.
Exercise 7.10 Consider a stock paying a dividend of 2 euros in 3 months and of

3 euros in 4 months. The stock price follows a geometric Brownian motion with
current price of 36 euros and volatility σ = 0.28 (per year).
1. Compute the discounted value of all dividends paid when the risk-free interest
rate is r = 6% per year.
2. Find the initial price of a Call option with strike of 32 euros and maturity of
1 year, written on the stock above. Using the Put-Call Parity, deduce the initial
price of the corresponding Put option.
3. Discuss whether the probability that the Call of item 1. is exercised is greater,
smaller than or equal to the probability to exercise the Call in item 2. written on
the same stock but without dividends.
4. Establish if it is possible to choose suitable drift, volatility and current price of

the stock (without dividends and distributed as a log-normal) so that the price of
the Call option with strike of 32 euros, maturity of 1 year and written on such a
stock can equal the Call price found in item 1.
5. Discuss what would happen to the prices of the Call and Put of item 1. if the
dividends were paid in 4 and 8 months and the maturity of the options were
unchanged.
6. What if the amount of dividends stayed unchanged while their dates of payment
were swapped?
Exercise 7.11 A spread is a strategy consisting in investing in two or more options
of the same kind. Whereas a butterfly spread is used when a small change in the
underlying price is expected, a bull spread is considered when an increase in the
underlying price is expected.
1. Build a butterfly spread based on the following three Call options with the same
maturity and with strikes and prices as below:
Call A: strike of 20 euros; price of 4 euros;
Call B: strike of 24 euros; price of 2 euros;
Call C: strike of 28 euros; price of 1 euro.
2. Build a bull spread using only two of the above options. [Hint: you should buy
the Call with lower strike (between the two options chosen) and sell the Call of
greater strike (same two).]
3. Suppose now that at maturity of the options nothing of what is expected actually
happens and the stock is, instead, ST = 16 euros.
(a) Compare the Profit and Loss due to the butterfly spread (respectively, bull
spread) with the profit of a strategy consisting in a long position in a Put
option with strike of 20 euros and price of 2 euros.
(b) Now compare with the Profit and Loss due to a bear spread built with the
options of item 1.
4. Supposethat the price
(at T = 1 year) of the underlying is given by ST =
S0 exp μ − 2 σ T + σ WT , with μ = 0.1 (per year), σ = 0.4 (per year)
1 2
and that the risk-free interest rate available on the market is r = 0.02 (per year).
(c) Compute the probability to lose at most one euro with the bear, with the bull
and with the butterfly spread, respectively.
(d) Compute the probability that the profit due to the bear spread is not smaller
than the profit of the butterfly, and the probability that the loss due to the bear
spread is not greater than the loss of the butterfly.
Is there any relationship between the probabilities just computed?
5. Say if one can obtain a more satisfactory result by means of long/short positions
in the stock and in the options.
Exercise 7.12 We take a short position in a European Call option with strike K1 =
8 euros and maturity T = 1 year, written on a stock “Smart-and-Fast” whose price
is distributed as a log-normal with S0 = 8 euros, σ = 40% per year. We also take a
long position in two European Put options with strike K2 = 10 euros and maturity
of 1 year, written on the same stock.
The risk-free interest rate available on the market is r = 4% per year.
1. After having computed the Delta of the Call option (denoted by Δ1 ) and of the
Put option (Δ2 ), find the number of shares of the stock “Smart-and-Fast” to be
bought/sold to make Delta-neutral a portfolio PORT1 consisting in four short
positions in the Call option and in a long position in the Put option.
2. Suppose that a Call option with strike K3 = 12 euros written on the same stock
as before is also available on the market.
(a) Compute Delta and Gamma (Δ3 and Γ3 ) of such a new option.
(b) Explain how to make Delta- and Gamma-neutral a short position in the first
Call option.
(c) Discuss what would change in items (a) and (b) if we had a Put option with
strike K3 (instead of the Call option with strike K3 ).
3. Establish whether it is possible to make the portfolio (PORT1) Delta-, Gamma-
and Vega-neutral when a third Put option with strike K4 = 6 euro and written on
stock “Smart-and-Fast” is also available. If yes, explain how.
(d) Would this be still possible if we were forced to buy/sell at least one option
with strike K3 and at least one option with strike K4 ?
(e) And if we had to buy/sell at most one option with strike K3 and at most one
option with strike K4 ?
Chapter 8
American Options
An American Option is a contract giving the buyer the right to buy (Call) or sell
(Put) a financial underlying asset for a strike price K at every instant between
the agreement date and the maturity. The main difference between American and
European options consists thus in the early exercise feature.
If the underlying does not pay any dividend, it is possible to prove that the early
exercise of an American Call option is never optimal. The American Call option
value—when written on dividend-free underlying—is then the same as that of the
European Call option with the same parameters. The same conclusion does not hold
for an American Call option on an underlying paying dividends, nor for an American
Put option. This means that Put-Call parity cannot be formulated in the same way
as for European options.
Since American options provide the holder with “at least” the same rights of
their European counterpart (actually some more), their value cannot be less than the
corresponding European options value. The initial values of both kind of options are
therefore related by the following inequalities:
C0Am ≥ C0Eur
.
P0Am ≥ P0Eur .
During the options’ lifetime it may happen that early exercise becomes more
remunerative than exercise at maturity, and the option holder will exploit this
opportunity, which is available for American options. The exercise time will be
then chosen so to optimize the income. A problem of this type is well known in
Stochastic Analysis as Optimal Stopping problem.
170 8 American Options
The American options valuation problem in a Black-Scholes setting can be

formulated as a free boundary problem for the Black-Scholes PDE, and this
formulation can be proved to be equivalent to the optimal stopping problem for
the stochastic process describing the underlying dynamics. The American option
fair value (i.e. the option price consistent with the no-arbitrage requirement) can
be equivalently characterized as the free boundary problem solution for the Black-
Scholes PDE, or as the supremum over all possible risk-neutral expectations of the
option payoff with respect to all exercise times. Both formulations do not enjoy
a closed-form solution and the methods to solve them require some numerical
techniques that give approximate solutions. A systematic presentation of these
techniques is beyond the purpose of this textbook.
We invite the reader interested in a deeper understanding of both the analytical
and the stochastic aspects of free boundary problems, and in particular American
options valuation problems, to consult some classical textbooks on the subject, like
Wilmott et al. [43], and Musiela and Rutkowski [33]. We are going to propose a
number of exercises on American options valuation in the simple context where
the detailed knowledge of more complex techniques is not required, and we mainly
focus on the binomial model setting. We will show how to evaluate an American
option by assigning to each node of the binomial tree an option value that takes the
early exercise feature into account, and by proceeding with a backward recursion in
time starting from maturity. It is useful to remember that a geometric Brownian
motion with √ diffusion coefficient .σ can√be approximated by a binomial process
with .u = eσ Δt and .d = 1/u = e−σ Δt , where .Δt is the time step size. This
approximation can always be used whenever an exact valuation formula is not
available and the quality of this approximation improves as the time step is smaller.
Finally we shall present a few approximate methods proposed for American options
valuation.
Exercise 8.1 A Put option (with strike of 8 euros) is written on an underlying stock
with current price of 8 euros and without dividends. At each ensuing two semesters
the stock price can move up by 40% with probability p = 0.8 or down by 60% with
probability 1 − p = 0.2. Denote by (St )t=0,1,2 the stochastic process representing
the underlying stock price, where t = 1 stands for one semester and t = 2 for two
semesters.
The risk-free interest rate available on the market is 4% (per year).

1.
(a) Compute the discounted expectation of the random variable S2 at the present
date.
Is the probability measure P obtained by the previous data an equivalent
martingale measure?
Is the stochastic process (St )t=0,1,2 describing the underlying stock dynam-
ics a martingale with respect to P ?
(b) By letting S̃t denote the underlying value at time t discounted at the present
date, find a probability measure Q with respect to which S̃t is a
t=0,1,2
martingale.
2. What is the fair initial value of a European Call option written on the underlying
considered before with maturity T = 1 year? And what is the fair value of the
Put option on the same underlying and with the same parameters?
3. Establish if it is optimal to exercise the American Call on the same underlying
and with the same parameters of the options of the previous item before maturity.
What would then be its fair value?
4. Verify if there exist arbitrage opportunities in the market where only the (risky)
stock and the (risk-free) bond are traded if, for each one of the next two
semesters, the stock price can rise by 4% with probability 80% or fall by 4%
with probability of 20% and the risk-free interest rate is 15% (per year).
5. In the market model of item 4., establish if it is possible to evaluate both the
European and the American options as in items 2. and in 3..
Solution From the data we see that we are working in a binomial model with factors
u = 1.4 and d = 0.4 in each time period (in the present case one semester). The
dynamics of the stock price can then be described as follows:
S2uu = 15.68
p
S1u = 11.2
p=0.8 1−p
S0 = 8 S2ud = 4.48
1−p=0.2 p
.
S1d = 3.2
1−p
S2dd = 1.28
− − −− − − − − − − −− − − − − − − −−
t =0 t = 1 (semester) T = 2 (semesters)
1.
(a) By the binomial tree just described, we immediately get:

.P S2 = S2uu = p2 = 0.64

P S2 = S2ud = 2p (1 − p) = 0.32

P S2 = S2dd = (1 − p)2 = 0.04
and
EP [S2 ] = 15.68 · 0.64 + 4.48 · 0.32 + 1.28 · 0.04 = 11.52.

.
Since EP [S2 ] = 11.52 = S0 , the process (St )t=0,1,2 is not a martingale with
respect to the probability measure P .
Let us check if the stochastic process S̃t , describing the discounted
t=0,1,2
(at the present date) underlying value, is a martingale with respect to P , i.e.
if P is a martingale measure.
We immediately verify that P is not a martingale measure because:
E [S ] 11.52
P 2
EP S̃2 =
. = = 11.076 = 8 = S0 = S̃0 .
1+r 1.04
(b) In order to determine the martingale measure Q, it is necessary to check if

there exists a probability measure Q such that
⎧ EQ [S2 ]
⎪ 1+r
⎪ = S0
⎪
⎪
⎪ EQ [S1 ]
⎨ (1+r) 1/2 = S0
. EQ [ S2 |S1 =S1u ]
⎪
⎪ = S1u
⎪
⎪ (1+r)1/2
⎪
⎩ EQ S2 |S1 =S1 = S d
d
(1+r) 1/2 1
or, equivalently, if there exists qu ∈ [0, 1] (the “new” growth probability

of the underlying for each time period) solving the following system of
equations:
⎧ uu 2
⎪
⎪ S · qu + S2ud · 2qu (1 − qu ) + S2dd · (1 − qu )2 = S0 (1 + r)
⎨ 2u
S1 · qu + S1d · (1 − qu ) = S0 (1 + r)1/2
. .
⎪ S2uu · qu + S2ud · (1 − qu ) = S1u (1 + r)1/2
⎪
⎩ ud
S2 · qu + S2dd · (1 − qu ) = S1d (1 + r)1/2
Since S2uu = S1u · u, S2ud = S1u · d = S1d · u, S2dd = S1d · d, S1u = S0 · u and
S1d = S0 · d, the system can be reduced to the following form:

S0 u2 qu2 + S0 ud · 2qu (1 − qu ) + S0 d 2 (1 − qu )2 = S0 (1 + r)
.
S0 uqu + S0 d (1 − qu ) = S0 (1 + r)1/2

u2 qu2 + ud · 2qu (1 − qu ) + d 2 (1 − qu )2 = 1 + r
uqu + d (1 − qu ) = (1 + r)1/2
Since the first equation is essentially the second one squared, we immediately
recognize that the solution of the previous system is the following:
√
(1 + r)1/2 − d 1.04 − 0.4
.qu = = = 0.62.
u−d 1.4 − 0.4
Moreover, we conclude that the martingale measure (required in order to

compute the options prices) corresponds to the following rise/fall probabili-
ties for each time step:
qu = 0.62
.
1 − qu = 0.38.
2. We remark that the European Call option payoff is 7.68, 0 and 0 when ST =
15.68, ST = 4.48 and ST = 1.28, respectively. The initial value of the European
Call is then:
1 2
C0Eur =
. qu · 7.68 + 2qu (1 − qu ) · 0 + (1 − qu )2 · 0 = 2.84 euros.
1+r
By the Put-Call parity we immediately get the European Put price:
K 8
P0Eur = C0Eur − S0 +
. = 2.84 − 8 + = 2.53 euros.
1+r 1.04
3. We know that for an American Call option, written on an underlying without

dividends, early exercise is never optimal; so the initial value of such an option
coincides with that of a European Call option on the same underlying and with
the same parameters:
C0Am = C0Eur = 2.84 euros.

.
4. In the present case, the framework for valuation is a binomial model with factors
u = 1.04 and d = 0.96 for each time step. The stock price dynamics is then the
following:
S2uu = 8.6528
p
S1u = 8.32
p=0.8 1−p
S0 = 8 S2ud = 7.9872
1−p=0.2 p
.
S1d = 7.68
1−p
S2dd = 7.3728
− − −− − − − − − − −− − − − − − − −−
t =0 t = 1 (semester) T = 2 (semesters)
Since the risk-free interest rate for each semester is equivalent to the yearly rate
1/2
ryear of 15%, we obtain rsem = 1 + ryear − 1 = 0.072, and 1 + rsem =
1.072 > u = 1.04. So in the market model considered, where a stock and a
bond are traded, there exist arbitrage opportunities. We are going to present one
of them:
t =0 t =2=T
short-sell the stock give back the stock

⇒ +S0 ⇒ −ST
.
invest S0 at the risk-free rate r

⇒ −S0 S0 (1 + rsem )2
S0 − S0 = 0 S0 (1 + rsem )2 − ST > S0 u2 − ST ≥ 0
Moreover,

.P S0 u2 − ST > 0 = P {S2 = S0 ud} ∪ S2 = S0 d 2
= 2p (1 − p) + (1 − p)2 > 0.
This shows that the strategy illustrated is an arbitrage opportunity.

5. By the previous item and from the First Fundamental Theorem of Asset Pricing
we can immediately conclude that an equivalent martingale measure for the
market model considered does not exist. This implies that the options previously
mentioned cannot be evaluated as in items 2. and 3.
The non-existence of an equivalent martingale measure can be proved also
directly. If such a measure (let us denote it by Q) existed, the following condition
should be satisfied:
EQ [S1 ]
. = S0 ,
1 + rsem
i.e. there should exist qu ∈ [0, 1] such that
S0 uqu + S0 d (1 − qu ) = S0 (1 + rsem ) .
.
Since the last equation is equivalent to
uqu + d (1 − qu ) = 1 + rsem
.
1 + rsem − d u−d
qu = > = 1,
u−d u−d
it follows immediately that an equivalent martingale measure does not exist.

Exercise 8.2 An option is written on a stock without dividends and with current
price of 10 euros. At the end of each one of next two semesters the stock price can
rise by 25% or fall by 20%, the risk-free interest rate is 4% (per year) and the strike
of the option is 11 euros.
1. Find the current price of a European Put option written on this underlying and
with maturity T = 1 year.
2. Is it optimal to exercise the corresponding (i.e. with the same parameters)
American option before maturity? What is its fair value?
3. Assume now that the underlying stock pays a dividend of 1.5 euros in 6 months.
(a) Compute the fair value of both the American and European Put options with
strike of 11 euros.
(b) Find the fair value of the American Call option with the same parameters of
the previous one.
4. Can the difference between the initial values of the American Put and the
American Call (both with strike K = 11 euros) be more than (or equal to) 0.5
and less than (or equal to) 3 euros?
Solution The market model described above corresponds to a binomial model with
factors u = 1.25 and d = 0.8 for each time step (in the present case one semester).
The stock price dynamics is then the following:
uu = 15.625
S1y

u = 12.5
S6m

S0 = 10 ud = 10
S1y

.
d
S6m =8

dd = 6.4
S1y
− − −− − − − − − − −− − − − − − − −−
t =0 t = 6 months T = 1 year
The risk-neutral probability measure, necessary to compute the options’ value,

can be easily obtained and is defined by the following rise/fall probabilities for each
time step:
√
(1 + r)1/2 − d 1.04 − 0.8
.qu = = = 0.488
u−d 1.25 − 0.8
1 − qu = 0.512
1. Since the European Put option payoff is 0, 1 and 4.6 (when ST = 15.625, ST =
10 and ST = 6.4, respectively), by the previous remark we can conclude that the
initial fair value of the European Put option is given by:
1 2
P0Eur =
. qu · 0 + 2qu (1 − qu ) · 1 + (1 − qu )2 · 4.6 = 1.64 euros.
1+r
2. We want to verify now if the early exercise of the American Put is optimal or not.
To check this we are going to proceed backwards along the tree, starting from
maturity and considering all subtrees corresponding to one time-step separately.
We shall denote by a capital letter each node of the tree:
15.625 (D)

12.5 (B)

10 (A) 10 (E)

.
8 (C)

6.4 (F)
− − −− − − − − −− − − − − −−
Moreover, for each node we write the payoff of the American Put considered:
15.625 (D)
payoff = 0

12.5 (B)
payoff = 0

10 (A) 10 (E)
.
payoff = 1 payoff = 1

8 (C)
payoff = 3

6.4 (F)
payoff = 4.6
Let us start by considering the subtree including nodes B, D and E. The American
Put, when restricted to this subtree is “European-like”. So we can calculate the
corresponding (i.e. with the same parameters) European Put value at the nodes
B, D and E. To avoid confusion between the European Put of item 1. (we mean
the European Put written at t = 0 and maturing in 1 year) and the “artificial” put
options just introduced, we shall call the latter “European”(*).
At the nodes D and E the Put price coincides with its payoff, so
∗
,D = payoff at node D = 0
PTEur
.
∗
,E = payoff at node E = 1
PTEur
∗ 1
Eur
P6m,B = [qu · 0 + (1 − qu ) · 1] = 0.50.
(1 + r)1/2
Since the payoff of the American Put in B is 0, therefore less than the European*
Eur ∗ = 0.50), early exercise at
Put price in t = 6 months with maturity 1 year (P6m,B
node B is not optimal.
The American Put value at each node is the maximum between the payoff and
the value of the corresponding European* Put. In other words, it coincides with
the European* counterpart when the early exercise is not optimal, while it equals
the payoff at the node considered when the early exercise is optimal.
At the nodes B, D and E we have then:
,D = PT ,D = payoff in D = 0
PTAm Eur
.
,E = PT ,E = payoff in E = 1
PTAm Eur
Am
P6m,B = max{P6m,B
Eur
; payoff in B} = max{0.50; 0} = 0.50.
Let us now consider the subtree with nodes C, E and F. The American Put option,
restricted to this subtree is again “European-like”. We can then compute the
corresponding European* Put value at nodes C, E and F.
At nodes E and F, the option value coincides with the payoff. Hence,
∗
,E = payoff in E = 1
PTEur
.
∗
,F = payoff in F = 4.6
PTEur
∗ 1
Eur
P6m,C = [qu · 1 + (1 − qu ) · 4.6] = 2.79.
(1 + r)1/2
Since the American Put payoff at node C is equal to 3 (hence greater than the
Eur ∗ =
European* Put value at t = 6 months and with maturity T = 1 year (P6m,C
2.79)), the early exercise at node C is optimal.
The American Put price at nodes C, E, F is then given by:
∗
,E = PT ,E = payoff at node E = 1
PTAm Eur
.
∗
,F = PT ,F = payoff at node F = 4.6
PTAm Eur
∗
Am
P6m,C = max{P6m,C
Eur
; payoff at node C} = max{2.79; 3} = 3.
Let us finally consider the subtree corresponding to nodes A, B and C. The

American Put option restricted to this subtree is again “European-like”. We can
compute the corresponding European* Put value at nodes A, B and C.
In B and C the value coincides with that of the American Put computed before,
so
∗
Eur
P6m,B
. = P6m,B
Am
= 0.50
∗
Eur
P6m,C Am
< P6m,C =3
∗ 1
Eur
P0,A = [qu · 0.50 + (1 − qu ) · 3] = 1.75.
(1 + r)1/2
Since the payoff of the American Put at node A is 1, so less than the value of the
European* Put in t = 0 with maturity t = 6 months (P0,A Eur ∗ = 1.75), the early
exercise in A is not optimal.

Eventually, we can conclude that the American Put value in A is given by:
∗
Am
P0,A
. = max{P0,A
Eur
; payoff at node A} = max{1.75; 1} = 1.75 euros.
We remark that, as one could expect, the initial value of the American Put is
greater than that of the corresponding European (as computed in 1. and equal to
1.64 euros).
3. If the underlying stock pays a dividend of 1.5 euros in 6 months, its dynamics
can be described as follows:
S1Div,uu Div,u
year = S6m · u = 13.75

Div,u
S6m = S0 u − 1.5 = 11
Div,ud Div,u
S1y = S6m · d = 8.8
S0Div = 10
Div,du Div,d
S1y = S6m · u = 8.125
.
Div,d
S6m = S0 d − 1.5 = 6.5

Div,dd Div,d
S1y = S6m · d = 5.2
− − −− − − − − − − − − − − −− − − − − − − − − − − −−
(a) We describe, with the help of the following diagram, the American Put and
the European Put payoffs, both with strike K = 11 euros:
13.75 (node D)
payoffAm = payoffEur = 0

11 (node B)
payoffAm = 0
8.8 (node E)

payoffAm = payoffEur = 2.2
10 (node A)
.
payoffAm = 1
8.125 (node F)

6.5 (node C)
payoffAm = 4.5

5.2 (node G)
Since
(1.04)1/2 − d
qu =
. = 0.488,
u−d
we have that
1
Eur
P6m,B
. = √ [0 + (1 − qu ) · 2.2] = 1.105
1.04
1
Eur
P6m,C = √ [qu · 2.875 + (1 − qu ) · 5.8] = 4.288.
1.04
Consequently,
Am
P6m,B
. = max{P6m,B
Eur
; payoffAm; node B } = max{1.105; 0} = 1.105
Am
P6m,C = max{P6m,C
Eur
; payoffAm; node C } = max{4.288; 4.5} = 4.5.
Finally we obtain:
1
P0Eur = √
. [qu · 1.105 + (1 − qu ) · 4.288] = 2.682 euro
1.04

1
P0Am = max √ [qu · 1.105 + (1 − qu ) · 4.5] ; payoffAm; node A
1.04
= max{2.788; 1} = 2.788 euros.
(b) If the American option considered in the previous point was a Call instead
of a Put, the procedure would change as follows:
13.75 (node D)
payoffCall Am = 2.75

11 (node B)
payoffCall Am = 0
8.8 (node E)

payoffCall Am = 0
10 (node A)
.
payoffCall Am = 0
8.125 (node F)

payoffCall Am = 0
6.5 (node C)
payoffCall Am = 0

5.2 (node G)
payoffCall Am = 0
By proceeding in the same way outlined before, we have that:
1
Eur
C6m,B
. = √ [qu · 2.75 + 0] = 1.316
1.04
Eur
C6m,C =0
and, consequently,
Am
C6m,B
. = max{C6m,B
Eur
; payoffCall Am; node B }
= max{1.316; 0} = 1.316
Am
C6m,C = max{C6m,C
Eur
; payoffCall Am; node C } = 0.
Finally, we obtain

1
.C0Am = max √ [qu · 1.316 + 0] ; payoffAm; node A
1.04
= max{0.629; 0} = 0.629.
4. From the previous items, we immediately get
P0Am − C0Am = 2.788 − 0.629 = 2.159 ∈ [0.5; 3] .

.
Consequently, the difference between the initial American Put and Call values
(both with strike K = 11 euros) is included in the interval [a, b] where a = 0.5
and b = 3 euros (endpoints included).
We could obtain the same conclusion without having to compute the two options
prices directly. We know, indeed, that
K
. − S0 < P0Am − C0Am < D + K − S0 ,
1+r
where D is the discounted value of the dividends distributed between the initial
date 0 and the maturity T . In the present case we have D = √1.5 = 1.471 and
1.04
K
0.577 =
. − S0 < P0Am − C0Am < D + K − S0 = 2.471.
1+r
Exercise 8.3 Consider the American Put option of item 2. in Exercise 8.2. This
option expires in 1 year, its strike is K = 11 euros and it is written on an underlying
stock without dividends with current price of 10 euros. During each one of next two
semesters the stock price can rise by 25% or fall by 20%. The risk-free interest rate
is 4% (per year).
Find the hedging strategy of the American option considered.
Solution From Exercise 8.2, we recall that the dynamics of the underlying stock
and the dynamics of the American Put can be described as follows:
uu = 15.625 (node D)
S1y
valuePut Am = 0

u = 12.5 (node B)
S6m
valuePut Am = 0.50

ud = 10 (node E)
S0 = 10 (node A) S1y
.
valuePut Am = 1.75 valuePut Am = 1

d = 8 (node C)
S6m
valuePut Am = 3

dd = 6.4 (node F)
S1y
valuePut Am = 4.6
and that at node C early exercise is optimal.

In order to compute the hedging strategy, it is necessary to determine the strategy
at time t = 0 and at node B, but not in C since at this node early exercise is viable.
To find the hedging strategy at time t = 0 means to find a quantity Δ0 of the
underlying stock and an amount of money x0 such that

Δ0 S0 u + x0 (1 + r)1/2 = P6m,B
Am
.
Am ,
Δ0 S0 d + x0 (1 + r)1/2 = P6m,C
i.e.
⎧ Am −P Am
⎨Δ = P6m,B 6m,C
= −0.556
.
0 S0 (u−d) .
⎩ x0 = 1
P6m,B − Δ0 S0 u
Am = 7.31
(1+r) 1/2
This means that the strategy consists in selling 0.556 stock units and investing 7.3
euros. As one could expect, its initial cost is given by:
Δ0 S0 + x0 = −0.556 · 10 + 7.31 = 1.75 = P0Am .

.

The replicating strategy ΔB6m , x6m to be performed at t = 6 months in the node
B
B must solve the following equations:

6m S6m u + x6m (1 + r) = P6Am
u 1/2
ΔB B
months, D .
.
Δ6m S6m d + x6m (1 + r) = P6m,E
B u B 1/2 Am
Then
⎧ Am −P Am
⎨ ΔB = P6m,D 6m,E
u (u−d) = −0.1778
.
6m S6m .
⎩ xB = 1
P Am − ΔB S u u = 2.724
6m (1+r)1/2 6m,D 6m 6m
As expected, then, at node B the cost of the strategy just outlined is given by:
6m S6m + x6m = Δ6m S0 u + x6m = 0.50 = P6m,B .

ΔB u B B B Am
.
Exercise 8.4 Compute the price at time t = 0 of a perpetual American Put option
with strike K and written on an underlying whose price dynamics is described by a
geometric Brownian motion with diffusion coefficient σ . The risk-free interest rate
available on the market is r.
Solution Since the option is perpetual, i.e. its maturity is T = +∞, the valuation
problem does not depend explicitly on t. By removing the t-dependence in the PDE
providing the American Put value, we obtain the following ODE:
σ 2 2 d 2P dP
. S + rS − rP = 0. (8.1)
2 dS 2 dS
This ordinary differential equation must be solved taking into account the proper
boundary conditions that, for the American Put option, turn out to be the following
P (+∞) = lim P (S) = 0,

.
S→+∞
P (Sf ) = K − Sf ,

dP
P (Sf ) = = −1.
dS S=Sf
We should remark that we need three conditions in order to determine the

solution of the ODE, since the free-boundary value Sf is unknown, i.e. we need
one more condition than for a “fixed” boundary. Actually, the free-boundary value
Sf has to be computed as part of the solution of the problem considered, and
becomes an extra unknown. The second and third conditions are called the “value
matching” condition and the “smooth pasting” condition, respectively. Together
with the condition at infinity they can be proved to be both necessary and sufficient
in order to guarantee the existence and uniqueness of the solution for the problem
under examination.
To compute the general solution of the ODE above, we can perform a simple
variable change:
.S = ex . (8.2)
Hence, x = ln S and the ODE for the new unknown P (x) becomes

σ 2 d 2P σ 2 dP
. + r − − rP = 0. (8.3)
2 dx 2 2 dx
This is a linear homogeneous ordinary differential equation, whose solution can be

immediately written as
P (x) = Aeλ1 x + Beλ2 x ,

. (8.4)
where A, B are constants determined by imposing the boundary conditions and

λ1 , λ2 (λ1 < λ2 ) are the two distinct solutions of the quadratic equation:

σ2 2 σ2
. λ + r− λ − r = 0. (8.5)
2 2
Hence,
⎡ ⎤
σ2 σ2
λ1,2 = σ −2 ⎣−(r −
. )± (r − )2 + 2rσ 2 ⎦
2 2

−2 σ2 σ2
=σ −(r − ) ± (r + ) ,
2 2
2r
.λ1 = − , λ2 = 1. (8.6)
σ2
Written in terms of the old independent variable S , the solution is then
− 2r2
P (S) = AS λ1 + BS λ2 = AS
. σ + BS. (8.7)
The first boundary condition, i.e. P (∞) = 0, implies that the constant B must
vanish, while the second and third conditions allow to determine simultaneously the
constant A and the free-boundary value Sf :
− 2r2
P (Sf ) = ASf
.
σ
= K − Sf ,
2r − 2r2 −1
P (Sf ) = − AS σ
= −1.
σ2 f
From these conditions, we immediately obtain
2r2
A = K − Sf Sfσ
.
K
Sf = .
σ2
1+ 2r
Exercise 8.5 The early exercise premium of an American option p(S, t) is defined
as p(S, t) P Am (S, t) − P Eur (S, t), i.e. as the difference at time t between the
values of an American and a European Put option on the same underlying and with
the same parameters. Assume as ansatz that the dependence of P on the time to
maturity τ = T − t can be factorized through the function H (τ ) = 1 − exp(−rτ )
as follows:
p(S, τ ) = H (τ )f (S, H (τ )),

. (8.8)
for a suitable function f .

1. Find the PDE satisfied by the function f in the Black-Scholes framework (i.e.
by assuming that the underlying price dynamics is described by a geometric
Brownian motion).
2. Assume that the nonlinear terms appearing in the PDE determined in the previous
step are much smaller than the remaining terms. Solve the ODE obtained by
neglecting those terms by taking into account the proper boundary conditions.
3. By using the solution obtained in the previous step, provide an approximate
formula for the American Put option value.
Solution We know already that the function p(S, τ ) is non-negative. Since p is the
difference between two functions satisfying the Black-Scholes equation, it must also
satisfy the same equation (due to the linearity of the Black-Scholes equation):
∂p σ 2 2 ∂ 2 p ∂p
. − + S + rS − rp = 0. (8.9)
∂τ 2 ∂S 2 ∂S
The initial condition in the variable τ for the function p must be p(S, 0) = 0, since
the American and the European options satisfy the same initial condition (they have
the same payoff).
1. By assuming that the dependence of p on τ can be factorized in the suggested
form, it is easy to compute the partial derivatives with respect to the new
independent variables:
∂p ∂f
. =H ,
∂S ∂S
∂ 2p ∂ 2f
= H ,
∂S 2 ∂S 2
∂p ∂f dH dH
=H +f ,
∂τ ∂H dτ dτ
dH
= r(1 − H ).
dτ
Hence, we get the following PDE for f :
2f
2∂ ∂f k 1 ∂f
.S + kS − f 1 + H (1 − H ) = 0, (8.10)
∂S 2 ∂S H f ∂H
where we introduced the parameter k = 2r/σ 2 .

2. By neglecting the quadratic term in H , and thus eliminating the explicit
dependence of f on H , we finally obtain the approximate ODE for f :
∂ 2f ∂f k
S2
. + kS − f = 0. (8.11)
∂S 2 ∂S H
This ODE, in which H plays the role of a parameter, can be solved explicitly.
By solving the ODE exactly as in the previous exercise, we obtain the solution:
.f (S) = AS λ , (8.12)
where λ is defined as follows:

1 4k
.λ − (k − 1) + (k − 1)2 + . (8.13)
2 H
The constant A must be determined by imposing the proper conditions.

3. Combining the results obtained, we can write an approximation for the American
Put option value for S > Sf :
P Am (S, τ ) ≈ P Eur (S, τ ) + AH (τ )S λ ,

. (8.14)
where the constant A must be such that the option value matches its payoff for
S = Sf :
P Eur (S, τ ) + AH (τ )Sfλ = K − Sf .

. (8.15)
There is still the unknown Sf to be determined, and this can be done by imposing
the “smooth pasting” condition
∂P Am ∂P Eur
. (Sf , τ ) = (Sf , τ ) + AλH Sfλ−1
∂S ∂S
= N(d1 (Sf )) − 1 + AλH Sfλ−1 = −1, (8.16)
where N (·) stands, as usual, for the cumulative distribution function of a standard
normal. Solving the last equation, we get an expression for A in terms of Sf :
N(d1 (Sf ))
A=−
. . (8.17)
λH Sfλ−1
By substituting into the Eq. (8.15), we obtain an equation for the remaining
unknown Sf :
Sf
P Eur (Sf , τ ) − N(d1 (Sf ))
. = K − Sf , (8.18)
λ
which, taking into account the Put-Call parity and the trivial identity N(−d) =
1 − N (d), can be written as
Sf
Sf N(d1 (Sf ))−Ke−rτ N(d2 (Sf ))−Sf +Ke−rτ −N(d1 (Sf ))
. −K +Sf = 0.
λ
Collecting terms, we write it as
1
Sf N(d1 (Sf ))(1 −
. ) + Ke−rτ 1 − N(d2 (Sf )) − K = 0. (8.19)
λ
Since d1 and d2 depend on Sf , this is an implicit equation in the unknown Sf
and must be solved by an approximate iterative method, like Newton’s method,
where the initial seed S0 = K may be used for starting the iterative procedure.
Exercise 8.6 F. Black [7] suggested the following approximate procedure for
evaluating American call options written on distributing dividends assets. He
proposed to evaluate both the European call maturing at time T and the European
call maturing just before the last dividend distribution date, then to approximate
the corresponding American option value by the maximum between the values of
the two European options considered. As an example, consider an American call
option written on the asset with initial value S(0) = 30 euros, maturity 1 year
(T = 1),strike K = 25, and assume that its dynamics is described by a geometric
Brownian motion with diffusion coefficient σ = 0.4, and the risk-free interest
rate r = 0.06. Suppose moreover that, during the lifetime of the option, the asset
distributes two dividends, one after 4 months and one after 8 months, of the same
amount di = 1 euro (i = 1, 2). Evaluate the American call written on the asset
considered by applying the procedure proposed by F. Black.
Solution We first have to evaluate the two European options written on the same
asset, the first one with the same maturity of the American option, and the second
one just before the last dividend distribution date, by applying the Black-Scholes
formula for options written on asset distributing dividends at discrete time:
. C1 = [S(0) − d1 e−0.333×0.06 − d2 e−0.667×0.06 ]N(d̂11 ) − Ke−0.06 N(d̂21 ) =

= 28.059N (2.8) − 25e−0.06 N(2.4) = 27.986 − 23.351 = 4.635euro.
. C2 = [S(0) − d1 e−0.333×0.06 ]N(d̂12 ) − Ke−0.667×0.06 N(d̂22 ) =

= 29.02N(3.511) − 25e−0.667×0.06 N(3.184) = 5.006euro.
Since the maximum between the two values is the second, we approximate the
American call option value with this: CA = C2 = 5.006 euro.
Remark 8.7 Notice that in this case the approximation proposed by Black is very
rough. By computing K[1 − e−r(T −t2 ) ] = Ke−0.333×0.06 = 24.5 > 1, it is
immediate to conclude that it is not optimal to exercise the option before the date
of the second dividend distribution, but it is optimal to exercise the option before
the first dividend distribution. This typically happens since in this case, when the
asset price approaches the strike value, the dividend yield of the asset is bigger than
the risk free rate. When this is the case, it is optimal to exercise the option as late
as possible. This implies that the true value of the option will be closer to that of
the corresponding European option. Intuitively speaking, the dividends are not big
enough to compensate the potential remuneration due to late exercise.
Exercise 8.8 The current price of a stock is 40 euros. At the end of this year the
stock price will be either 42 or 36 euros. The growth and fall factors of the stock
in the following years will remain the same and the risk-free interest rate is 4% per
year.
1. Compute the price of the American Put with maturity T = 2 years and with strike
of 40 euros. Is it optimal to exercise this option before maturity?
2. Suppose a first dividend of 1 euro is paid after 1 year and a second one of 3 euros
after 2 years: what changes in the valuation procedure? Which option is more
expensive: the American Put or the American Call?
3. For which American option is the early exercise more likely to take place: a Call
or a Put option?
4. Compute the price of the American Put and Call options with the same
parameters as in the previous point, but with maturity delayed by 2 years.
5. Is it possible to find a different value of the growth factor in a binomial model, so
that the value of a Call option with maturity T = 2 years and strike K = 40 euros
written on a stock without dividends coincides with the value of an American Put
option with the same parameters written on a stock with the same dividends as
item 2.?
Exercise 8.9 Compute the price (at time t = 0) of a perpetual American Call
option, written on an underlying paying dividends at a constant rate q, and whose
price dynamics is described by a geometric Brownian motion with diffusion
coefficient σ . The risk-free interest rate is r, the strike K.
Exercise 8.10 By applying the approximation procedure proposed by F. Black,
compute the price (at time t = 0) of an American Call option, written on an
underlying asset with initial value S(0) = 50 euros, maturity T = 16 months,
paying two dividends during the option lifetime, one after 6 months, one after 1 year,
both of the same amount di = 1.5 (i = 1, 2) euros. Assume the asset price dynamics
is described by a geometric Brownian motion with diffusion coefficient σ = 0.2.
The risk-free interest rate is r = 0.04, the strike K40 euros.
Chapter 9
Exotic Options
An Exotic option is an option that is neither a European Call or Put option nor an
American Call or Put. A path-dependent option is an option whose payoff does not
depend only on the underlying value at maturity, but also on one or more values that
it can assume during its lifetime. Although the two option classes do not coincide,
many exotic options exhibit path-dependence features. Among the most popular
Exotic options that are not path-dependent we just recall binary (or digital) options,
compound options and chooser options.
Binary Options The most traded kind of binary option is the so called “cash or
nothing” Call. This is a contract guaranteeing the owner a fixed amount of money
B if the underlying value at maturity exceeds a threshold L, and nothing if the
underlying value is less than L.
The payoff of this option can be expressed in the following form:
B1{ST ≥L} = B · H (ST − L),

.
where the function H is the so-called Heaviside step-function:

0, x < 0
H (x) =
. . (9.1)
1, x ≥ 0
Another kind of binary option is the “asset or nothing” Call. This option expires
out-of-the-money if the underlying value at maturity is less than the threshold L,
while it guarantees a unit of the underlying asset if the threshold value is exceeded.
The expression of this second kind of binary option payoff is given by:
ST 1{ST ≥L} = ST · H (ST − L).

. (9.2)
192 9 Exotic Options
“Chooser” Options are options with other options as underlying assets, giving
the owner the right to decide, at maturity .T1 < T2 , whether the underlying option
must be a Call or a Put. The chosen option will be (possibly) exercised at maturity
.T2 and its strike will be .K2 .
The valuation of these options can be performed in a way that is strictly similar
to the European options, but with the following final condition:

.F (ST2 , T2 ) = max C(ST2 , T2 ) − K2 ; P (ST2 , T2 ) − K2 ; 0 .
As far as path-dependent options are concerned, we are going to propose a few

exercises on Asian options, Barrier options and Lookback options.
Asian Options are options depending on the average of the values assumed by
the underlying during the option lifetime.
Asian options can be of different kinds according to the average considered:
this can be the arithmetic mean or the geometric mean, and the underlying values
contributing to the average can be “observed” in continuous or discrete time.
A further distinction among Asian options arises according to the functional
dependence of the payoff on the average considered; this can in fact substitute the
underlying final value or the strike in the payoff involved. In the former case we
shall refer to average rate options, in the latter to average strike options.
The valuation problem for these kind of options can be cast into the Black-
Scholes framework and, for some particular cases, closed-form solutions providing
options prices exist. Except for the very simple case of the geometric Asian options
(with continuous time monitoring), these solutions do not admit a representation
that is easy to handle for applications. We shall therefore focus our attention mainly
on examples of valuation problems in a binomial setting.
Barrier Options are options with the same payoff at maturity of the European
options, but with one extra feature: they expire valueless (or not) if, at any instant
during the option lifetime, the underlying reaches some threshold value from above
or from below. In the first case they will be called up-and-out options, in the second
down-and-out (or up-and-in and down-and-in respectively).
For Barrier options (with continuous monitoring), explicit valuation formulas are
available in a Black-Scholes setting. We recall these formulas below.
Down-and-Out Call Pricing Formula
2r −1
L σ2 L2
CDO (St , t) = C(St , t) −
. C ,t , (9.3)
St St
where .C(St , t) is the value of the European Call option written on the same
underlying and with the same parameters (strike, maturity, volatility and risk-free
interest rate) of the Barrier option, and L is the barrier value. The previous formula
holds when .K > L. When .K < L, the down-and-out Call pricing formula reduces
to the following:
CDO (St , t) = St [N(d3 ) − b(1 − N(d6 ))]

.
−Ke−r(T −t) [N(d4 ) − a(1 − N(d5 ))] , (9.4)
where .a, .b, .d3 , .d4 , .d5 , .d8 are defined as follows:
2r 2r +1
L σ 2 −1 L σ2
.a = , b= ,. (9.5)
St St
2

ln SLt + r + σ2 (T − t)
d3 = √ ,. (9.6)
σ T −t
2
ln( SLt ) + (r − σ2 )(T − t)
d4 = √ ,. (9.7)
σ T −t
2

ln SLt − r − σ2 (T − t)
d5 = √ ,. (9.8)
σ T −t
2

ln SLt − r + σ2 (T − t)
d6 = √ . (9.9)
σ T −t
Up-and-Out Call Pricing Formula

For up-and-out Call, in case .U > K, the expression providing the price at time t
is the following:
CU O (St , t) = C(St , t) − St [N(d3 ) + b(N(d6 ) − N(d8 ))]

.
+Ke−r(T −t) [N(d4 ) + a(N(d5 ) − N(d7 ))] . (9.10)

= St [N(d1 ) − N(d3 ) − b (N(d6 ) − N(d8 ))]
−Ke−r(T −t) [N(d2 ) − N(d4 ) − a(N(d5 ) − N(d7 )] , (9.11)
where
2

St K
ln − r − σ2 (T − t)
U2
.d7 = √ ,. (9.12)
σ T −t
2

ln SUt K2 − r + σ2 (T − t)
d8 = √ (9.13)
σ T −t
and the quantities .d3 , .d4 , .d5 , .d6 are defined as before, with U instead of L. It is easy
to observe that the value of an up-and-out Call option with .U < K is zero.
For Barrier options of “in” type it is possible to obtain immediately the pricing
formulas from those just presented, by observing that a portfolio composed by two
Barrier options of in and out type respectively, written on the same underlying, with
the same parameters and with the same threshold values, is equivalent to a European
option, still written on the same underlying and with the same parameters. As an
example, we provide the following pricing formula for a down-and-in Call option,
holding if .K > L:
2r/σ 2 −1
L L2
CDI (St , t) =
. C ,t . (9.14)
St St
Lookback Options are options with payoff assuming one of the following forms:
FP M (ST , Smax , T ) = max {Smax − ST ; 0} = max {M − ST ; 0}

. (9.15)
for the Put,
FCm (ST , Smin , T ) = max {ST − Smin ; 0} = max {ST − m; 0}

. (9.16)
for the Call, where .m = Smin , .M = Smax are, respectively, the minimum and the
maximum values assumed by the underlying during the option lifetime. We remark
that M and m “substitute the strike” in the payoff of the European Put and Call
options respectively.
Explicit analytical expressions are available for the Lookback options just
mentioned in the Black-Scholes setting; they can be obtained both by solving the
relative PDE with similarity methods or by computing the discounted expectation
of their payoff, since the joint probability distributions of the running maximum and
minimum of a Brownian motion and of the Brownian motion itself is known. We
briefly recall these pricing formulas for the reader’s convenience:
FCm (St , m, t) = St N(d1 ) − me−r(T −t) N(d2 )

2 √
+St e−r(T −t) σ2r · ( Smt )−2r/σ N(−d1 +
2 2r
σ T − t)
−er(T −t) N(−d1 ) ,
.
FP M (St , M, t) = Me−r(T −t) N(−d2 ) − St N(−d1 )

2 √
+St e−r(T −t) σ2r · −( M
St −2r/σ 2
) N(d1 − 2r
σ T − t)
+er(T −t) N(d1 ) .
It is useful to remember that a geometric Brownian motion with diffusion √

.u = e
coefficient .σ can be approximated by a binomial process with σ Δt and
√
.d = 1/u = e
−σ Δt , where .Δt is the time step size. This approximation can always
be used whenever an exact valuation formula is not available and the quality of
this approximation improves as the time step is smaller. It is worth mentioning a
few issues related to Exotic options hedging. While some of these options are easy
to hedge—like Asian options (since the average becomes more and more stable as
time goes on, and the number of observations grows), other options—like barrier
options—can be extremely difficult to hedge: a small variation of the underlying
price, close to the barrier value, can give rise to a huge variation of the option
value, i.e. the Delta can be discontinuous across the barrier. In these cases, a static
replication approach can be used. The static replication methodology is based on
a standard result of functional analysis that can be roughly resumed as follows:
every continuous real function with compact support can be approximated (with the
desired accuracy) by piecewise linear functions; this means, in practice, that every
payoff (with compact support) can be approximated by the payoff of a portfolio
consisting of a suitable number of units of the underlying and a suitable number of
Put and/or Call European options. We shall illustrate this procedure by an example
in the exercises.
The reader interested in a more detailed presentation of pricing methods for
Exotic options can find many results in El Karoui [18], Hull [25], Musiela and
Rutkowski [33] and Wilmott et al. [43].
Exercise 9.1 Consider a stock with value at time t denoted by (St )t≥0 and with
current value S0 = 8 euros. Compute the price at time t = 0 of a “cash or nothing”
Call option, written on the stock considered, with maturity T = 3 months, strike
L = 10 euros and rebate B = 20 euros. The stock price dynamics is assumed to be
described by a geometric Brownian motion with volatility σ = 0.25 (per year). The
risk-free interest rate on the market is r = 0.04 (per year).
We recall that, for a “cash or nothing” Call the strike is the threshold value L such
that, if the underlying at maturity is greater than or equal to L, the option buyer gets
the cash amount B (the rebate).
Solution In order to compute the value at time t = 0 of a binary option of “cash or
nothing” Call type, we can apply the usual risk-neutral valuation procedure. Hence,
we need to determine the discounted expectation of the option payoff with respect
to the equivalent martingale measure:
CD (St , t) = e−r(T −t) EQ [Φ(ST )] ,

. (9.17)
where we denote by CD the binary option’s value at time t, by Φ its payoff and by
EQ the expectation with respect to Q (the risk-neutral measure). We recall that the
payoff of a binary option of “cash or nothing” Call type is given by the following
expression: Φ(ST ) = B · H (ST − L), where L is the “threshold” value and B the
rebate.
In the present case, since we are working in a Black-Scholes framework, we can
write:
+∞
CD (St , t) = e−r(T −t)
. B · H St e r−σ 2 /2 (T −t)+σ x
−L
−∞
1 x2
×√ e− 2(T −t) dx
2π(T − t)
+∞ B · H (St ez − L)
= e−r(T −t) √
−∞ σ 2π(T − t)
2
z − (r − σ 2 /2)(T − t)
× exp − dz.
2σ 2 (T − t)
The integral appearing in the last formula can be explicitly computed:

2
+∞B · H (St ez − L) z − (r − σ 2 /2)(T − t)
. √ exp − dz
−∞ σ 2π(T − t) 2σ 2 (T − t)
2
+∞ z − (r − σ 2 /2)(T − t) dz
=B exp − √
ln L−ln St 2σ (T − t)
2 σ 2π(T − t)
+∞ 1 u2
=B ln L−ln St − r−σ 2 /2 (T −t)
( ) √ e− 2 du
√
σ T −t
2π

ln (L/St ) − r − σ 2 /2 (T − t)
= B 1−N √
σ T −t
= B · N(d2 ).
In the last expression we have denoted by N the cumulative distribution function of

a standard normal and set
ln(St /L) + (r − σ 2 /2)(T − t)

d2 =
. √
σ T −t
(in strict analogy with the term appearing in the Black-Scholes formula for the
European Call option price).
The “cash or nothing” Call option value is then given by
CD (St , t) = Be−r(T −t) N(d2 ).

.
In our case, we have t = 0, T = 1/4, S0 = 8, B = 20, L = 10, r = 0.04,

σ 2 = 0.0625, so
ln(8/10) + (0.04 − 0.0625/2) 14

d2 =
. √ = −1.767
0.25 · 1/4
and
1
CD (0, 8) = 20 · e−0.04· 4 N(−1.767) = 0.764 euros.
.
Exercise 9.2 Consider a chooser option with maturity T2 and strike K, i.e. an
option written at time t = 0 with maturity T2 for which at time T1 (with T1 < T2 )
the option buyer can choose if the option will be exercised as a Call or a Put on the
same underlying asset.
Compute the initial price of the chooser option with maturity T1 = 3 months,
T2 = 6 months and strike K = 30 euros, if the underlying dynamics is described by
a geometric Brownian motion with initial value S0 = 20 euros, volatility σ = 0.25
(per year), and the risk-free interest rate is r = 0.06 (per year).
Solution A chooser option can be easily valuated by showing that it is equivalent to
a portfolio composed by a European Call with maturity T1 and strike Ke−r(T2 −T1 )
and by a European Put with maturity T2 and strike K (see Hull [25] for a detailed
proof). We denote by C0 Ke−r(T2 −T1 ) ; T1 and P0 (K; T2 ) the prices at t = 0 of
these options.
The initial price Ch0 of the chooser option is then given by:

Ch0 = P0 (K; T2 ) + C0 Ke−r(T2 −T1 ) ; T1 ,
. (9.18)
i.e. the initial prices of the Put and Call options mentioned before. By substituting
the data assigned, we get:
1
Ch0 = P0 (30, 1/2) + C0 (30e−0.06 4 ; 1/4).
.
By applying the Black-Scholes formulas for European options directly, one obtains:
1
. C0 (30e−0.06 4 ; 1/4) = 0.038
P0 (30; 1/2) = 9.172,
by which we can compute the initial value of the chooser option:
Ch0 = 0.038 + 9.172 = 9.21 euros.

.
Exercise 9.3 Consider a stock with initial value S0 = 8 euros, with dynamics
described by a geometric Brownian motion with volatility σ = 0.36 (per year)
in a market model with risk-free interest rate r = 0.04 (per year).
1. Compute the value at time t = 0 of an up-and-in and of an up-and-out Call option
with barrier U = 12 euros, strike K = 9 euros and maturity T = 4 months.
2. What is the (risk-neutral) probability that, at t = 2 months, the underlying value
will reach the barrier?
Solution
1. First of all, we recall the valuation formula for a Barrier Call option of up-and-out
type. By denoting by CU O the value of this Barrier option and by C0 the value
the corresponding vanilla option (both at time t = 0), from (9.10) we get:
CU O,0 = C0 − S0 [N(d3 ) + b(N(d6 ) − N(d8 ))] .

. (9.19)
+Ke−rT [N(d4 ) + a(N(d5 ) − N(d7 ))] (9.20)
where a, b, d3 , d4 , d5 , d6 , d7 and d8 are defined by (9.5)–(9.13).

By the formulas just mentioned, with the assigned data, we have that:
0.08 0.08 +1
12 (0.36)2 −1 12 (0.36)2
a=
. = 0.856; b = = 1.927
8 8
2

ln 128
+ 0.04 + (0.36)
2
4
12
d3 = = −1.783
0.36 · 124
2

ln 128
+ 0.04 − (0.36)
2
4
12
d4 = = −1.991
0.36 · 124
2

ln 128
− 0.04 − (0.36)
2
4
12
d5 = = −1.911
0.36 · 124
2

ln 128
− 0.04 + (0.36)
2
4
12
d6 = = −2.119
0.36 · 124

(0.36)2
ln − 0.04 −
8·9 4
(12)2 2 12
d7 = = −3.295
0.36 · 12
4

(0.36)2
ln 8·92 − 0.04 + 2
4
12
(12)
d8 = = −3.50
0.36 · 12
4
Since
2
ln(S0 /K) + (r + σ2 )T
d1 =
. √ = −0.399
σ T
√
d2 = d1 − σ T = −0.606,
we conclude that the initial (at time t = 0) value of the European Call option is
C0 = S0 · N(d1 ) − Ke−rT · N(d2 )

. = 8 · N (−0.399) − 9 · e−0.04·4/12 · N (−0.606)
= 0.342 euros.
Finally, we obtain the initial value of the up-and-out Call:
CU O,0 = 0.20 euros.

.
By the relation holding between up-and-out, up-and-in and vanilla option

values (strictly analogous to that between down-and-out, down-and-in and
vanilla), we
have that:
C0 = CU I,0 + CU O,0 ,
. (9.21)
where CU I,0 denotes the initial value of the up-and-in Call.

From (9.21) we finally obtain:
.CU I,0 = C0 − CU O,0 = 0.342 − 0.20 = 0.142 euros.

We remark that both the up-and-in Call and the up-and-out Call values are
smaller that the value of the corresponding European Call. That is because, in
general, there is a non-vanishing probability that the barrier will be reached. In
general, the following relations hold:
C0 ≥ CDI,0
.
C0 ≥ CDO,0 .
2. We now want to compute the probability that, at t = 2 months, the underlying

value will reach the barrier, i.e. we want to compute P (St ≥ U ) at t = 2/12
(2 months).
Since the risk-neutral dynamics of the underlying’s value is described by a
geometric Brownian motion,

St = S0 exp
. r − σ 2 /2 t + σ Wt
and the probability that its value at t = 2/12 will be greater than U is the
following:
P (St ≥ U ) = P (ln (St ) ≥ ln (U ))

.

σ2
= P ln (S0 ) + r − t + σ Wt ≥ ln (U )
2
⎛ 2
⎞
W ln (U ) − ln (S0 ) − r − σ2 t
= P ⎝√ ≥ ⎠
t
√
t σ t
⎛ ⎞
σ2
ln (U ) − ln (S0 ) − μ − 2 t
= 1−N⎝ √ ⎠.
σ t
With the assigned data we obtain:

⎛ ⎞
(0.36)2
ln (12) − ln (8) − 0.04 − 2
2
12
P (S2 months
. ≥ U) = 1 − N ⎝ √ ⎠
0.36 2/12
= 1 − N (2.480) = 0.0066.
Exercise 9.4 In a binomial setting, consider a Lookback Call option with maturity
T = 8 months, written on an underlying stock (without dividends) whose price
can rise or fall by 50% during each 2-months period of the next 8 months. The
underlying’s initial value is S0 = 10 euros, the risk-free interest rate is 8% (per
year).
Compute the initial value of the Lookback option considered. What is the
probability (risk-neutral) that its payoff will be strictly positive?
Solution The data provide u = 1.5 and d = 0.5 as parameters of the binomial
model and S0 = 10 as initial stock price. In the present setting, the underlying’s
dynamics can be described as follows:
50.625

33.75

22.5 16.875

15 11.25

S0 = 10 7.5 5.625

. 5 3.75

2.5 1.875

1.25

0.625
− − −− − − −− − − −− − − −− − − −−
t =0 t =1 t =2 t =3 T =4
(2 months) (4 months) (6 months) (8 months)
In order to simplify the notation, we shall denote as follows the nodes of the tree
just described:
M

G

D N

B H

. A E O

C I

F P

L

Q
We recall that the payoff of a Lookback Call option is (ST − Smin ) , where Smin
denotes the minimum value assumed by the underlying during the option lifetime.
The initial value of this option (written on an underlying with a binomial dynamics)
is then given by:
1
FCm (0) =
. EQ [ST − Smin ] , (9.22)
(1 + r)T
where Q is the risk-neutral probability measure.

Moreover, since Smin depends on the “trajectory” of the underlying value process,
we must examine the contribution of each single path followed by the underlying
along the binomial tree.
Before considering all possible final nodes, let us compute the risk-neutral
probability measure with respect to which the expectation must be calculated. As
usual, in the present binomial setting, qu is given by:
(1 + r)2/12 − d (1.08)2/12 − 0.5

qu =
. = = 0.513
u−d 1.5 − 0.5
and qd = 1 − qu = 0.487.
Now we must consider all possible paths arriving at the final nodes M, N, O, P
and Q.
final node: M If the underlying value at maturity was 50.625 (node M), then the
only possible path would be that described by a growth of the underlying value
at each time step. In this case we should have

Smin = 10
. ⇒ ST − Smin = 40.625.
ST = 50.625
The probability that this occurs is qu4 = 0.0693.

final node: Q If the underlying value at maturity was 0.625 (node Q), then the
only possible path would be that described by a fall of the underlying value at
each time step. In this case we should have

Smin = 0.625
. ⇒ ST − Smin = 0.
ST = 0.625
The probability that this occurs is qd4 = (1 − qu )4 = 0.0562.

final node: N If the underlying value at maturity was 16.875 (node N), then there
would be 4 possible paths to consider.
First Path The price dynamics is the following:
G

D N
.
B

A
and the corresponding probability is qu3 (1 − qu ) = 0.0654. In this case we have:

Smin = 10
. ⇒ ST − Smin = 6.875.
ST = 16.875
Second Path The price dynamics is the following:
D N

. B H

A
with probability qu3 (1 − qu ) = 0.0654. In this case we have:

Smin = 10
. ⇒ ST − Smin = 6.875.
ST = 16.875
Third Path The price dynamics is the following:
N

. B H

A E
and the probability of this path is qu3 (1 − qu ) = 0.0654. In this case we have:

Smin = 7.5
. ⇒ ST − Smin = 9.375.
ST = 16.875
Fourth Path The price dynamics is the following:
N

H
.
A E

C
and the probability is qu3 (1 − qu ) = 0.0654. In this case we have:

Smin = 5
. ⇒ ST − Smin = 11.875.
ST = 16.875
final node: O If the underlying value at maturity was 5.625 (node O), there would
be 6 possible paths to consider.
D

. B H

A O
and the probability is qu2 (1 − qu )2 = 0.0560. In this case we have:

Smin = 5.625
. ⇒ ST − Smin = 0.
ST = 5.625
B H
.
A E O

Smin = 5.625
. ⇒ ST − Smin = 0.
ST = 5.625
A E O
.
C I

Smin = 3.75
. ⇒ ST − Smin = 1.875.
ST = 5.625
A O

. C I

F

Smin = 2.5
. ⇒ ST − Smin = 3.125.
ST = 5.625
Fifth Path The price dynamics is the following:
B

. A E O

I

Smin = 3.75
. ⇒ ST − Smin = 1.875.
ST = 5.625
Sixth Path The price dynamics is the following:
H

. A E O

C
and the probability that such occurs is qu2 (1 − qu )2 = 0.0560. In this case we
have:

Smin = 5
. ⇒ ST − Smin = 0.625.
ST = 5.625
final node: P If the underlying value at maturity was 1.875 (node P), there would
be 4 possible paths to consider.
B

A E
.
I

P
and the probability is qu (1 − qu )3 = 0.0593. In this case we have:

Smin = 1.875
. ⇒ ST − Smin = 0.
ST = 1.875
A

C
.
F P

L

Smin = 1.25
. ⇒ ST − Smin = 0.625.
ST = 1.875
A

. C I

F P

Smin = 1.875
. ⇒ ST − Smin = 0.
ST = 1.875
A E

. C I

P

Smin = 1.875
. ⇒ ST − Smin = 0.
ST = 1.875
By summing up the contributions of all cases, we conclude that the Lookback

Call option value at time t = 0 is given by:
1
FCm (0) =
. [qu4 · 40.625
(1 + r)8/12
+qu3 (1 − qu ) (6.875 + 6.875 + 9.375 + 11.875)
+qu2 (1 − qu )2 (0 + 0 + 1.875 + 3.125 + 1.875 + 0.625)
+qu (1 − qu )3 (0 + 0.625 + 0 + 0) + (1 − qu )4 · 0]
= 5.561 euros.
Moreover, we can easily find that the (risk-neutral) probability that the payoff of
the Lookback Call will be strictly positive is given by:
Q ({payoff > 0}) = qu4 + 4qu3 (1 − qu ) + 4qu2 (1 − qu )2 + qu (1 − qu )3 = 0.6142.

.
Exercise 9.5 With the same data of the previous exercise and again in a binomial
setting, compute the value at time t = 0 of an average strike Call and the (risk-
neutral) probability that its payoff will be strictly positive.
Solution An “average strike” option is an Asian option with payoff at maturity
explicitly depending on the average of the values assumed by the underlying asset
during the option lifetime, according to the following expression:
FAS (ST , Smed , T ) = (ST − Smed )+ = max {ST − Smed ; 0} .

.
We shall assume that the average considered is the arithmetic mean, which in our
binomial setting with 4 time steps is:
4
t=1 St
Smed =
. . (9.23)
4
By recalling the scheme provided in the previous exercise we have to consider
the following possible paths.
final node: M If the underlying value at maturity was 50.625 (node M), then the
only possible path for the underlying price would be that of a growth at each time
step. In this case we have:

Smed = 30.469
. ⇒ (ST − Smed )+ = 20.156.
ST = 50.625
The probability that this occurs is qu4 = 0.06935.

final node: Q If the underlying value at maturity was 0.625 (node Q), then the
only possible path for the underlying price would be that of a fall at each time
step. In this case we have:

Smed = 2.344
. ⇒ (ST − Smed )+ = 0.
ST = 0.625
The probability that this occurs is qd4 = (1 − qu )4 = 0.0562.

final node: N If the underlying value at maturity was 16.875 (node N), there
would be 4 possible paths for the underlying price.
G

D N
.
B

A
and the probability of this is qu3 (1 − qu ) = 0.0654. In this case we have:

Smed = 22.031
. ⇒ (ST − Smed )+ = 0.
ST = 16.875
D N

. B H

A

Smed = 16.406
. ⇒ (ST − Smed )+ = 0.469.
ST = 16.875
N

. B H

A E

Smed = 12.656
. ⇒ (ST − Smed )+ = 4.219.
ST = 16.875
N

H
.
A E

C

Smed = 10.156
. ⇒ (ST − Smed )+ = 6.719.
ST = 16.875
final node: O If the underlying value at maturity was 5.625 (node O), then there
would be 6 possible paths for the underlying price.
D

. B H

A O

Smed = 13.594
. ⇒ (ST − Smed )+ = 0.
ST = 5.625
B H
.
A E O

Smed = 9.844
. ⇒ (ST − Smed )+ = 0.
ST = 5.625
A E O
.
C I

Smed = 5.469
. ⇒ (ST − Smed )+ = 0.156.
ST = 5.625
A O

. C I

F

Smed = 4.219
. ⇒ (ST − Smed )+ = 1.406.
ST = 5.625
Fifth Path The price dynamics is the following:
B

. A E O

I

Smed = 7.969
. ⇒ (ST − Smed )+ = 0.
ST = 5.625
Sixth Path The price dynamics is the following:
H

. A E O

C

Smed = 7.344
. ⇒ (ST − Smed )+ = 0.
ST = 5.625
final node: P If the underlying value at maturity was 1.875 (node P), there would
be 4 possible paths for the underlying price.
B

A E
.
I

P

Smed = 7.031
. ⇒ (ST − Smed )+ = 0.
ST = 1.875
A

C
.
F P

L

Smed = 2.656
. ⇒ (ST − Smed )+ = 0.
ST = 1.875
A

. C I

F P

Smed = 3.281
. ⇒ (ST − Smed )+ = 0.
ST = 1.875
A E

. C I

P

Smed = 4.531
. ⇒ (ST − Smed )+ = 0.
ST = 1.875
The initial value of the option considered is then given by:
FAS (ST ; Smed ; 0)

= (1.08)1
8/12 [0.0693 · 20.156 + 0.0654 · (0.469 + 4.219 + 6.719)
.
+0.0560 · (0.156 + 1.406) + 0.0593 · 0 + 0.0562 · 0]
= 2.26.
Moreover, we can easily find that the (risk-neutral) probability that the payoff of
the average strike Call will be strictly positive is given by:
Q ({payoff > 0}) = qu4 + 3qu3 (1 − qu ) + 2qu2 (1 − qu )2

.
= 0.0693 + 3 · 0.0654 + 2 · 0.0560 = 0.3775.
Exercise 9.6 Compute the value at time t = 0 of a geometric Asian option, of the
average-rate kind, written on an underlying stock whose price dynamics is described
by a geometric Brownian motion with diffusion coefficient σ . The risk-free interest
rate is r, the strike K.
Solution We first remark that the payoff of a geometric Asian option of average-
rate type can be written as follows:
CAR (T , ŜT ) = max{ŜT − K; 0} = max{S0 exp(X̄T ) − K; 0},

. (9.24)
where ŜT denotes the geometric average of the values assumed by the underlying
during the time interval [0, T ], while X̄T denotes the arithmetic mean of the values
assumed by the log-returns during the same time interval. The arithmetic average
of the log-returns can be easily derived when the underlying price dynamics is
described by a geometric Brownian motion. If the underlying price evolves as
follows:
dSt = rSt dt + σ St dWt ,

.
i.e. St = S0 exp(Xt ) where
dXt = (r − σ 2 /2)dt + σ dWt ,

.
then, by integrating by parts, we can write immediately:

t t
. Xu du = tXt − udX(u). (9.25)
0 0
By inserting the dynamics of X and dividing by t, (9.25) becomes:

σ2 t u t u
.X̄t = r− 1− du + 1− σ dWu .
2 0 t 0 t
The last equation implies that X̄T is distributed as a normal random variable with
mean:
T
σ2 u σ2 T
.E[X̄T ] = r− 1− du = r − , (9.26)
2 0 T 2 2
and variance:
T u 2 2 σ2
V (X̄T ) =
. 1− σ du = T. (9.27)
0 T 3
Now, since the value at time t = 0 of an average rate option is just the discounted
value of its expected payoff (computed with respect to the risk-neutral measure):

CAR (0, S0 ) = e−rT E max{S0 exp(X̄T ) − K; 0} ,
. (9.28)
we can obtain this value by applying the Black-Scholes formula for European Call
options, simply by comparing the payoffs and substituting the new values of the
mean and the variance in the arguments. From the previous results, by remembering
that Ŝt = eX̄t , we can write the stochastic differential equation satisfied by Ŝt :
σ2 1 σ2 σ
. d Ŝt = [(r − ) + ]Ŝt dt + √ Ŝt dWt . (9.29)
2 2 6 3
In order to simplify the computation, we can consider (just as a computational trick)

the extra term subtracted by r in the previous equation like a constant continuous
dividend rate:
σ2 1 σ2
(r −
. ) + = r − q, (9.30)
2 2 6
where:
r σ2 σ2 r σ2 1 σ2
.q= + − = + = [r + ] (9.31)
2 2 6 2 12 2 6
A direct substitution provides the following formula for the (geometric) average rate
Call option:

2
− r+ σ6 T2 T
CAR (0, S0 ) = S0 e
. N d∗ + σ − Ke−rT N(d∗ ), (9.32)
3
where d∗ is defined as follows:

σ2
ln( SK0 ) + r − 2
T
2
.d∗ . (9.33)
σ T3
Exercise 9.7 Consider the following down-and-out Call option with maturity T =
1 year, strike K = 30 euros and barrier threshold L = 25 euros. The underlying
price dynamics is described by a geometric Brownian motion with volatility σ =
0.40 (per year). The current price of the underlying is S0 = 30 euros, while the
risk-free interest rate is r = 0.05 (per year).
Construct a portfolio consisting of European options (approximately and stati-
cally) replicating the Barrier option considered.
Solution There is a large amount of arbitrariness in constructing a portfolio
replicating the Barrier option. We start the process with a simple remark: if two
portfolios have the same value on the contour of a region in the (t, S)-space, then
they must have the same value inside the region. The idea is then to approximate
the Barrier option on the contour of such region by means of European options with
suitable maturities and strikes.
In the present case, since the barrier option is of down-and-out type, the domain
is the strip delimited by vertical half-line t = 0 on the left, the interval t ∈ [0, 1] on
the horizontal line S = 25 below, and by the vertical half-line t = T = 1 on the
right, and this is the contour on which we want to replicate the barrier option. The
down-and-out Call takes the following values on these lines:
C(S, 1) = max{S − 30; 0}

. (for S > 25) (9.34)
and
.C(25, t) = 0 (for 0 ≤ t ≤ 1). (9.35)
The simplest way to replicate the option on the right contour (t = 1) is to buy a
European Call option with strike KA = 30 and maturity T = 1. We denote by A
such an option and by CA its value.
To replicate the option on lower contour we can divide the time interval [0, 1]
on the line S = 25 in three sub-intervals of 4 months (i.e. 1/3 of a year) each, and
choose three different European options satisfying condition 9.35 for t = 0, 13 , 23 ,
respectively.
So, a second option (denoted by B) to be considered for replication on the lower

contour can be a Put with strike KB = 25 and maturity TB = 1. Please, note that
the value of this option vanishes on the previous contour considered, as it has to do,
while a Call option with the same strike and maturity does not. By the Black-Scholes
formula for European options (see (7.5)), the values of options A and B at t = 2/3
are CA (2/3, 25, 30, 1) = 0.86 and PB (2/3, 25, 25, 1) = 2.07 respectively, where
the first and the second arguments denote the valuation point coordinates (t, S) and
the third and the fourth denote the strike and the maturity (K, T ).
In order to satisfy the condition (9.35) for t = 1/3 we need another option
(vanishing at all the points previously considered): this third option (denoted by
C) can be a Put option with maturity TC = 2/3 and strike KC = 25 and its value at
t = 1/3, S = 25 is PC (1/3, 25, 25, 2/3) = 2.07.
Finally, we need to satisfy the condition (9.35) at time t = 0. To this end we
choose a fourth European option of type Call (denoted by D) with maturity TD =
1/3 and strike KD = 25 and its value at t = 0, S = 25 is CD (0, 25, 25, 1/3) =
2.48.
The number of options of each kind to be held in the replicating portfolio is
determined by the condition that the replicating portfolio assumes value zero at
each of the points considered on the contour. So, by assuming that one unit of the
option A is held in the replicating portfolio, the number β of options B to buy (or
sell) is obtained by imposing that at point (t = 2/3, S = 25)
CA (2/3, 25, 30, 1) + βPB (2/3, 25, 25, 1) = 0.

.
Since the values of options A and B at (t = 2/3, S = 25) are 0.86 and 2.07,
respectively, this condition provides for β the value β = −0.42.
As far as the number of options C is concerned, we can repeat the same argument
and impose the condition at (t = 1/3, S = 25):
CA (1/3, 25, 30, 1) + βPB (1/3, 25, 25, 1) + γ PC (1/3, 25, 25, 2/3) = 0.
.
Since A, B, and C at (S = 25, t = 1/3) are valued 1.83, 2.79 and 2.07, respectively,
this gives γ = −0.32.
Finally, we can determine the number of options of type D by imposing the
analogous condition at (t = 0, S = 25):
CA (0, 25, 30, 1) + βPB (0, 25, 25, 1) + γ PC (0, 25, 25, 2/3)
.
+ δCD (0, 25, 25, 1/3) = 0.
Since the options’ values are 2.70, 3.29, 2.79 and 2.48 for A, B, C, and D,
respectively, the value of δ = −0.35 follows.
The replicating portfolio has then the following composition:
(1, −0.42, −0.32, −0.35)

.
in options A, B, C, D, respectively.
Exercise 9.8 Compute the value at time t = 0 of a Chooser option written on

an underlying stock whose price dynamics is described by a geometric Brownian
motion, with strike K = 20 euros and maturities T1 = 6 months and T2 = 1 year.
The initial value of the underlying is S0 = 20 euros, the risk-free interest rate r =
10% (per year) and the volatility σ = 20% (per year).
Exercise 9.9 Compute the value at time t = 0 of a Barrier option of “up-and-in”
Call type, written on an underlying stock whose price dynamics is described by
a geometric Brownian motion, with initial value S0 = 10 euros, σ 2 = 0.016, r =
0.04, K = 6 euros and barrier value U = 8 euros. By using the relationship between
Barrier and European Call options, find the initial price of the corresponding “up-
and-out” Call.
Exercise 9.10 Compute the value at time t = 0 of a Binary option of “asset or
nothing” Call type, with the same parameters (except the strike K) of Exercise 9.1.
Exercise 9.11 With the same data of Exercises 9.4 and 9.5, and again in a binomial
setting, compute the initial price of a Lookback Call option (on the minimum) with
maturity T = 10 months. Find, moreover, the initial value of the corresponding Put
option.
setting, compute the initial price of an average strike Call option, with a payoff
at maturity expressed by (ST − SMG )+ = max {ST − SMG ; 0}, where SMG is the
geometric mean
nof the values assumed by the underlying during the option lifetime,
i.e. SMG = n Πi=1 Si .
setting, compute the price at time t = 0 of an average rate Call option, with a
payoff at maturity expressed by (Smed − K)+ = max {Smed − K; 0}, where Smed
denotes the arithmetic mean of the values assumed by the underlying during the
option lifetime and K = 5 euros.
Exercise 9.14 Determine the payoff at maturity of a portfolio composed by the
Lookback Call of Exercises 9.4 and by the corresponding Lookback Put option.
Compute the value of this portfolio at time t = 0.
Exercise 9.15 Compute the initial value of a geometric Asian option, of the
average-rate type, with maturity T = 1 year, strike K = 30 euros, written on an
underlying stock with dynamics described by a geometric Brownian motion and
with initial value S0 = 36 euros. The risk-free interest rate is r = 0.04 (per year)
and the volatility σ = 0.4 (per year).
Exercise 9.16 Compute the value at t = 0 of a (floating strike) Lookback Call
option on the minimum, with maturity T = 1 year, written on an underlying stock
with dynamics described by a geometric Brownian motion and with initial value
S0 = 36 euros. The risk-free interest rate is r = 0.04 (per year) and the volatility
σ = 0.4 (per year).
Consider the binomial model with four time steps (each of 3 months) approximat-
ing the geometric Brownian motion above. After computing the Lookback option
value at time t = 0 in this binomial setting, evaluate the binomial approximation’s
quality by comparing the two results obtained.
Exercise 9.17 Compute the initial value of a down-and-out Call option with the
same data of Exercise 9.7, except the barrier value (which is now a lower threshold),
here assumed to be L = 40 euros, and construct a static replicating portfolio
composed by 4 European options.
Do Call options give a better or worse replication than Put options?
Exercise 9.18 Compute the initial value of a financial derivative with payoff
{Smax − Smin }, i.e. with payoff represented by the difference between the maximum
and the minimum of the values assumed by S(t), t ∈ [0, T ]. Hint: the financial
derivative above can be replicated by a portfolio composed by a lookback put on the
maximum (long position) and a lookback call on the minimum (short position).
Chapter 10
Interest Rate Models
In the valuation problems presented in the previous chapters, the interest rate was
assumed to be a deterministic parameter, constant in most cases. This is of course
a crude approximation of the real situation where interest rates vary over time in
an unpredictable way. This assumption can be reasonable only when dealing with
financial contracts with short maturities (like options), while for financial derivatives
with longer lifetime (like bonds) or any other fixed-income products, it can be
quite misleading. It is necessary, then, to adopt stochastic models for interest-rate
dynamics.
Let us start by defining the relevant quantities that we wish to describe.
A zero-coupon bond (ZCB) is a contract that guarantees the buyer one unit (by
convention) of the reference currency at maturity T . Since the value of this contract
depends both on the date at which it is written and on its maturity T , we will denote
by .Z(t, T ) its value at time t. If the zero-coupon bond guarantees the buyer the cash
amount N at maturity T , we will specify that the zero-coupon bond’s face value is
N . In this case, its value at time t will be expressed by .N · Z (t, T ).
A coupon bond with face value N, maturity T and periodically paid coupons
(at the end of each semester, for example) at the annual rate .rc is a contract that
guarantees the buyer the cash amount N at time T and the coupons with value c at
the end of each period (in the case of one semester: .c = r2c · N) until maturity T .
We define instantaneous forward rate with respect to maturity T the following
quantity:
∂ ln Z(t, T )
f (t, T ) −
. (10.1)
∂T
and short rate the quantity .rt f (t, t).
222 10 Interest Rate Models
The bond valuation is then strictly related to the interest-rate dynamics. Without
any further specification, in the following we shall refer to short rates. When
dealing with the valuation problem for zero-coupon bonds the first difficulty is the
incompleteness of a short-rate stochastic model, so the no-arbitrage requirement
does not provide a unique price for a generic derivative (in our case the zero-
coupon bond). Nevertheless, no arbitrage implies some consistency condition on
the zero-coupon bonds’ values (with different maturities) which allows to reduce
the valuation problem to the final-value problem for a partial differential equation;
the coefficient of this PDE must be determined according to some “exogenous”
criterion.
If the short-rate dynamics is described by a model of the following kind:
drt = u(rt , t)dt + w(rt , t)dWt ,

. (10.2)
then the zero-coupon value .Z(t, T ) must satisfy the following PDE:
∂Z 1 ∂ 2Z ∂Z
. (t, T ) + w 2 (t, T ) 2 (t, T ) + (u − λw) (t, T ) (t, T ) − (rZ) (t, T ) = 0
∂t 2 ∂r ∂r
together with the final condition .Z(T , T ) = 1. The term .λ in the previous equation
is called the Market Price of Risk and, by the consistency relation implied by the no-
arbitrage condition mentioned above, it can be shown to be the same for all bonds
available in the market considered. The computation of the market price of risk
turns out to be equivalent to the determination of the risk-neutral measure, which is
not unique any more due to market incompleteness, chosen by the market in order
to assign prices to traded securities. Several different approaches can be followed
in order to determine the coefficient .λ: among these we just mention the general
equilibrium framework. The most popular in practice, anyway, is the one based on
a “calibration” procedure. In the following, we will assume that the coefficient .λ is
known.
An explicit solution of the bond pricing PDE for short-rate models is available
only in a few relevant cases. Among these the most relevant are those assuming the
following general form:
Z(t, T ) = exp {A(t, T ) − rt B(t, T )} ,

. (10.3)
where A and B are deterministic functions of t and T . These solutions are called
affine term structures. It is possible to prove that a sufficient condition for the zero-
coupon bond equation to admit a solution of the form (10.3) is that the coefficients
.u, w of the stochastic differential equation describing the short-rate dynamics in
(10.2) have the following form:
u(r, t) = α(t)r + β(t).

. (10.4)

w(r, t) = γ (t)r + δ(t). (10.5)
Moreover, the functions .A, B are related to the coefficients .α, β, γ , δ by the
following ordinary differential equations (of Riccati type):
∂B(t,T )
∂t + α(t)B(t, T ) − 12 γ (t)B 2 (t, T ) = −1, B(T , T ) = 0
. ∂A(t,T ) (10.6)
∂t − β(t)B(t, T ) + 12 δ(t)B 2 (t, T ) = 0, A(T , T ) = 0.
Among these models, we briefly recall the most popular (and easiest to handle)
on which we will focus in view of their applications. In particular, we shall deal with
the models proposed by Vasiček, Ho and Lee, Hull and White.
Vasiček Model
drt = a (b − rt ) dt + σ dWt ,
. (10.7)
i.e. .α(t) = −a, .β(t) = ab, .γ (t) = 0 and .δ(t) = σ 2 .

In this case, we have:
1 − e−a(T −t)
B(t, T ) =
. . (10.8)
a
[B(t, T ) − (T − t)] (a 2 b − σ 2 /2) σ 2 B 2 (t, T )
A(t, T ) = − . (10.9)
a2 4a
The explicit solution of the stochastic differential Eq. (10.7) describing the short-rate
dynamics is the following:

−at
t
rt = r0 e
. + b 1 − e−at + σ e−a(t−s) dWs . (10.10)
0
Moreover:

.E [rt ] = r0 e−at + b 1 − e−at . (10.11)
Ho-Lee Model
drt = ϑt dt + σ dWt ,
. (10.12)
i.e. .α(t) = 0, .β(t) = ϑt , .γ (t) = 0 and .δ(t) = σ 2 .

In this case, we have that:
B(t, T ) = T − t.
. (10.13)
T
σ 2 (T − t)3
A(t, T ) = ϑs (s − T )ds + . (10.14)
t 6
Hull-White Model
drt = (ϑt − art ) dt + σ dWt ,

. (10.15)
i.e. .α(t) = −a, .β(t) = ϑt , .γ (t) = 0 and .δ(t) = σ 2 .

In this case, we have that:
1 − e−a(T −t)
B(t, T ) =
. . (10.16)
a
T
1 2 2
A(t, T ) = σ B (s, T ) − ϑs B(s, T ) ds. (10.17)
t 2
By a suitable calibration procedure it is possible to obtain the functions .ϑt

appearing in the Ho-Lee and Hull-White valuation formulas:
∂f (0, t)
ϑtH L =
. + σ 2 t. (10.18)
∂t
∂f (0, t) σ2
ϑtH W = + af (0, t) + (1 − e−2at ). (10.19)
∂t 2a
In both expressions .f (0, t) denotes the instantaneous forward rate at time .t = 0.

When the above models are taken as pricing framework, the zero-coupon bonds
can be easily valuated, and so are the coupon bonds and the bond options. We shall
examine in a more detailed way these valuation problems in the exercises.
The formulas providing the value of a Call option written on a zero-coupon bond
with face value N are, for the three models presented, the following.
Price of a Call Option on a ZCB in the Vasiček Model
. C(t, T , K, T ∗ ) = N · Z(t, T ∗ ) · N(d) − K · Z(t, T ) · N(d − σp ), (10.20)
where .K, T denote the strike and the maturity of the Call option, respectively, while
T ∗ is the bond maturity, .N (·) the cumulative distribution function of a standard
.
normal random variable and the quantities .d and .σp are defined as follows:

1 N · Z(t, T ∗ ) 1
.d = ln + σp. (10.21)
σp K · Z(t, T ) 2

∗
1 − e−a(T −T ) σ 2
σp = 1 − e−2a(T −t) . (10.22)
a 2a
Price of a Call Option on a ZCB in the Ho-Lee Model
. C(t, T , K, T ∗ ) = N · Z(t, T ∗ ) · N(d) − K · Z(t, T ) · N(d − σp ), (10.23)

where

1 N · Z(t, T ∗ ) 1
.d = ln + σp. (10.24)
σp K · Z(t, T ) 2
√
σp = σ (T ∗ − T ) T . (10.25)
Price of a Call Option on a ZCB in the Hull-White Model
C(t, T , K, T ∗ ) = N · Z(t, T ∗ ) · N(d) − K · Z(t, T ) · N(d − σp ),

. (10.26)
where

1 N · Z(t, T ∗ ) 1
d=
. ln + σp. (10.27)
σp K · Z(t, T ) 2

∗
1 − e−a(T −T ) σ 2
σp = 1 − e−2a(T −t) . (10.28)
a 2a
Strictly analogous formulas exist for European Put options. These formulas can
be proved quite easily with the help of a useful technique, called the Change of
Numéraire. The latter provides a solution to several valuation problems arising in
several, different contexts, like interest-rate models, currency derivatives, and some
exotic options. We briefly recall the main ideas underlying this powerful method.
Roughly speaking, a Numéraire is the unit in which all assets in a market model
are expressed. This can be a currency unit, or a money market account, which is
t
simply given by .Bt = exp 0 ru du in market models with a deterministic short
rate. Changing numéraire means that a new unit is assumed to express financial
values, and the new numéraire does not have to be constant, nor deterministic, but
it can be described by a stochastic dynamics; the only requirement it must fulfill
is to be strictly positive for all t. The fundamental theorems of asset pricing are
usually expressed with respect to the numéraire .Bt (bond) in such a way that, in
order to avoid arbitrage opportunities, all asset price processes in the market models
considered (including derivatives), divided by .Bt , must be local martingales with
respect to the risk-neutral measure. The same results can be expressed in terms of
a different numéraire, but the risk-neutral measure for the market model will be
different as well. It can be proved that, under suitable integrability conditions, the
relationship between the two risk-neutral measures is provided by the following
formula:

dQ1 St1 S00
.Lt = = , (10.29)

dQ0 Ft St0 S01
where .Lt denotes the Radon-Nikodym derivative of the risk-neutral measure .Q1
(with .S 1 as numéraire) with respect to the risk-neutral measure .Q0 (with .S 0 as
numéraire) at time t.
A special role in interest rate models is played by the T-forward measure, that is
the risk-neutral measure adopting the value .p(t, T ) (.= Z(t, T )) of a zero-coupon
bond with maturity T as numéraire. We shall illustrate in the exercises how this
method can be used to obtain pricing formulas.
Up to now we have focused our attention on short-rate models. In market
practice, however, other relevant models play a role; among these, a very important
one is the LIBOR (London InterBank Offer Rate) market model describing the
dynamics of rates. The simple forward LIBOR rate at time t for the time interval
.[S, T ] is defined by:
p(t, S) − p(t, T )
L(t, S, T )
. , (10.30)
(T − S)p(t, T )
where .0 ≤ S ≤ T and .p(t, T ) (.= Z(t, T )) denotes the value at time t of a zero-
coupon bond with maturity T (supposed to be quoted on the market). The LIBOR
spot rate for the time interval .[S, T ] is defined by:
1 − p(S, T )
L(S, T )
. , (10.31)
(T − S)p(S, T )
and it is simply the forward rate for .t = S.

If the dynamics of the LIBOR rate for each interval .[Ti−1 , Ti ] (where .Ti are
the revision dates for the interest rate) is assumed to be described by a geometric
Brownian motion, the well-known Black Formula provides an explicit value for the
Caplets. These are financial contracts with the following payoff:
Xi αi · max {L(Ti−1 , Ti ) − R; 0} ,
.
where .αi = Ti − Ti−1 is the tenor and R is the Cap rate. The Black Formula for
Caplets is the following:
Capli (t) = αi · p(t, Ti ) [L(t, Ti−1 , Ti ) · N(d1 ) − R · N(d2 )] ,

. i = 1, ..., N,
(10.32)
where:

1 L(t, Ti−1 , Ti ) σi2
.d1 = √ ln + (Ti−1 − t)
σi Ti−1 − t R 2

1 L(t, Ti−1 , Ti ) σi2
d2 = √ ln − (Ti−1 − t)
σi Ti−1 − t R 2
and .σi denotes the Black volatility for the i-th Caplet. The Black volatilities must
be calibrated on the market values and, in order to simplify the treatment, in the
exercises we shall assume that .σi take a constant value (we shall assume a flat
structure for the Black volatilities).
The reader interested in a rigorous and systematic treatment of the valuation
problems for interest rate derivatives should consult the books by Björk [6], Brigo
and Mercurio [9], El Karoui [18] and Mikosch [32].
Exercise 10.1 Consider a short rate model of Vasiček type, that is (rt )t≥0 satisfying
.
with a = 0.4, b = 0.01, σ = 0.2 and initial value r0 = 4% on an annual basis.

1.
(a) Compute the value at time t = 0 of a zero-coupon bond with face value 40
euros and maturity T ∗ = 2.5 years.
(b) Discuss whether there exists a constant interest rate value r ∗ such that the
value of a zero-coupon bond with face value 40 euros and maturity T ∗ = 2.5
coincides with the bond value considered in item 1.(a).
2. Consider a European Call option with maturity T of 1 year and 2 months, strike
K = 36 euros and written on a zero-coupon bond with face value 40 euros.
(a) Compute the initial value of the European Call considered.
(b) Compute the value of the same Call option of the previous item under the
assumption that the short rate has constant value re = E [rT ] , where T
denotes the maturity of the Call option.
3. Consider now a coupon bond with face value N ∗ , maturity T ∗ = 2.5 years, with
coupon paid at the end of each semester with annual rate of 8%.
(a) Find the value of N ∗ such that the zero-coupon bond of item 1.(a) coincides
with the present value of the coupon bond, if this value is computed under
the assumption of a constant short rate with value r0 .
(b) Compute the value (at time t = 0) of a European Call option with maturity
T of 1 year and 2 months, strike K = 36 euros, written on the coupon bond
with face value N ∗ and maturity T ∗ = 2.5.
(c) Compute the value (at time t = 0) of a European Call option with maturity
T of 1 year and 2 months, strike K = 36 euros and written on a zero-coupon
bond with face value N ∗ .
4. Compute the value of the European Put options with the same parameters of the
Call options considered in items 2.(a) and 3.(b).
Solution
1.
(a) The price of the zero-coupon bond (with unit face value and maturity T ∗ ) at
time t in the Vasiček model is given by:
∗ ∗
Z t, T ∗ = eA(t,T )−rt B (t,T ) ,
.
where
∗ −t
1 − e−a (T )
B t, T ∗ =
.
a
and
σ2
B (t, T ∗ ) − (T ∗ − t) a2b − 2 σ 2 B 2 (t, T ∗ )
∗
A t; T
. = − .
a2 4a
To find the initial value of a zero-coupon bond with face value 40 euros and
maturity T ∗ = 2.5 years, we need to compute the quantities A (0; 2.5) and
B (0; 2.5).
Since
1 − e−0.4·2.5
B (0; 2.5) =
. = 1.58
0.4
(0.2)2
[B (0; 2.5) − 2.5] (0.4)2 · 0.01 − 2 (0.2)2 B 2 (0; 2.5)
A (0; 2.5) = −
(0.4) 2 4 · 0.4
= 0.043,
the initial value of the zero-coupon bond with face value 40 euros is then
given by:
N · Z (0; 2.5) = 40 · eA(0;2.5)−r0 ·B(0;2.5) = 40 · e0.043−0.04·1.58 = 39.2.

.
This means that we have to invest 39.2 euros today in order to get 40 euros
in two-and-a-half years.
By the previous results, it follows that the initial value of a zero-coupon bond
(with unit face value) is Z (0; 2.5) = 0.980 euros.
(b) We have to find a constant short rate value r ∗ such that the price of a zero-
coupon bond with face value 40 euros and maturity T ∗ = 2.5 years coincides
with the value found in item (a).
This short rate r ∗ must then satisfy the following equation:
∗ ∗
N · Z 0; T ∗ = N · e−r T
.
∗ ·2.5
or, equivalently, Z (0; 2.5) = e−r . We get then:
ln Z (0; 2.5)
r∗ = −
. = 0.008.
2.5
2. We focus now on a European Call option with maturity T of 1 year and 2 months
(that is T = 7/6 years) and strike K = 36, written on a zero-coupon bond with
face value N = 40 euros and maturity T ∗ = 2.5 years.
(a) We recall that the initial value of a European Call option written on a zero-
coupon bond with face value N is given by:

C0 = N · Z 0; T ∗ · N (d) − K · Z (0; T ) · N d − σp ,
. (10.33)
where T ∗ and T stand, respectively, for the maturity of the ZCB and of the
option, and

1 N · Z(0, T ∗ ) 1
.d = ln + σp. (10.34)
σp K · Z(0, T ) 2

∗
1 − e−a (T −T ) σ 2
σp = 1 − e−2aT . (10.35)
a 2a
First of all, we remark that the previous formula extends the “classical” Black-
Scholes formula and that the factor N ·Z (0; T ∗ ) simply represents the current
value of the underlying security, while K · Z (0; T ) represents the current
value (at t = 0) of the strike. This happens just because Z (0; s) represents
the present value of a unit of currency paid at time s.
Now, we turn to the Call option valuation problem. Looking at formula
(10.33), we can immediately observe that only Z (0; T ) must be computed,
since the value of Z (0; T ∗ ) has been already computed in 1.(a).
By proceeding in the same way as in 1.(a), we obtain that, for maturity T =

7/6 years (i.e. 1 year and 2 months),
7
1 − e−0.4· 6
B (0; 7/6) =
. = 0.9323
0.4

(0.2)2
B 0; 76 − 76 (0.4)2 · 0.01 − 2 (0.2)2 B 2 0; 76
A (0; 7/6) = −
(0.4)2 4 · 0.4
= 0.0052.
The zero-coupon bond (with unit face value) price is then given by:
Z (0; 7/6) = eA(0;7/6)−r0 ·B(0;7/6) = e0.0052−0.04·0.9323 = 0.9684.

.
Hence

1 − e−0.4(2.5−7/6) (0.2)2
σp =
. 1 − e−2·0.4·7/6 = 0.18
0.4 2 · 0.4

40·Z(0;2.5)
ln 36·Z(0;7/6) 0.18
d= + = 0.741.
0.18 2
As a direct consequence, the initial value of the European Call with maturity
T = 7/6, strike K = 36, written on the zero-coupon bond with face value
N = 40 and maturity T ∗ = 2.5 years is
C0 = 40 · 0.980 · N (0.741) − 36 · 0.9684 · N (0.741 − 0.18) = 5.365 euros.

.
(b) Assume that the short rate is constant with value re = E [rT ] and consider
a European Call option with maturity T = 7/6 years, strike K = 36 euros,
written on a zero-coupon bond with face value N = 40 euros and maturity
T ∗ = 2.5 years.
We recall that for the Vasiček model the explicit solution of the stochastic
differential equation describing the short-rate dynamics is the following:

t
rt = r0 e−at + b 1 − e−at + σ
. e−a(t−s) dWs .
0

Moreover, E [rt ] = r0 e−at + b 1 − e−at = r0 + (b − r0 ) 1 − e−aT . This
implies that:
re = E [rT ] = r0 + (b − r0 ) 1 − e−aT
.
7
= 0.04 + (0.01 − 0.04) 1 − e−0.4· 6 = 0.029.
At maturity, the Call option payoff is then the following:
+ ∗ +
. 40 · Z re T ; T ∗ − 36 = 40 · e−re (T −T ) − 36
+
= 40 · e−0.029·(2.5−7/6) − 36 = 2.483
and the value of the Call option with this payoff is its present value (computed
by using the short rate re ), i.e.
+
C0re = Z re (0; T ) · 40 · Z re T ; T ∗ − 36
.
+
= e−re T · 40 · Z re T ; T ∗ − 36 = 2.40 euros.
3. We focus now on a coupon bond with face value N ∗ , maturity T = 2.5 and
coupons paid at the end of each semester with annual rate 8%.
(a) We need to compute the value N ∗ that makes the price of the zero-coupon
bond of item 1.(a) equal to the present value of the coupon bond, when the
present value is computed assuming a constant short rate r0 . Since this value
is:
4N ∗ −r0 /2 104N ∗
. e + e−r0 + e−3r0 /2 + e−2r0 + e−5r0 /2 ,
100 100
we have to find N ∗ as a solution of the following equation:
4N ∗ −r0 /2 104N ∗
. e + e−r0 + e−3r0 /2 + e−2r0 + e−5r0 /2 = 40 · Z (0; 2.5) .
100 100
We obtain then:
40 · Z (0; 2.5)
N ∗ = 100 ·
. = 35.86 euros.
4 e−r0 /2 + e−r0 + e−3r0 /2 + e−2r0 + 104e−5r0 /2
(b) Let us compute now the initial value of a European Call option with maturity
T = 7/6, strike 36 euros, written on the coupon bond with face value N ∗ =
35.86, with maturity T ∗ = 2.5 and coupons paid at the end of each semester
at annual rate 8%.
A few preliminary remarks are in order. After the option maturity (which
does not coincide with any coupon payment date), the payment flow of the
coupons will take place as follows: the cash amount c = N ∗ · 0.04 = 1.43
euros) will be paid at times (always expressed in yearly units) T1∗ = 1.5,
T2∗ = 2 and T3∗ = T ∗ = 2.5 (this is the bond maturity). Since at this date
the bond cash value is paid back, the total cash amount received by the bond
buyer is N ∗ + c = 37.29 euros.
The basic idea in computing the price of a European option written on a
coupon bond is the following:
– “decompose” the underlying coupon bond as a linear combination of
n zero-coupon bonds of suitable face value, where n is the number of
coupons paid after the option maturity;
– “decompose” the option considered as a sum of n options, each one
written on the zero-coupon bonds of the previous step, and with suitable
strike;
– compute the value of the relevant option as the sum of the options
constructed in the previous step.
According to the previous remarks, let us try to decompose the Call option
strike into three contributions (K1 , K2 and K3 ) representing: the strike (K1 )
of the “virtual” option with maturity T and written on the zero-coupon bond
(ZCB1 ), with face value N1 = c = 1.43 and maturity T1∗ = 1.5 years;
the strike (K2 ) of a further “virtual” option with maturity T and written the
zero-coupon bond (ZCB2 ) with face value N2 = c = 1.43 and maturity
T2∗ = 2 years; and the strike (K3 ) of another “virtual” option with maturity T
and written on the zero-coupon bond (ZCB3 ) with face value N3 = N ∗ +c =
37.29 and maturity T3∗ = T ∗ = 2.5 years.
We impose the following requirement:

N1 · Z T ; T1∗ + N2 · Z T ; T2∗ + N3 · Z T ; T3∗ = K
. (10.36)
and define

.K1 N1 · Z T ; T1∗

K2 N2 · Z T ; T2∗

K3 N3 · Z T ; T3∗ .
We remark that condition (10.36) requires just the decomposition of K into

the three strikes K1 , K2 and K3 taking into account the respective face values
and the times to maturity of the three zero-coupon bonds.
In order to
compute
K1 , K2 and K3 , we just need to calculate Z T ; T1∗ ,
Z T ; T2∗ and Z T ; T3∗ .
By proceeding in a strictly similar way to that outlined in item 1.(a), we get:

∗
∗
1 − e−a (T1 −T ) 1 − e−0.4·(1.5−7/6)
.B T ; T1 = = = 0.312
a 0.4

A T ; T1∗ = 0.00001
∗
1 − e−a (T2 −T ) 1 − e−0.4·(2−7/6)
B T ; T2∗ = = = 0.709
a 0.4

A T ; T2∗ = 0.0018
∗
1 − e−a (T3 −T ) 1 − e−0.4·(2.5−7/6)
B T ; T3∗ = = = 1.033
a 0.4

A T ; T3∗ = 0.0078
and
∗ ∗ ∗ ∗
Z T ; T1∗ = eA(T ;T1 )−rT ·B (T ;T1 ) = e0.00001−0.312·rT
.
∗ ∗ ∗ ∗
Z T ; T2∗ = eA(T ;T2 )−rT ·B (T ;T2 ) = e0.0018−0.709·rT
∗ ∗ ∗ ∗
Z T ; T3∗ = eA(T ;T3 )−rT ·B (T ;T3 ) = e0.0078−1.033·rT ,
where rT∗ is unknown and can be determined via (10.36). By the previous
remarks and using N1 = N2 = 1.43 and N3 = 37.29, indeed, equality
(10.36) can be written as follows:
∗ ∗ ∗
1.43 · e0.00001−0.312·rT + 1.43 · e0.0018−0.709·rT + 37.29 · e0.0078−1.033·rT = 36.
.
The solution of this equation is rT∗ = 0.1168. The required decomposition is

then given by:

K1 = N1 · Z T ; T1∗ = 1.38
.

K2 = N2 · Z T ; T2∗ = 1.32

K3 = N3 · Z T ; T3∗ = 33.31.
By collecting the results available at this stage:

– the underlying coupon bond has been written as a linear combination of
three zero-coupon bonds with face values N1 = 1.43, N2 = 1.43 and
N3 = 37.29, respectively;
– the relevant option has been decomposed into three options with maturity
T = 7/6 years in the following way:
one option (Call 1) with strike K1 and written on ZCB1 with face value
N1 and maturity T1∗ = 1.5 years;
a second option (Call 2) with strike K2 and written on ZCB2 with face
value N2 and maturity T2∗ = 2 years;
a third option (Call 3) with strike K3 and written on ZCB3 with face value
N3 and maturity T3∗ = 2.5 years;
– the initial value of the option we want to evaluate C0CB is then:
(1) (2) (3)

C0CB = C0 + C0 + C0 ,
.
(i)
where C0 , i = 1, 2, 3, are the initial values of Call i.
(1) (2) (3)
So, we need to compute C0 , C0 and C0 . Since Call 1, Call 2 and Call 3
are options written on zero-coupon bonds, the following relation holds:

C0 = Ni · Z 0; Ti∗ · N d (i) − Ki · Z (0; T ) · N d (i) − σp(i) ,
(i)
.
where, for i = 1, 2, 3,

Ni ·Z(0,Ti∗ )
ln Ki ·Z(0,T )
1
d (i) =
.
(i)
+ σp(i)
σp 2

∗
1 − e−a (Ti −T ) σ 2
σp(i) = 1 − e−2aT .
a 2a
In our case, we have:

∗
1 − e−a (T1 −T ) σ 2
σp(1)
. = 1 − e−2aT = 0.054
a 2a

N ·Z(0,T ∗ )
ln K11 ·Z(0,T1 ) 1
d (1) = (1)
+ σp(1) = 0.6446.
σp 2

Moreover, it is immediate to verify that Z 0; T1∗ = 0.9664 and Z (0; T ) =
0.9684. Consequently,
(1)
C0 = 1.43·0.9664·N (0.664)−1.38·0.9684·N (0.664 − 0.054) = 0.058.
.
In the same way we can obtain:
(2)
C0 = 0.1335
.
(3)
C0 = 5.1871.
We can then conclude that the price of the European Call option considered
is:
C0CB = C0(1) + C0(2) + C0(3) = 0.058 + 0.1335 + 5.1871 = 5.3786.

.
(c) If we have a European Call written on a zero-coupon bond with face value
N ∗ = 35.86 euros, by the same procedure outlined before we obtain:

1 − e−0.4(2.5−7/6) (0.2)2
.σp = 1 − e−2·0.4·7/6 = 0.18
0.4 2 · 0.4

ln 35.86·Z(0;2.5)
36·Z(0;7/6) 0.18
d= + = 0.135.
0.18 2
Finally,

C0ZCB = N ∗ · Z 0; T ∗ · N (d) − K · Z (0; T ) · N d − σp
.
= 35.86 · 0.980 · N (0.135) − 36 · 0.9684 · N (0.135 − 0.18)

= 2.65 euros.
4. In order to compute the prices of the Put options with the same parameters of the
Call examined in items 2.(a) and 3.(b) we can adopt two different approaches.
The first consists in applying the European Put valuation formula, the second in
applying the Put-Call Parity relation for options written on zero-coupon bonds.
Let us compute the price of the Put option with the same parameters of the Call
in item 2.(a).
We recall that the initial value of a Put with maturity T and strike K, written on
a zero-coupon bond with face value N and maturity T ∗ , is given by:

P0 = K · Z (0; T ) · N σp − d − N · Z 0; T ∗ · N (−d) ,
. (10.37)
where d and σp are defined as in (10.34) and in (10.35).

Since all the quantities appearing in (10.37) have been already computed in item
2.(a), we get immediately that:
.P0 = 36 · 0.9684 · N (0.18 − 0.741) − 40 · 0.98025 · N (−0.741) = 1.029 euros.
Alternatively, and by directly applying the Put-Call parity holding for a European
Call and European Put with maturity T and strike K, written on the same zero-
coupon bond with face value N and maturity T ∗ :

C0 − P0 = N · Z 0; T ∗ − K · Z (0; T ) ,
.
we obtain the same result as before, i.e. P0 = 1.0274 ∼

= 1.029 euros.
Let us compute now the price of the Put option with the same parameters of the
Call considered in item 3.(b). As in the previous case, we have two possibilities:
one consists in repeating step-by-step the computations performed for the Call
option, the other consists in applying the Put-Call Parity to each of the Call
options appearing in the decomposition proposed (Call 1, Call 2 and Call 3,
in the present case) and finally sum up the values of the Put options in the
“decomposition”.
We are going to compute the Put value relying on the second approach, while
the calculations with the former strategy can be performed as an exercise by the
motivated student.
We have that:
P0CB = P0(1) + P0(2) + P0(3) ,

.
where Put i is a European Put with maturity Ti∗ , strike Ki and is written on a
zero-coupon bond with face value Ni . By the Put-Call Parity we obtain that

= C0 − N1 · Z 0; T1∗ + K1 · Z (0; T ) = 0.012
(1) (1)
P0
.

= C0 − N2 · Z 0; T2∗ + K2 · Z (0; T ) = 0.025
(2) (2)
P0

= C0 − N3 · Z 0; T3∗ + K3 · Z (0; T ) = 0.889,
(3) (3)
P0

since Z 0; T1∗ = 0.9664, Z 0; T2∗ = 0.9697, Z 0; T3∗ = 0.9803 and
Z (0; T ) = 0.9684. Finally,
.P0CB = P0(1) + P0(2) + P0(3) = 0.926 euros.
Exercise 10.2 Consider the instantaneous forward rate with the following affine
dynamics:
. f (0, t) = c + mt
for t ≥ 0.
1. Determine ϑt and the price of a zero-coupon bond with maturity t, as a function
of the parameters c and m when the short rate dynamics is described by one of
the following models: (a) the Ho-Lee model; (b) the Hull-White model.
2. Assume that the short rate (rt )t≥0 dynamics is described by the Ho-Lee model:
.
with r0 = c = 0.04, σ = 0.2 and with the ϑt obtained in the previous item.
(a) Find m such that the current value of a zero-coupon bond with maturity T =
7/6 and face value 40 euros is 36 euros.
(b) By using the parameters r0 and σ above and the value of m found in the
previous item, compute the value of a European Call option with maturity
T = 7/6, strike 36 euros and written on a zero-coupon bond with face value
40 euros and maturity of two years and a half.
3. Suppose now that the short rate (rt )t≥0 is described by the Hull-White model:

.
with r0 = c = 0.04, σ = 0.2 and with the value of m found in item 2.(a).
(a) Since the parameter a value is not known beforehand, but we have ϑ1/(2a) −
ϑ0 = 0.084, determine a.
(b) Compute the value of a European Call option with maturity T = 7/6, strike
36 euros and written on a zero-coupon bond with face value 40 euros and
maturity of two years and a half.
4. Determine ϑt and the value of a zero-coupon bond with maturity t, as a function
of c and m, when the instantaneous forward rate is of quadratic type, i.e.
m 2
f (0, t) = c + mt +
. t ,
2
and when the short rate dynamics is described by the Ho-Lee model with the
parameters of item 2.(a).
Compute the value of a zero-coupon bond with maturity one year and a half and
face value 40 euros.
Solution
1. Let us start by considering case (a) of a short rate following a dynamics of Ho-
Lee type. In this case we have:
∂f
ϑtH L = ϑt =
. (0, t) + σ 2 t = m + σ 2 t. (10.38)
∂t
As far as the price Z H L (0, t) of a zero-coupon bond with maturity t is concerned,

we need to find the quantities A (0, t) and B (0, t). Recall that for the Ho-Lee
model they are given by:
B(t, T ) = T − t
.
T
σ 2 (T − t)3
A(t, T ) = ϑs (s − T )ds + ,
t 6
so we immediately get B (0, t) = t. It is then enough to determine A (0, t) in

order to find Z H L (0, t).
Since r0 = f (0, 0) = c and

t σ 2t 3
A (0, t) =
. ϑs (s − t)ds +
0 6
t σ 2t 3
= m + σ 2 s (s − t)ds +
0 6
2 t
ms σ 2s3 σ 2 s 2 t σ 2t 3
= + − mst − +
2 3 2 0 6
mt 2 σ 2t 3 σ 2t 3 σ 2t 3
= + − mt 2 − +
2 3 2 6
mt 2
=− ,
2
we obtain that, in a Ho-Lee model with the assigned parameters, the initial value
of a zero-coupon bond with maturity t is given by:
mt 2 mt 2
Z H L (0, t) = eA(0,t)−r0 B(0,t) = e−
. 2 −r0 t = e− 2 −ct .
In case (b) of a Hull-White type dynamics, we have that:
∂f σ2
ϑtH W = ϑt =
. (0, t) + a · f (0, t) + 1 − e−2at
∂t 2a
σ2
= m + a (c + mt) + 1 − e−2at
2a
σ2
= m + ac + amt + 1 − e−2at .
2a
As far as the price of a zero-coupon bond with maturity t is concerned, we just
recall that Z H W (0, t) = eA(0,t)−r0 B(0,t) with
1 − e−at
B(0, t) =
.
a
t
1 2 2
A(0, t) = σ B (s, t) − ϑs B(s, t) ds.
0 2
Hence
t 2
1 2 1 − e−a(t−s) 1 − e−a(t−s)
.A(0, t) = σ − ϑs ds.
0 2 a a
In order to simplify the notation, we denote by Cs the integrand:

2
1 2 1 − e−a(t−s) 1 − e−a(t−s)
.Cs σ − ϑs .
2 a a
Since
σ2 σ2

Cs = 2a 2
1 + e−2a(t−s) − 2e−a(t−s) − m a − c − ms − 2a 2 1 − e
−2as
+m −a(t−s) + c · e−a(t−s) + ms · e−a(t−s) + σ e−a(t−s) − σ 2 e−at−as

2
ae 2a 2 2a 2
.
σ 2 −2a(t−s)
m σ 2 −2as σ2
= 2a 2
e − a + c − ms + 2a 2
e + m
a + c − 2a 2
e−a(t−s)
σ 2 −at−as
+ms · e−a(t−s) − 2a 2e ,
we get:
m t
σ 2 −2a(t−s) ms 2 σ 2 −2as
A(0, t) = 4a 3
e − a + c s − 2 − 4a 3
e
t 0
+ am2 + ac − 2a
2
−a(t−s) + σ e−at−as +
2 t −a(t−s) ds
0 ms · e
σ
3 e 2a 3

2 2
−2at − m + c t − mt − σ e−2at + 2 2
0
σ2 σ2
= σ
4a 3 − σ
4a 3 e a 2 4a 3 4a 3
+ am2 + ac − 2a 3
σ2 −at + σ 2 e−2at − σ 2 e−at
− am2 + ac − 2a 3 e 3 2a 3
. 2a
t
ms −a(t−s) m −a(t−s)
+ a e − a2 e
m 2
0
= − a + c t − mt2 + am2 + ac − am2 + ac e−at + mt
a − m
a2
+ m −at
a2
e
2
= −ct − mt2 + ac − ac e−at

−at − ct − mt 2 .
= a 1−e
c
2
Since r0 = f (0, 0) = c, in the Hull-White model considered the initial value of

a zero-coupon with maturity t is given by:
Z H W (0, t) = eA(0,t)−r0 B(0,t)

.

r0 −at
mt 2 1 − e−at
= exp 1−e − r0 t − − r0
a 2 a
2
mt
= exp −r0 t − .
2
2. Assume now that the short rate (rt )t≥0 is described by the following Ho-Lee
model:
.
with r0 = c = 0.04, σ = 0.2 and ϑt as obtained in (10.38), i.e.
ϑtH L = ϑt = m + σ 2 t.
.
(a) Let us start by determining the value m such that the current price of a zero-
coupon bond with face value 40 euros and maturity T = 7/6 years is equal
to 36 euros.
We have already obtained (see item 1.) the current price of a zero-coupon
bond as a function of m, that is:
mt 2
Z H L (0, t) = eA(0,t)−r0 B(0,t) = e−
. 2 −r0 t .
We need then to find the value m such that 40·Z (0; T ) = 36 or, equivalently,
mT 2 36
e−
. 2 −r0 T = .
40
It follows that:
ln (36/40) + r0 T ln (36/40) + 0.04 · 7/6
m = −2
. = −2 = 0.0862.
T2 (7/6)2
The value m required is then m = 0.0862. Consequently, ϑtH L = 0.0862 +

0.04 · t and f (0, t) = 0.04 + 0.0862 · t.
(b) We have to find the price of a European Call with maturity of 1 year and
2 months, strike 36 euros and written on a zero-coupon bond with face value
40 euros and maturity of two years and a half years.
Since, in a Ho-Lee model, the initial value of such a Call option is given by
C0 = 40 · Z(0, T ∗ ) · N(d) − 36 · Z(0, T ) · N(d − σp ),

.
with

40 · Z(0, T ∗ ) 1
d = σp−1 ln
. + σp
36 · Z(0, T ) 2
√
σp = σ (T ∗ − T ) T ,
it is necessary to compute Z (0, T ∗ ), Z (0, T ), d and σp .

The value Z (0; T ) is immediate to obtain: m has been actually chosen in
order to get Z (0; T ) = 36/40 = 0.9 euros.
Moreover: since T = 7/6, T ∗ = 2.5 and σ = 0.2, we obtain also
m (T ∗ )2 ∗ 0.0862·(2.5)2
.Z 0, T ∗ = e− 2 −r0 T = e− 2 −0.04·2.5
= 0.691

σp = 0.2 (2.5 − 7/6) 7/6 = 0.288

ln 40·0.691
36·0.9 0.288
d= + = −0.408.
0.288 2
Finally, the initial value of a European Call with maturity T = 7/6 (i.e. 1 year
and 2 months), strike 36 euros and written on a zero-coupon bond with face
value 40 euros and maturity of two years and a half, is given by:
C0 = 40 · 0.691 · N(−0.408) − 36 · 0.9 · N(−0.408 − 0.288) = 1.563 euros.

.
3. Let us consider now a short rate (rt )t≥0 with dynamics described by the Hull-
White model with ϑt as in item 1., a to be determined and with parameters r0 =
0.04 and σ = 0.2.
(a) We need to find the value of a first. We know that a must satisfy the following
condition:
. ϑ1/(2a) − ϑ0 = α, (10.39)
with α = 0.084.
By condition (10.39) we can obtain a. More precisely, (10.39) is equivalent
to:
1
+ σ2a 1 − e−2a· 2a
2
am · 1
2a =α
.
σ2

m
2 + 2a 1 − e−1 = α,
because (see item 1.)
σ2
ϑt = m + ac + amt +
. 1 − e−2at .
2a
Then:

σ 2 1 − e−1 σ 2 (e − 1)
.a = =
2 α− 2 m e (2α − m)
(0.2)2 (e − 1)
= = 0.309.
e (2 · 0.084 − 0.0862)
(b) We compute now the current price of a European Call with maturity of 1 year
and 2 months, strike 36 euros and written on a zero-coupon bond with face
value 40 euros and maturity of two years and a half.
Since, in the Hull-White model, the initial value of such a Call option is given
by:
.C0 = 40 · Z(0, T ∗ ) · N(d) − 36 · Z(0, T ) · N(d − σp ),
with

40 · Z(0, T ∗ ) 1
.d = σp−1 ln + σp
36 · Z(0, T ) 2

∗
1 − e−a(T −T ) σ 2
σp = 1 − e−2aT ,
a 2a
we need then to compute Z (0, T ∗ ), Z (0, T ), d and σp .

Since T = 7/6, T ∗ = 2.5, σ = 0.2, m = 0.0862 and a = 0.309, we obtain:
m (T ∗ )2 ∗ 0.0862·(2.5)2
Z 0, T ∗ = e− 2 −r0 T = e−
. 2 −0.04·2.5
= 0.691
mT 2 0.0862·(7/6)2
Z (0; T ) = e− 2 −r0 T = e− 2 −0.04·7/6
= 0.9

1 − e−0.309·(2.5−7/6) (0.2)2
σp = 1 − e−2·0.309·7/6 = 0.199
0.309 2 · 0.309

ln 40·0.691
36·0.9 0.199
d= + = −0.699.
0.199 2
Finally, the initial price of the European Call option (with maturity of 1 year
and 2 months, strike of 36 euros and written on a zero-coupon bond with face
value 40 euros and maturity of two years and a half) equals
C0 = 40 · 0.691 · N(−0.699) − 36 · 0.9 · N(−0.699 − 0.199) = 0.716 euros.

.
4. If the short rate dynamics is described by the Ho-Lee model and the instantaneous
forward rate is f (0, t) = c + mt + m2 t 2 , we have that:
∂f
ϑt =
. (0, t) + σ 2 t = m + mt + σ 2 t = m + m + σ 2 t.
∂t
As far as the price Z (0, t) of a zero-coupon bond with maturity t is concerned,
the unique quantity we still have to compute is A (0, t), since B (0, t) = t.
As r0 = f (0, 0) = c and
t σ 2t 3
A (0, t) =
. ϑs (s − t)ds +
0 6
t σ 2t 3
= m + m + σ 2 s (s − t)ds +
0 6
t
ms 2 m + σ 2 s3 m + σ 2 s 2 t σ 2t 3
= + − mst − +
2 3 2 6
0
mt 2 mt 3 σ 2t 3 mt 3 σ 2t 3 σ 2t 3
= + + − mt 2 − − +
2 3 3 2 2 6
mt 2 mt 3
=− − ,
2 6
we obtain that
mt 2 mt 3 mt 2 mt 3
Z (0, t) = eA(0,t)−r0 B(0,t) = e−
. 2 − 6 −r0 t = e− 2 − 6 −ct .
The value (at time t = 0) of a zero-coupon bond with maturity of 1 year and
2 months and face value 40 euros is then equal to:
0.0862·(7/6)2 0.0862·(7/6)3
40 · Z (0; 7/6) = 40 · e−
. 2 − 6 −0.04·7/6
= 35.189 euros.
Exercise 10.3 Assume the short rate (rt )t≥0 dynamics is described by the Hull-
White model:

. (10.40)
with r0 = 0.04, σ = 0.2 and ϑt associated to an instantaneous forward rate with the
following affine dependence on t:
f (0, t) = c + mt,
.
with m = 0.0862 and a = 0.309—as in Exercise 10.2, items 2.(a), 3.(a).

1. Compute the current price of a zero-coupon bond with maturity T of 1 year and
2 months and face value 32 euros, in the case of a short rate dynamics described
by the Vasiček model (considered as a particular case of Hull-White model with
ϑt = ab) with parameter b satisfying the following condition: ab = ϑT4 /ϑ0 .
2. Consider now the Hull-White dynamics for the short rate as in (10.40), with flat
term structure, i.e. with constant instantaneous forward rate f (0, t) = r0 for any
t ≥ 0.
Compute the value (at time t = 0) of a European Call option with maturity of
1 year and 8 months, strike of 38 euros and written on a coupon bond with face
value 40 euros, maturity of two years and a half and coupons paid at the end of
each semester at the annual coupon rate of 8%.
Solution
1. We already know from Exercise 10.2 that
∂f σ2
ϑt =
. (0, t) + a · f (0, t) + 1 − e−2at
∂t 2a
σ2
= m + a (c + mt) + 1 − e−2at
2a
σ2
= m + ar0 + amt + 1 − e−2at .
2a
Let us now consider the Vasiček model corresponding to the Hull-White model
considered, i.e.
. (10.41)
where the parameter b satisfies the condition ab = ϑT4 /ϑ0 . The value of b is then:
2 4
ϑT4 m + ar0 + amT + σ2a 1 − e−2aT
.b = = = 0.228,
aϑ0 a (m + ar0 )
because a = 0.309, m = 0.0862, r0 = 0.04, σ = 0.2 and T = 20/12 (1 year

and 8 months).
We need to compute the current value of a zero-coupon bond with face value 32
euros and maturity of 1 year and 8 months. In a Vasiček model, we obtain the
following results:
20
1 − e−aT 1 − e−0.309· 12
.B(0, T ) = = = 1.303
a 0.309
[B(0, T ) − T ] (a 2 b − σ 2 /2) σ 2 B 2 (0, T )
A(0, T ) = − = −0.06166
a2 4a
Z (0, T ) = eA(0,T )−r0 B(0,T ) = e−0.06166−0.04·1.303 = 0.8925.
The value (at t = 0) of a zero-coupon bond with face value 32 euros and maturity
of 1 year and 8 months is then equal to:
32 · Z (0; 20/12) = 28.56 euros.

.
2. Let us determine first ϑt in the case f (0, t) = r0 for any t ≥ 0. Since we are
working now with the Hull-White model, we can write
∂f σ2 σ2
ϑt =
. (0, t) + a · f (0, t) + 1 − e−2at = ar0 + 1 − e−2at .
∂t 2a 2a
We just recall that for any affine model the price of zero-coupon bond with
maturity t can be written as Z (0, t) = eA(0,t)−r0 B(0,t) with
1 − e−at
B(0, t) =
.
a
t
1 2 2
A(0, t) = σ B (s, t) − ϑs B(s, t) ds.
0 2
We get then A (0, t):

t 2
1 2 1 − e−a(t−s) 1 − e−a(t−s)
A(0, t) =
. σ − ϑs ds.
0 2 a a
In order to simplify the notation, denote by Cs the integrand, that is

2
1 2 1 − e−a(t−s) 1 − e−a(t−s)
.Cs σ − ϑs .
2 a a
Since
σ2
σ2

Cs = 2a 2
1 + e−2a(t−s) − 2e−a(t−s) − r0 1 − e−a(t−s) − 2a 2 1−e
−2as
2 2
+ 2a
σ −a(t−s) − σ e−at−as
. 2e
2
2a 2
−2a(t−s) − σ 2 e−a(t−s) − r 1 − e−a(t−s) + σ 2 e−2as − σ 2 e−at−as ,
= σ
2a 2 e 2a 2 0 2a 2 2a 2
the following relation holds:

t
σ 2 −2a(t−s) σ 2 −a(t−s) r0 −a(t−s)
.A (0, t) = e − 3e − r0 s + e
4a 3 2a a
0
t
σ2 σ2
+ − 3 e−2as + 3 e−at−as
4a 2a 0
r0 r0 −at
= −r0 t + − e .
a a
Consequently,
Z (0, t) = eA(0,t)−r0 B(0,t) = e−r0 t ,

.
as we could actually have guessed from the beginning.

In a strictly analogous way we can prove that:
r0
.A (t, T ) = −r0 (T − t) + [1 − e−a(T −t) ]
a
σ2
+ [2e−a(T −t) − 1 − e−2a(T −t)
4a 3
+e−2aT − 2e−a(T +t) + e−2at ]. (10.42)

r0 −a(T −t) 1 − e−a(T −t)
Z (t, T ) = exp −r0 (T − t) + (1 − e ) − rt
a a
2
σ
· exp 2e−a(T −t) − 1 − e−2a(T −t)
4a 3

+e−2aT − 2e−a(T +t) + e−2at . (10.43)
Let us now turn our attention to the initial problem, that is compute the initial
value of a European Call option with maturity of 1 year and 8 months, strike of
38 euros and written on a coupon bond with face value 40 euros, maturity of two
years and a half and with coupons paid at the end of each semester with coupon
rate of 8%.
We remark that, after the option maturity (which does not coincide with any of
the coupon payment dates), the coupons are paid at dates T1∗ = 2 years and T2∗
= T ∗ = 2.5 years (the bond maturity). At this last date, moreover, the face value
is paid back, the total amount paid then is N + c = 41.6 euros since the coupons
(paid in each semester) are equal to c = N · 0.04 = 1.6 euros.
In strict analogy with Exercise 10.2, we try to decompose the Call option strike
in two components (K1 and K2 ) representing, respectively, the strike of the
“virtual” option with maturity T and written on the zero-coupon bond (ZCB1 )
with face value N1 = c = 1.6 and maturity T1∗ = 2 years; the strike of the
“virtual” option with maturity T and written on the zero-coupon bond (ZCB2 )
with face value N2 = N + c = 41.6 and maturity T2∗ = T ∗ = 2.5 years.
We then impose the condition:

.N1 · Z T ; T1∗ + N2 · Z T ; T2∗ = K (10.44)
and define:

K1 N1 · Z T ; T1∗
.

K2 N2 · Z T ; T2∗ .
We remark that condition (10.44) only requires to decompose K into two strikes
K1 and K2 as a function of the face values and the time to maturity of the two
zero-coupon bonds.
In order to
determine K1 and K2 as mentioned above, we need Z T ; T1∗ and
Z T ; T2∗ .
From (10.43) we deduce

∗ r0 −a(T1∗ −T )
∗ −a(T1∗ −T ) ∗1−e
.Z T ; T1 = exp −r0 T1 − T + 1−e − rT
a a
σ2 ∗ ∗
+ [2e−a(T1 −T ) − 1 − e−2a(T1 −T )
4a 3
∗ ∗
+e−2aT1 − 2e−a(T1 +T ) + e−2aT ]
∗
= e−0.0021−0.3167·rT
∗
Z T ; T2∗ = e−0.0112−0.7347·rT ,
where rT∗ is unknown but can be obtained by (10.44). In fact, as a consequence

of our previous remarks and since N1 = 1.6 and N2 = 41.6, equality (10.44) can
be written as follows:
∗ ∗
1.6 · e−0.0021−0.3167·rT + 41.6 · e−0.0112−0.7347·rT = 38.
.
By solving explicitly this equation, we find the solution rT∗ ∼

= 0.16337. We get
then:

K1 N1 · Z T ; T1∗ = 1.516
.

K2 N2 · Z T ; T2∗ = 36.484.
At this stage we have:

– the decomposition of the underlying coupon bond as a linear combination
of two zero-coupon bonds with face values N1 = 1.6 and N2 = 41.6
respectively;
– the option considered has been then decomposed into two options (both with
maturity T = 20/12): the first one (Call 1) with strike K1 , written on ZCB1
with face value N1 and maturity T1∗ = 2 years; the second one (Call 2) with
strike K2 , written on ZCB2 with face value N2 and maturity T2∗ = 2.5 years;
– the initial price of our option C0CB is then given by:
(1) (2)
C0CB = C0 + C0 ,
.
(i)
where C0 , i = 1, 2 denotes the initial price of the European Call i.
(1) (2)
The remaining quantities to compute are then C0 and C0 . Since Call 1 and
Call 2 are options written on zero-coupon bonds, in the Hull-White model, the
following relation holds:

C0 = Ni · Z 0; Ti∗ · N d (i) − Ki · Z (0; T ) · N d (i) − σp(i) ,
(i)
.
where, for i = 1, 2,

Ni ·Z(0,Ti∗ )
ln Ki ·Z(0,T )
1
d (i) =
.
(i)
+ σp(i)
σp 2

∗
1 − e−a (Ti −T ) σ 2
σp(i) = 1 − e−2aT .
a 2a
In this case we have:

∗
1 − e−a (T1 −T ) σ 2
σp(1)
. = 1 − e−2aT = 0.0646
a 2a

N ·Z(0,T ∗ )
ln K11 ·Z(0,T1 ) 1
d (1) = (1)
+ σp(1) = 0.6606.
σp 2

It is immediate to verify that Z 0; T1∗ = 0.9231 and Z (0; T ) = 0.9355.
Consequently:
(1)
C0 = 1.6 · 0.9231 · N (0.6606) − 1.516 · 0.9355 · N (0.6606 − 0.0646)
.
= 0.0739 euros.
By proceeding in a strictly similar way we obtain
(2)
C0 = 4.3384.
.
We conclude then that the value at t = 0 of the European Call option considered
at the beginning is the following:
(1) (2)
.C0CB = C0 + C0 = 0.0739 + 4.3384 = 4.4123 euros.
Exercise 10.4 By using the Change of Numéraire technique, prove that the pricing
formula for a European Call option with strike K, maturity T , written on a zero-
coupon bond (with unit face value and maturity T ∗ ) when the short rate dynamics
is described by the Ho-Lee model, is the following:
C(t, T , K, T ∗ ) = p(t, T ∗ )N (d) − Kp(t, T )N (d − σp ),

. (10.45)
where N denotes the cumulative density of a standard normal and d and σp are
defined as follows:

p(t,T ∗ ) σp2
ln Kp(t,T ) + 2
.d =
σp
√
σp = σ (T ∗ − T ) T − t.
Solution The payoff of a Call option is the following: Φ = max {p(T , T ∗ ) − K; 0}.
A risk-neutral valuation argument provides the following expression for the Call
option price at time t:

T
C(t, T , K, T ∗ ) = EQ exp −
. rs ds max p(T , T ∗ ) − K; 0 | Ft .
t
(10.46)
Using the zero-coupon bond with maturity T as a numéraire, this formula can be
rewritten as:

p(T , T ∗ )
∗
.C(t, T , K, T ) = p(t, T ) · EQT max − K; 0 Ft , (10.47)
p(T , T )
where EQT denotes expectation computed with respect to the T-forward measure
QT , defined as the probability measure with respect to which the value of every
asset divided by p(t, T ) is a martingale.
Set now
p(t, T ∗ )
Zt
. . (10.48)
p(t, T )
Zt is a QT -martingale and its QT -dynamics can be easily obtained by that of the

bond prices. By the zero-coupon bond pricing formula in the Ho-Lee model, the
bond dynamics with respect to Q is in fact:
dp(t, T ) = rt p(t, T )dt − σ B(t, T )p(t, T )dWt .

. (10.49)
By applying Itô’s formula, we obtain the dynamics of Z:
dZt = σ 2 Zt B(t, T )(B(t, T ) − B(t, T ∗ ))dt + σ Zt [B(t, T ) − B(t, T ∗ )]dWt

.
= σ Zt (B(t, T ) − B(t, T ∗ ))[σ B(t, T )dt + dWt ]

= Zt σ B(t, T ) − B(t, T ∗ ) dWtT

= Zt σ (T − t) − (T ∗ − t) dWtT

= Zt σ T − T ∗ dWtT ,
t
where WtT = Wt + σ 0 B(s, T )ds and, by Girsanov’s Theorem, QT is the new
probability measure for which (WtT )t≥0 is a Brownian motion.
By comparing the present case with the usual one in the Black-Scholes model,
we deduce that
C(t, T , K, T ∗ ) = p(t, T )EQT [max {Z(T ) − K; 0} | Ft ]

.
= p(t, T ∗ )N (d) − Kp(t, T )N (d − σp ).
Exercise 10.5 By applying the Black Formula, find the value (at time t = 0) of a
Caplet with maturity of 2 years and tenor of one semester in a LIBOR market model.
Assume a flat term structure for the Black (forward) implied volatility σi = 0.6 (on
annual basis) and for the spot LIBOR at level 2% and a CAP rate R = 0.06 (i.e. the
forward implied volatility and the LIBOR rate are assumed to be constant over the
period considered). The face value is M = 100.000 euros.
Solution We recall the Black formula for Caplets:
Capl(0) = M · αi · p(0, Ti ) [L(0, Ti−1 , Ti )N (d1 ) − R · N(d2 )] ,

. (10.50)
where αi = Ti − Ti−1 and

L(0,Ti−1 ,Ti )
ln R + 12 Σi2 (0, Ti−1 )
d1
.
Σi (0, Ti−1 )

L(0,Ti−1 ,Ti )
ln R − 12 Σi2 (0, Ti−1 )
d2
Σi (0, Ti−1 )
with

Ti −1
Σi (0, Ti−1 )
. σi2 (u)du. (10.51)
0
In the present case, αi = Ti − Ti−1

= 0.5 years, σi = 0.6 , Ti = 2 years and
Ti −1 2 √
Σi (0, Ti−1 ) 0 σi (u)du = σi Ti−1 = σi Ti−1 = 0.7348.
2
From the definition of LIBOR spot rate we get:
1 = p(0, T ) [1 + L(0, T )T ] ,
. (10.52)
hence
1 1
.p(0, T ) = = . (10.53)
1 + L(0, T )T 1 + 0.02 · T
In particular,
1 1
p(0, Ti ) =
. = = 0.96
1 + 0.02 · Ti 1 + 0.02 · 2
1 1
p(0, Ti−1 ) = = = 0.97.
1 + 0.02 · Ti−1 1 + 0.02 · 1.5
It follows that the forward LIBOR rate at time t = 0 and for the interval [Ti−1 , Ti ]
equals
p(0, Ti−1 ) − p(0, Ti )

L(0, Ti−1 , Ti ) =
. = 0.0208. (10.54)
αi · p(0, Ti )
Consequently,
⎛ ⎞
ln 0.0208
0.06 + 12 (0.7348)2
.N (d1 ) = N ⎝ ⎠
0.7348
= N(−1.07) = 1 − N(1.07) = 1 − 0.8577 = 0.1423

⎛ ⎞
ln 0.0208
0.06 − 12 (0.7348)2
N (d2 ) = N ⎝ ⎠
0.7348
= N(−1.81) = 1 − N(1.81) = 1 − 0.9649 = 0.0351.
Finally, Capl(0) = 40.98 euros.
Exercise 10.6
1. Compute the current value of a zero-coupon bond with maturity of 1 year and
2 months and face value of 24 euros, by assuming that the short rate dynamics is
described by the Hull-White model with the same value of a as in item 3.(a) of
Exercise 10.2.
2. Compare the price obtained in the previous item with the one obtained by
assuming for the short rate dynamics of Vasiček type with parameter b satisfying
the condition ab = ϑ0 , where a and ϑ0 are the same as for the Hull-White model
of item 1.
Exercise 10.7 Assume that the short rate dynamics is described by the Vasiček
model:
.
where the values of the parameters a, b and σ are: a = 0.2, b = 0.1 and σ = 0.2.
Assume that, at time t = 0, the short rate is r0 = 4% (per year).
1. Compute the value (at time t = 0) of a zero-coupon bond with maturity T ∗ =
4 years.
2. Keeping the results of item 2. in mind, compute the price (at time t = 0) of a
zero-coupon bond with maturity T ∗ = 4 years and face value of 36 euros.
3. Establish if there exists a meaningful value of r0 (i.e. r0 ∈ (0, 1)) such that the
current value of the bond of item 2. coincides with the current value of the same
bond under the hypothesis of a constant short rate equal to the expected value of
rT ∗ .
4. Consider a European Call option with maturity T = 11/3 years and strike K =
32, written on a coupon bond with maturity T ∗ = 4 years, face value of 36 euros
and coupon rate of 8% per year (coupon rate paid at the end of each semester).
(a) Compute the value (at t = 0) of this European Call option.
(b) How would the price of the option considered change if its maturity was
3 years and 2 months?
(c) Compute the price of the Put option with the same parameters of the Call of
item 4.(b).
(1) (2) (3)
Exercise 10.8 Consider three short rates rt , rt and rt , with
t≥0 t≥0 t≥0
the following dynamics:
(1) (1)
drt
. = a1 b − rt dt + σ dWt
(2)
drt = ϑt dt + σ dWt
drt(3) = ϑt − a2 rt(3) dt + σ dWt ,
with real parameters a1 , a2 , b ∈ R, σ > 0 and a2 = 0.

1. Establish what is the model describing the dynamics of the short rates defined as
follows:
(2) (3) a1 (1)
r̂t rt
. − rt − r
a2 t
rt∗ rt
(2) (3) (1)
− rt + rt .
(2)
2. For a1 = a2 = a, find the probability distribution of rT with T = 1 year.
(2)
Discuss whether P rT ≥ 0 = 0.9 holds for some value of the parameters
b and σ .
Exercise 10.9 Assume that the short rate dynamics is described by the Hull-White
model:

.
where ϑt is related to f (0, t) of Exercise 10.2, item 4., r0 = 0.04, σ = 0.2 and with
the same parameters a and m of Exercise 10.2.
1. Compute the initial value of a European Put option of strike 36 euros, maturity
of 1 year and 2 months and written on a zero-coupon bond with face value of 40
euros and maturity of two years and a half.
2. In the corresponding Ho-Lee model (obtained by assigning the values a = 0 and
σ ), compute the price of the European Put of item 1.
3. Compute the value of the Put option of items 1.–2. in a Vasiček model with
parameters a, σ and r0 as before, ϑt = c for every t and with c constant and
arbitrary (but reasonable).
Exercise 10.10 By applying the Change of Numéraire technique, prove that the
initial value of a European Call option with maturity T and strike K, written on a
zero-coupon bond (with unit face value) with maturity S, when the short rate
dynamics is described by the Vasiček model, is given by the following formula:
C(0, T , K, S) = p(0, S) · N(d) − p(0, T ) · K · N(d − σp ),

.
where:
√
σp = σ (S − T ) T
.

1 p(0, S) 1
d= ln + σp .
σp p(0, T )K 2
Exercise 10.11 By the Black Formula, compute the value (at time t = 0) of a
Caplet with maturity of 2 years and tenor of one semester in a LIBOR market model,
by assuming a flat term structure for the Black (forward) implied volatility σi =
0.4 (on annual basis), and for the initial spot LIBOR at level 4% and a CAP rate
R = 0.04 (i.e. the forward implied volatility and the LIBOR rate are assumed to be
constant over the period considered). The face value is M = 400.000 euros.
Exercise 10.12 Solve the previous exercise proposed with the same data, but now
assume that the Black (forward) implied volatility is not constant in time, and it
grows linearly through different time intervals: σ0 = 0.4, σi = 1.5σi−1 , i = 2, 3, 4
(it is still constant inside each time interval), then solve again the previous exercise
proposed with the same data, but now assume that the LIBOR initial structure is not
flat any more and L1 (0) = 0.04, Li (0) = 1.5Li−1 (0), i = 2, 3, 4.
Chapter 11
Pricing Models Beyond Black-Scholes
In the previous chapters we presented several pricing and hedging problems both in
a discrete- and in a continuous-time setting. The basic model assumed in the first
case was the binomial model, while for the continuous-time case the Black-Scholes
model was assumed to be the framework, and in this last case the dynamics of the
risky assets was described by a geometric Brownian motion.
Empirical evidence suggests that these models provide a very rough description
of the financial markets’ behaviour, and they can be used just as a first approximation
of real markets’ modelling.
These models are nevertheless still somehow popular since they give several
explicit results that are not so easily available in more sophisticated models, and
they provide a simple and consistent approach to valuation and hedging problems.
In spite of this remark, some empirical features exhibited by asset prices that
cannot be explained by the Black-Scholes model are impossible to ignore; we
briefly mention the so-called Aggregational Gaussianity, Volatility Clustering, Fat
Tails and Leverage Effects. The first indicates that prices tend to be more normally
distributed when observed on a bigger time-scale, the second that volatility exhibits
some kind of persistence behaviour; moreover, the log-returns seem to assume
extreme values more likely than expected by a Gaussian distribution (fat tails),
and prices look negatively correlated with volatility (when volatility grows, prices
go down on average). Finally, a well-known effect exhibited by option prices is
completely outside the reach of the Black-Scholes description, and this is the so-
called Volatility Smile. This can be shortly summarized as follows. The coefficient .σ
entering in the Black-Scholes formula for European Call options can, in principle, be
estimated by the asset prices from historical data, but in practice it can be obtained
by inverting the same formula if some of these options are actively traded on the
market. The volatility value obtained by this second procedure is called Black-
Scholes implied volatility. Empirical evidence shows that implied volatility exhibits
256 11 Pricing Models Beyond Black-Scholes
a strong dependence both on the strike price and on time to maturity, while, if its
behaviour is consistent with the Black-Scholes description, it should be independent
from both.
In order to obtain a description of financial markets incorporating the previous
features, more complex models must be introduced. Models proposed in the
literature in the attempt to capture at least some of the previous features can be
collected into two main groups: the so-called Stochastic Volatility models and
models with jumps.
The Stochastic Volatility models propose to model the dynamics of volatility
(which is assumed simply to be constant in the Black-Scholes model) by a stochastic
differential equation driven by a Brownian motion correlated with the one driving
the asset’s dynamics. The most popular models of this kind are the following:
√ (1)
dSt = μSt dt + Yt St dWt
. (2) , (11.1)
dYt = ξ(η − Yt )dt + βdWt
√ (1)
. (2) , (11.2)
dYt = αYt dt + βYt dWt
√ (1)
. √ (2) , (11.3)
dYt = ξ(η − Yt )dt + θ Yt dWt

(1) (2)
where Y is the square of the volatility process and . Wt and . Wt
t≥0 t≥0
(1) (2)
are two correlated Brownian motions with .dWt dWt = ρdt. The first of the
models just written was proposed by Stein and Stein [41], the second by Hull and
White [26], and the third by Heston [24]. All three models allow to obtain, under
suitable simplifying assumptions, some kind of semi-closed form expression for
European option prices. Typically, it is possible to obtain an explicit expression
for the characteristic function of the log-return distribution, and hence—through
a technique pioneered by Carr and Madan [11] and now available in several
versions—get an explicit formula for the Fourier transform of the option prices.
Once this is available, a standard algorithm for inverting the Fourier transform can
provide the option prices for different strikes and maturities. Only the last step of this
procedure requires a numerical treatment, and this is the reason why these pricing
methods are said to provide solutions in semi-closed form. We shall illustrate by an
example how the log-return characteristic function can be computed.
Another direction explored in the attempt to explain some of the relevant features
of the market behaviour is the one introducing jumps in the stochastic process
describing the asset price dynamics.
The first model of this kind was proposed by Merton [31], and is called the Jump-
Diffusion model since it combines a geometric Brownian motion dynamics with a
Compound Poisson process.
The asset price dynamics is then assumed to be:

Nt
.St = S0 exp μt + σ Wt + Zi , (11.4)
i=1
where the term .μt represents the usual deterministic drift part, .σ Wt represents the
t
diffusion term and the term . N i=1 Zi consists of a Compound Poisson process, i.e.
it is the sum on .Nt independent and identically distributed random variables .Zi ,
where .Nt is a random variable (independent of .Wt and all the .Yi ) distributed as a
Poisson with intensity .λt (that is, .P (Nt = k) = e−λt (λt)k /k!). The original choice
proposed by Merton for the jump-size distribution was Gaussian with mean .γ and
variance .δ 2 , i.e. .Zi ∼ N(γ , δ 2 ).
Also in the jump-diffusion model described above, under suitable simplifying
assumptions, some explicit formulas for option prices are available. Merton [31]
proved, assuming that the risk-neutral dynamics of the underlying price differs from
the same dynamics under the historical measure only in the drift coefficient, that the
initial price of a European Call option with strike K and maturity T is provided by
the following formula:
+∞
(λkT )n
CJ D (S0 , T , K, r, σ, λ, γ , δ 2 ) = e−λkT
. CBS (S0 , T , K, σ (n), r(n)),
n!
n=0
(11.5)
where .CBS (S0 , T , K, σ (n), r(n)) denotes the value expressed by the Black-Scholes
formula of a European Call option with volatility parameter .σ (n) and risk-free
interest rate parameter .r(n) with

δ2
.k = exp γ + ,
2
δ2
σ (n) = σ 2 + n ,
T

n δ2
r(n) = r + λ(1 − k) + γ+ .
T 2
Still in the attempt to capture some of the features exhibited by asset price
behaviour, some stochastic volatility models in discrete-time have been proposed.
Among them we just mention the class of GARCH (Generalized Auto-Regressive
Conditional Heteroskedasticity) models introduced by Bollerslev [8], generaliz-
ing the class of ARCH models previously proposed by Engle [19]. The asset
price dynamics described by the simplest version of the GARCH model, called
GARCH(1,1), is the following:
Sn = Sn−1 exp (Xn ) = S0 exp {X1 + ... + Xn } ,

.
Xn = σn εn ,
σn2 = α0 + α1 Xn−1
2
+ βσn−1
2
= α0 + (α1 εn−1
2
+ β)σn−1
2
,
where .α0 > 0, .α1 ≥ 0 and .β ≥ 0 are real parameters, .S0 and .σ0 are the initial values
of the asset price and the volatility, respectively, and .(εi )i are i.i.d. (independent and
identically distributed) random variables, often assumed to be standard normal. This
model can explain the volatility persistence mentioned before and can provide fatter
tails for the log-returns distributions than those predicted by the log-normal model.
The option pricing issues arising in this model have been addressed by some authors
[14, 17], but their application goes far beyond the purpose of the present textbook.
Although both the stochastic volatility models and the models with jumps
improve the description of financial markets with respect to the Black-Scholes
model, they still provide a description of the volatility term structure that is not
completely realistic: while stochastic volatility models usually perform quite well
in describing volatility smiles for long maturities, models with jumps perform better
in capturing volatility smiles for short maturities. Some models have been proposed
that incorporate both features, stochastic volatility and jumps, in order to remove
the inconsistencies exhibited by both classes of the previous models. Some of them
allow to obtain an explicit expression for the characteristic function of log-returns
distribution, in such a way that semi-explicit valuation formulas for European
options are available. We shall provide a single example of how to compute the
log-returns characteristic function for a model of this class. An exhaustive list of
examples related to these topics is again far beyond the purpose of this textbook.
Exercise 11.1 Compute the value (at time t = 0) of a European Call option with
maturity T = 1 year, strike K = 30 euros, written on an underlying asset whose
dynamics is of jump-diffusion type with σ = 0.4, r = 0.05, and S0 = 30 euros; the
jump process is a compound Poisson with intensity λ = 1 (this means an average
of one jump each year) and jump-size distributed as a Gaussian with mean γ = 1
and variance δ 2 = 0.49. Assume that the change of measure only affects the drift
coefficient in the SDE describing the asset price dynamics.
Solution Assume (as suggested by Merton) that, when the historical probability
is replaced by the risk-neutral one, only the drift part of the dynamics is affected.
The explicit valuation formula for European Call options obtained by Merton can
therefore be applied. Hence we need to compute:
+∞
(λkT )n
CJ D (S0 , T , K, r, σ, λ, γ , δ 2 ) = e−λkT
. CBS (S0 , T , K, σ (n), r(n)),
n!
n=0
(11.6)
where

δ2
.k = exp γ + ,
2
δ2
σ 2 (n) = σ 2 + n ,
T

n δ2
r(n) = r + λ(1 − k) + γ+ ,
T 2
and CBS denotes the Black-Scholes value of a European Call option computed
by assigning the values σ (n), r(n) to the volatility and the risk-free interest rate,
respectively.
The valuation formula for the Call option in a jump-diffusion setting is actually
provided by a series, and only an approximate value can be computed by truncating
the series.
We remark that the general term contains a negative exponential-like factor (with
n! in the denominator) which makes the convergence rate of this series fast enough;
for this reason we decide to consider only the first 5 terms.
(n)
By adopting the following notation, CBS CBS (S0 , T , K, σ (n), r(n)), we
obtain:
(0) (1) (2) (3) (4)
. CBS 0, CBS = 16.41, CBS = 12.90, CBS = 23.37, CBS = 25.59.
We remark that the first term is similar to the value of a European Call option
valuated according to the Black-Scholes formula, since for n = 0 no jumps occur in
the underlying dynamics. Note that with more than two jumps the probability that
the Call option will expire out of the money is negligible. On the other hand, for
n ≥ 4, the factor 1/n! makes the contribution very small, allowing us to consider
only the contribution of the first 5 terms of the series relevant for the result.
By summing up all these contributions, we finally obtain:
(λkT )2 (2) (λkT )3 (3) (λkT )4 (4)

CJ D e−λkT CBS + λkT CBS +
(0) (1)
. CBS + CBS + CBS
2 6 24
= 14.05.
Exercise 11.2 Compute the conditional characteristic function of the log-returns

for the stochastic volatility model proposed by Heston:
√ (1)
. √ (2) .
dYt = ξ(η − Yt )dt + θ Yt dWt
Solution The market model proposed by Heston [24] is incomplete. There is

one Brownian motion appearing in the dynamics of the asset price and a second
Brownian motion appearing in the square of the diffusion coefficient’s dynamics:
these are two different (although correlated) sources of randomness, and only one
risky asset is traded on the market. We must specify then which one, among
the infinitely many equivalent measures turning the discounted asset price into a
martingale, we are going to assume describes the market behaviour. For simplicity
reasons we are going to assume that the market price of risk associated to the
volatility vanishes, i.e. the risk-neutral volatility dynamics is the same under both
the historical and the risk-neutral measures.
Applying Itô’s formula and a change of measure (from historical to risk-neutral),
it is easy to check that the dynamics of the log-return Xt ln (St /S0 ) is described
by the following SDE:
√
Yt (1)
dXt = r − 2 dt + Yt dWt
. . (11.7)
X0 = 0
The characteristic function (conditionally to the log-return and volatility values

at time t) is defined as follows:
f (x, y, t) E exp (iuXT ) |Xt = x, Yt = y ,

. u ∈ R. (11.8)
The function f , depending on t, x, y, must satisfy a stochastic differential equation

which can be easily obtained by a straightforward application of Itô’s Lemma:

1 ∂ 2f ∂ 2f
df =
. Yt 2 + ρθ Yt
2 ∂x ∂x∂y

1 ∂ 2f Yt ∂f ∂f ∂f
+ θ 2 Yt 2 + r − + ξ(η − Yt ) + dt
2 ∂y 2 ∂x ∂y ∂t
∂f (1)
∂f (2)
+ Yt dWt + θ Yt dWt . (11.9)
∂x ∂y
By looking at the definition of the function f we should remark that this is a

stochastic process obtained by conditioning an FT -adapted random variable with
respect to the information available at time t (i.e. with respect to the filtration Ft ).
This means that it must be an Ft -martingale. In a diffusion framework, like the
present one, this means that the drift coefficient appearing in the SDE describing
the dynamics of f must vanish identically, so that:

1 ∂ 2f ∂ 2f 1 2 ∂ 2f Yt ∂f ∂f ∂f
. Yt + ρθ Yt + θ Yt 2 + r − + ξ(η − Yt ) + = 0.
2 ∂x 2 ∂x∂y 2 ∂y 2 ∂x ∂y ∂t
(11.10)
In order to compute the solution of this PDE we need to impose a final condition
(this is a backward parabolic PDE). This turns out to be:
f (x, y, T ) = eiux .
. (11.11)
Inspired by PDEs of similar kind encountered in the previous chapters, we make an

educated guess of the following type:
f (x, y, t) = exp {A(τ ) + yB(τ ) + iux} ,

. (11.12)
where τ = T − t is the time to maturity. By plugging this expression for f in PDE

(11.10), we get

1 1 iu
. f Yt − u2 + ρθ iuB(τ ) + B 2 (τ ) − − ξ B(τ )
2 2 2
+f riu + ξ ηB(τ ) − A (τ ) − yB (τ ) = 0.
By imposing that both the sum of terms multiplying y and the sum of the terms
in which y does not appear vanish separately, we obtain the following system of
ordinary differential equations together with their initial conditions:

A (τ ) = ξ ηB(τ ) + iur, A(0) = 0
.
θ2 2 u2 +iu .
B (τ ) = (iuρθ − ξ ) B(τ ) + 2 B (τ ) − 2 , B(0) = 0
Although non-linear, the previous system exhibits some nice features: once the
differential equations satisfied by B is solved, the first equation can be immediately
reduced to quadratures and the unknown function A can be computed as an integral.
The second equation does not contain the unknown A and is a differential equation
which admits a closed-form solution; it is actually a Riccati equation with constant
coefficients, and the explicit solution can be found by separating the variables as
follows:
dB
. = dτ, (11.13)
aB 2 + bB + c
where a = θ 2 /2, b = iuρθ − ξ , c = −(u2 + iu)/2. The right-hand side of the

previous equation can be written as follows:
[a (y2 − y1 )]−1 dB [a (y2 − y1 )]−1 dB

. − , (11.14)
B − y2 B − y1
where y1,2 are the roots of the second-order algebraic equation aB 2 + bB + c = 0:

√
b2 − 4ac
−b − 1
.y1 = = 2 ξ − iuρθ − (iuρθ − ξ ) + θ u + iu
2 2 2
2a θ
√
−b + b2 − 4ac 1
y2 = = 2 ξ − iuρθ + (iuρθ − ξ ) + θ u + iu .
2 2 2
2a θ
By integrating the two rational functions and inverting the solution obtained, we
finally get:
⎧
⎪
⎨ A(τ ) = iruτ + ξ ητ (ξ −iρθu)
− 2ξθ 2η ln cosh βτ + ξ −iρθu
sinh βτ
θ2 2 β 2
.
u +iu
2 ,
⎩ B(τ ) =
⎪ −
β coth βτ
,
2 +ξ −iρθu
(11.15)

where β = (ξ − iρθ u)2 + θ 2 u2 + iu . By substituting A(τ ) and B(τ ) in
(11.12), we obtain the characteristic function we were looking for.
Exercise 11.3 Compute the characteristic function of the log-returns in the jump-
diffusion model proposed by Merton:

Nt
St = S0 exp μt + σ Wt +
. Yi . (11.16)
i=1
Solution The log-return process in the jump-diffusion model proposed by Merton

is the following:

Nt
St
Xt = ln
. μt + σ Wt + Yi . (11.17)
S0
i=1
It is the sum of a linear deterministic component (μt), a diffusion component (a

Brownian motion with diffusion coefficient σ ), and a jump part represented by a
compound Poisson process with intensity λ and normal jump-size distribution.
The characteristic function of Xt is given by:

Nt
E exp (iuXt ) = E exp iu μt + σ Wt +
. Zi , u ∈ R. (11.18)
i=1
We remark that in the Merton model the jump and the diffusion processes are
assumed to be independent, so that the expectation above can be computed as
the product of the expectations of each contribution. The first contribution is the
characteristic function of a Brownian motion with drift, i.e. it is the characteristic
function of a normal random variable with mean μt and variance σ 2 t:

σ2
E exp {iu (μt + σ Wt )} = exp iuμt − iu2 t .
. (11.19)
2
The characteristic function of the compound Poisson process can be computed

as follows, by conditioning with respect to the variable Nt :

Nt
Nt

E exp iu
. Zi = E E exp iu Zi Nt

i=1 i=1

+∞

Nt

= E exp iu Zi Nt = n P (Nt = n).

n=0 i=1
Since the Zi are independent and identically distributed, the last conditional
expectation in the previous expression becomes:

Nt
n

E exp iu
. Zi Nt = n = E exp(iuZi )

i=1 i=0

n
= E[exp(iuZi )]
i=0
= [φ(u)]n ,
where φ(u) is the characteristic function of the random variable Zi . For a normal
jump-size distribution with mean γ and variance δ 2 , we have:

δ2
φ(u) = exp iγ u − u2 .
. (11.20)
2
Finally, we obtain:

Nt +∞

Nt

E exp iu
. Zi = E exp iu Zi Nt = n P (Nt = n)

i=1 n=0 i=1
+∞
e−λt (λt)n
= [φ(u)]n
n!
n=0
= exp {λt (φ(u) − 1)} .
Putting together the results in (11.19) and (11.21), we conclude that

2
2σ
.E exp (iuXt ) = exp iuμt − iu t · exp {λt (φ(u) − 1)} . (11.21)
2
Exercise 11.4 By combining the results obtained in Exercises 11.2–11.3, compute

the characteristic function of the log-returns for the following model proposed by
Bates [5]:
√ t !
(1)
St = S0 exp μt + Yt Wt + N i=1 Zi .
. √
dYt = ξ(η − Yt )dt + θ Yt dWt(2)
Specify by which equivalent martingale measure you are going to describe the risk-
neutral dynamics of the asset price.
Solution As far as the jump-diffusion dynamics of the asset price is concerned, we
shall work under the usual assumption that the change of measure from the historical
to the risk-neutral probability affects only the drift component, in such a way that
in order to get a local martingale for the discounted asset price dynamics we can
write:

Yt (1)

Nt
St = S0 exp
. r− − λκ t + Yt Wt + Zi , (11.22)
2
i=1
where κ is defined by:

δ2
.κ = λ exp γ + −1 . (11.23)
2
Moreover, we shall assume a vanishing risk-premium for the volatility component,

so that the variance dynamics under the risk-neutral measure remains unchanged.
Under these assumptions, the characteristic function can be computed by observ-
ing that in the model under consideration the jump process (the compound
Poisson) appearing in the asset price dynamics is independent of the diffusion part.
Consequently, the characteristic function of the log-returns is simply the product

of the two characteristic functions of the diffusion and the jump components,
which can be computed separately. The diffusion component is just the same as
the one appearing in the Heston model (computed explicitly in Exercise 11.2),
while the characteristic function of the compound Poisson has been computed in
Exercise 11.3. The characteristic function of the log-return Xt = ln(St /S0 ) is the
given by:

ξ ηt (ξ − iρθ u)
E exp (iuXt ) = exp i (r − λκ) ut +
. ·
θ2
2ξ η
βt ξ − iρθ u βt − θ 2
· cosh + sinh ·
2 β 2
⎧ ⎫
⎨ u + iu
2 ⎬
· exp − Y0 · exp {λt (φ(u) − 1)} ,
⎩ β coth βt + ξ − iρθ u ⎭
2
where φ(u) is the characteristic function of a normal random variable with mean γ
and variance δ 2 , that is,

δ2
φ(u) = exp iγ u − u2 .
. (11.24)
2
Exercise 11.5 Prove that the solution of the recurrence equation describing the
GARCH(1,1) model is the following:
⎧ n %i
⎪
⎨ σn2 = α0 1 + i=1 j =1 (α1 εn−j
2 + β)
& .
.
⎪ % (11.25)
⎩ Xn = εn α0 1 + ni=1 ij =1 (α1 εn−j
2 + β)
Solution The equation characterizing the GARCH(1,1) model is a particular case

of a class of recurrence equations of the following form:
Mn = An Mn−1 + Bn ,
. (11.26)
where (An )n∈N and (Bn )n∈N are sequences of independent and identically dis-
tributed (i.i.d.) random variables. The solution of this equation can be easily
obtained by induction
Mn = An (An−1 Mn−2 + Bn−1 ) + Bn

.
⎛ ⎞
k
i−1
k
= Bn + ⎝Bn−i An−j ⎠ + Mn−k−1 An−i .
i=1 j =0 i=0
In order to ensure the convergence of this expression, as k → +∞ the third term

must vanish and the second must converge. By the Strong Law of Large Numbers
we know that:
1
k
. ln |An−i | → E[ln |An |], P -almost surely. (11.27)
k+1
i=0
So, under the assumption E [ln |An |] < 0, we have:

k
k
. |An−i | = exp ln |An−i |
i=0 i=0
k+1
1
k
= exp ln |An−i | → 0, P -a.s.
k+1
i=0
By proceeding in a similar way we can prove that the condition E [ln |An |] < 0
ensures that the term:
⎛ ⎞

k
i−1
. ⎝Bn−i An−j ⎠ (11.28)
i=1 j =0
converges as k → ∞, in such a way that the unique solution is

⎛ ⎞
+∞

i−1
Mn = Bn +
. ⎝Bn−i An−j ⎠ . (11.29)
i=1 j =0
In the particular case of the GARCH(1,1) models, Mn = σn2 , An = α1 εn−1

2 +β
and Bn = α0 (with Bn−i = 0 for i > n). Equation (11.29) implies:
⎡ ⎤

n
i
σn2 = α0 ⎣1 +
.
2
(α1 εn−j + β)⎦ . (11.30)
i=1 j =1
Since Xn = σn εn , (11.25) follows.

Exercise 11.6 Recall that a time series (Xn )n∈Z is called stationary (in strict sense)
if the random vector (Xt1 , ..., Xtn ) has the same distribution of (Xt1 +k , ..., Xtn +k )
for all t1 , ..., tn , k ∈ Z and for all n ∈ N.
Prove that for the GARCH(1,1) model the condition α1 + β < 1 is sufficient to
ensure that the sequence (σn )n∈N is stationary.
Solution The unique solution of the recurrence relation Mn = An Mn−1 + Bn (n ∈

Z), under the assumption E [ln |An |] < 0, has been shown (see Exercise 11.5) to be
the following:
⎛ ⎞
+∞

i−1
Mn = Bn +
. ⎝Bn−i An−j ⎠ .
i=1 j =0
For a GARCH(1,1) model, An = α1 εn−1 2 + β and Bn = α0 (with Bn−i = 0

for i >
n). The condition
E [ln |A n |] < 0, ensuring (σn )n∈Z is stationary, becomes
E ln α1 εn2 + β < 0, which in turn translates into the condition α1 + β < 1,
since the innovations (εi )i are i.i.d. standard normal random variables. The solution
obtained is a function of the i.i.d. random variables An , Bn , and this implies that the
process (Mn )n∈Z is strictly stationary.
Exercise 11.7 Compute the value (at time t = 0) of a European Call option with
maturity T = 1 year, strike K = 50 euros, written on an underlying asset whose
dynamics is of jump-diffusion type with σ = 0.8, r = 0.04, and S0 = 45 euros;
the jump process is a compound Poisson with intensity λ = 0.5 and jump-size
distributed as a normal with mean γ = 2 and variance δ 2 = 0.64. Assume that the
change of measure only affects the drift coefficient in the SDE describing the asset
price dynamics.
Exercise 11.8 Compute the conditional characteristic function of the log-returns
for the stochastic volatility model proposed by Stein and Stein and under the same
assumptions of Exercise 11.2:
√ (1)
. (2) .
dYt = ξ(η − Yt )dt + βdWt
Exercise 11.9 Consider a GARCH(1,1) model with parameters α0 = 0.7, α1 = 0.5

and β = 0.4. Verify that the sequence σn2 is stationary and write explicitly σ10
2.
Exercise 11.10 An ARCH(1) model describes the log-returns (squared) Xn2 of a

financial asset through the following recurrence equation:
Xn2 = α0 εn2 + α1 εn2 Xn−1

.
2
.
Assume that (εn )n is a sequence of i.i.d. standard normal random variables,

and prove that the condition α1 < 3.562 is sufficient to ensure Xn2 n is strictly
stationary, and also that the solution of the recurrence equation is the following:
⎛ ⎞
+∞

i
Xn2 = α0
. ⎝α1i 2 ⎠
εn−j .
i=0 j =0
Chapter 12
Risk Measures: Value at Risk and Beyond
In the late 90s an increasing interest has been developing towards risk measures, in
particular the Value at Risk (VaR) and the Conditional Value at Risk (CVaR). The
use of such risk measures is due, on the one hand, to the rules imposed by the Basel
Accord on the deposit of margins by banks and financial institutions because of
the financial risks they are exposed to. On the other hand, these tools are important
to quantify the riskiness assumed by an investor or an intermediary because of his
financial transactions.
We recall briefly some notions on risk measures. For a detailed treatment and
further details, we refer to Föllmer and Schied [21], Hull [25], Jorion [27] and
Barucci et al. [4], among many others.
Consider the Profit & Loss (P&L) or the return of a financial position at a
future date T , and denote with .X the family of random variables representing
P&Ls or returns of the financial positions taken into account. The space .X =
L∞ (Ω, F , P ) is often considered, where .L∞ (Ω, F , P ) (or, simply, .L∞ ) is
formed by all the essentially bounded random variables, i.e. by random variables
X satisfying .P (|X| ≤ K) = 1 for some .K > 0.
Given any random variable .X ∈ X , the Value at Risk at the level .α ∈ (0, 1) of
X is defined as
V aRα (X) −qα+ (X) = − inf{x ∈ R : FX (x) > α},

. (12.1)
i.e. as the opposite of the greatest .α-quantile of X. Notice that, for continuous
random variables X, .V aRα (X) is nothing but the opposite of the quantile .qα (X)
solution of .FX (x) = α.
The financial interpretation of the V aR is the following. If .V aRα (X) > 0, then
.V aRα (X) represents the minimal amount to be deposited as a margin for X.
270 12 Risk Measures: Value at Risk and Beyond
We list now some results on the V aR for stock returns of the following form:
ΔS ∼ √
. = μ · Δt + σ · ε · Δt,
S0
where .Δt = T is the length of the period of time considered, .S0 > 0 is the initial
stock price, .μ ∈ R its drift, .σ > 0 its volatility and .ε ∼ N (0; 1).
The V aR of the Profit & Loss of a stock of current price .S0 and daily return
distributed as a normal with drift .μ ∈ R and volatility .σ > 0, is:
√
V aRα (ΔS) = S0 σ Δt · N −1 (1 − α) − μ · Δt .
.
The V aR of the Profit & Loss of a portfolio formed by .xi (for .i = 1, 2, . . . , n)

stocks i with current prices .S0i , where the daily returns are jointly normal distributed
with drift .μi ∈ R, volatility .σi > 0 and correlation .ρij (.i, j = 1, 2, . . . n), is:

√ n
n
n
−1
(1 − α) · Δt
j j
.V aRα (ΔP ) = N xi xj S0i S0 σi σj ρij − xj S0 μj (Δt).
i=1 j =1 j =1
(12.2)
In particular: if all drifts are zero, then

√ n
n
V aRα (ΔP ) = N −1 (1 − α) · Δt
j
. xi xj S i S σi σj ρij .
0 0 (12.3)
i=1 j =1
The first-order approximation (Delta approximation) of the V aR for the Profit

& Loss of a portfolio formed by .xi (.i = 1, 2, . . . , n) options having Delta .Δi ,
underlying stocks of current prices .S0i and daily returns that are jointly normal with
null drifts, volatility .σi > 0 and correlation .ρij (.i, j = 1, 2, . . . n), is given by:

√ n
n
V aRαΔ approx (ΔP ) ∼
= N −1 (1 − α) · Δt
j j
. xi xj Δi Δ S i S σi σj ρij .
0 0 0 0
i=1 j =1
(12.4)
Although the V aR satisfies monotonicity (and, more in general, consistency

with respect to first-order stochastic dominance), cash-invariance and positive
homogeneity (recalled below), many authors (see, among others, Artzner et al. [1])
have underlined its several drawbacks. First of all, the V aR does not encourage
diversification of risk (namely, it does not satisfies subadditivity); furthermore, it
does not distinguish tails that are even very different at the left of .−V aR (i.e.
for losses greater than V aR). The need to overcome the latter problem led to
the introduction of the so-called Tail Conditional Expectation. More precisely, if

X is a random variable having finite expected value .E [X], the Tail Conditional
Expectation (TCE) of X at level .α ∈ (0, 1) is defined as:

EP −X1{X≤−V aRα (X)}
.T CEα (X) EP [−X|X ≤ −V aRα (X)] = . (12.5)
P (X ≤ −V aRα (X))
Unfortunately, neither the T CE encourages diversification in general. For further

details, see Artzner et al. [1] and Delbaen [15].
In order to amend the drawbacks of V aR and T CE but at the same time preserve
the main properties, Artzner et al. [1] and Delbaen [15] introduced the so-called
coherent risk measures. More details on coherent risk measures and, in particular,
on the CV aR can be found in Artzner et al. [1], Delbaen [15], Föllmer and Schied
[21], Rockafellar and Uryasev [38], Barucci et al. [4] and Fiori et al. [20], among
many others.
A coherent risk measure .ρ is defined (see [1], [15]) as a risk measure .ρ : X → R
satisfying the following properties:
• monotonicity: if .X ≤ Y , P -a.s., with .X, Y ∈ X , then .ρ(X) ≥ ρ(Y );
• cash-invariance: .ρ(X + c) = ρ(X) − c for any .X ∈ X , .c ∈ R;
• subadditivity: .ρ(X + Y ) ≤ ρ(X) + ρ(Y ) for any .X, Y ∈ X ;
• positive homogeneity: .ρ(λX) = λ · ρ(X) for any .X ∈ X , .λ ≥ 0.
When .X = L∞ , the most relevant examples of coherent risk measures are the
Conditional Value at Risk and the risk measures generated by a set of probability
measures. The Conditional Value at Risk at level .α ∈ (0, 1) is defined by:
E[(x − X)+ ]
CV aRα (X) inf
. −x ; (12.6)
x∈R α
while a risk measure associated to a set .Q of probability measures (known as set of

generalized scenarios) is given by:
ρQ (X) sup EQ [−X] .

. (12.7)
Q∈Q
Any coherent risk measure defined on a finite sample space or satisfying a further
hypothesis on continuity can be represented as in (12.7) (see Artzner et al. [1]), and
conversely.
It is also well known (see Föllmer and Schied [21]) that
E[(qα − X)+ ]
CV aRα (X) =
. − qα , (12.8)
α
where .qα is an arbitrary .α-quantile of X. Moreover, if X is a continuous random
variable, then .CV aRα (X) = T CEα (X).
Since .ρ (X) > 0 can be understood as the amount to deposit as a margin for X,
the acceptance set .A of .ρ is defined as the set
Aρ {X ∈ X : ρ (X) ≤ 0} .
.
Coherent risk measures and acceptance sets are in one-to-one correspondence.

Consider, for simplicity, a finite sample space. Then
• If .A satisfies the following properties .(A):
.A ⊇ {X ∈ X : X ≥ 0}
.A ∩ {X ∈ X : X ≤ 0 with X (ω) < 0 for some ω ∈ Ω} = ∅
.A is a convex cone,
then .ρA (X) inf {m ∈ R : m + X ∈ A } is a coherent risk measure.

• If .ρ is a coherent risk measure, then .Aρ satisfies properties .(A) and .ρAρ = ρ.
Several extensions of coherent risk measures are studied in the literature both
in a static and in a dynamic setting. Problems concerning the choice of portfolios
that minimize risk, capital allocations, and dynamic risk measures and their time-
consistency will be addressed in some of the exercises below. For portfolio
optimization we refer to Gaivoronski and Pflug [23], for the capital allocation
problem to Kalkbrener [28], Centrone and Rosazza Gianin [12], Dhaene et al. [16],
Tasche [42], while for (dynamic time-consistent) risk measures see Artzner et al.
[2], Cheridito and Stadje [13], Föllmer and Schied [21], Frittelli and Rosazza Gianin
[22], among many others.
Exercise 12.1 Consider two stocks whose Profit & Loss (per year) are represented,
respectively, by the following random variables:
⎧
⎪ −20; on ω1 ⎧
⎪
⎨ ⎨ 6; on ω1
−8; on ω2
.X = ; Y = 0; on {ω2 , ω3 }
⎪
⎪ 0; on ω3 ⎩
⎩ −2; on ω4
12; on ω4
with P (ω1 ) = 0.01, P (ω2 ) = 0.09, P (ω3 ) = 0.8 and P (ω4 ) = 0.1.
1. Consider the portfolio composed by 3 shares of the first stock (corresponding
to X) and by 4 shares of the other (corresponding to Y ). Establish whether risk
diversification is encouraged or not by V aR at 1%.
2. Consider the portfolio composed by one share of the first stock and by 8 shares of
the second. Establish whether risk diversification is encouraged or not by V aR
at 10%.
3. Consider the same portfolio as in item 2. Establish whether diversification of risk

is encouraged by CV aR at 10% and compute the profit due to diversification.
Solution
1. By the Profit & Loss of the two stocks we get the cumulative distribution
functions of X and Y :
⎧
⎪
⎪ 0; if x < −20 ⎧
⎪
⎪ ⎪ 0; if x < −2
⎪
⎨ 0.01; if − 20 ≤ x < −8 ⎪
⎨
0.1; if − 2 ≤ x < 0
.FX (x) = 0.1; if − 8 ≤ x < 0 FY (x) =
⎪
⎪ ⎪
⎪ 0.99; if 0 ≤ x < 6
⎪
⎪ 0.9; if 0 ≤ x < 12 ⎩
⎪
⎩ 1; 1; if x ≥ 6
if x ≥ 12
As a consequence, we deduce that

+
V aR0.01 (X) = −q0.01
. (X) = − inf{x ∈ R : FX (x) > 0.01} = 8
+
V aR0.01 (Y ) = −q0.01 (Y ) = − inf{x ∈ R : FY (x) > 0.01} = 2.
Let us consider a portfolio composed by 3 shares of X and by 4 shares of Y .

The Profit & Loss of such a portfolio is then given by
⎧ ⎧
⎪
⎪ −60 + 24; on ω1 ⎪
⎪ −36; on ω1
⎨ ⎨
−24; on ω2 −24; on ω2
.3X + 4Y = =
⎪
⎪ 0; on ω3 ⎪ 0;
⎪ on ω3
⎩ ⎩
36 − 8; on ω4 28; on ω4
whose cumulative distribution function is

⎧
⎪
⎪ 0; if x < −36
⎪
⎪
⎪
⎨ 0.01; if − 36 ≤ x < −24
.F3X+4Y (x) = 0.1; if − 24 ≤ x < 0 .
⎪
⎪
⎪
⎪ 0.9; if 0 ≤ x < 28
⎪
⎩
1; if x ≥ 28
Since
V aR0.01 (3X + 4Y ) = 24 < 3 · V aR0.01 (X) + 4 · V aR0.01 (Y ) = 32,

.
we can conclude that, for such a portfolio, V aR at 1% encourages diversification

of risk. The profit due to diversification (in terms of cash-saving on the margin to
be deposited) is then 8 euros.
2. By the cumulative distribution functions of X and of Y found above, we deduce

that
+
V aR0.1 (X) = −q0.1
. (X) = − inf{x ∈ R : FX (x) > 0.1} = 0
+
V aR0.1 (Y ) = −q0.1 (Y ) = − inf{x ∈ R : FY (x) > 0.1} = 0.
Consider now a portfolio composed by one share of X and by 8 shares of Y .

The Profit & Loss of such a portfolio is then given by
⎧
⎪
⎪ 28; on ω1
⎨
−8; on ω2
.X + 8Y = .
⎪
⎪ 0; on ω3
⎩
−4; on ω4
The cumulative distribution function of the Profit & Loss of the portfolio above
is then
⎧
⎪
⎪ 0; if x < −8
⎪
⎪
⎪
⎨ 0.09; if − 8 ≤ x < −4
.FX+8Y (x) = 0.19; if − 4 ≤ x < 0 ,
⎪
⎪
⎪
⎪ 0.99; if 0 ≤ x < 28
⎪
⎩ 1; if x ≥ 28
hence we can conclude that risk diversification is not encouraged by V aR at

10%. In fact,
V aR0.1 (X + 8Y ) = 4 > V aR0.1 (X) + 8V aR0.1 (Y ) = 0.

.
3. By formula (12.8) for the computation of CV aR and by item 2., we obtain
E[(q0.1 (X)−X)+ ]
CV aR0.1 (X) = α − q0.1 (X) = 20·0.01+8·0.09
0.1 = 9.2
. CV aR0.1 (Y ) = 2·0.1
0.1 =2
CV aR0.1 (X + 8Y ) = 4·0.09
0.1 + 4 = 7.6,
+ + +
by considering the quantiles q0.1 (X) = q0.1 (Y ) = 0 and q0.1 (X + 8Y ) = −4 =
−
q0.1 (X + 8Y ). Since
CV aR0.1 (X + 8Y ) = 7.6 < CV aR0.1 (X) + 8 · CV aR0.1 (Y ) = 25.2,

.
we conclude that there is diversification of risk with respect to CV aR for the

portfolio considered above (as well as for any other portfolio, because of the
properties of the CV aR) with a profit (in terms of cash-saving on the margin to
be deposited) of 17.6 euros.
Exercise 12.2 Consider a portfolio whose Profit & Loss is represented by the
following random variable:
⎧
⎪
⎪ −400; on ω1
⎪
⎪
⎪
⎪ −160; on ω2
⎪
⎨
−40; on ω3
.X =
⎪
⎪ 0; on ω4
⎪
⎪
⎪
⎪ 80; on ω5
⎪
⎩
1200; on ω6
with P (ω1 ) = 0.01, P (ω2 ) = p, P (ω3 ) = q, P (ω4 ) = 0.8, P (ω5 ) = 0.05 and
P (ω6 ) = 0.05, and p, q > 0 so that P is a probability measure.
1. Compute the Value at Risk at 1% and at 10% of X. Would we find a different
result with a maximum loss of 4000 (instead of 400)?
2. Find (if possible) p and q such that the V aR at 5% of X is equal to 40 and the
Tail Conditional Expectation at 5% of X does not exceed 100.
Solution The condition p + q = 0.09 has to be satisfied in order for P to
be a probability measure. Moreover, the cumulative distribution function of X
(depending on p, q) is
⎧
⎪
⎪ 0; if x < −400
⎪
⎪
⎪
⎪ 0.01; if − 400 ≤ x < −160
⎪
⎪
⎪
⎨ 0.01 + p; if − 160 ≤ x < −40
.FX (x) = 0.1; if − 40 ≤ x < 0 ,
⎪
⎪
⎪
⎪ 0.9; if 0 ≤ x < 80
⎪
⎪
⎪
⎪ 0.95; if 80 ≤ x < 1200
⎪
⎩
1 if x ≥ 1200
because 0.01 + p + q = 0.1 (by the arguments above).

1. By the results above, it is straightforward to deduce that V aR0.01 (X) = 160
and V aR0.1 (X) = 0. It is easy to check (also using FX ) that such results would
remain unchanged even if the maximum loss was 4000 instead of 400.
2. The conditions to be imposed on p and q (so that the V aR at 5% of X is equal
to 40 and the Tail Conditional Expectation at 5% of X does not exceed 100) are
the following:
⎧
⎨ p + q = 0.09 (P prob. measure)
. 0.01 + p ≤ 0.05 (condition on V aR) (12.9)
⎩
40 + 1600p + 400q ≤ 100 (condition on T CE)
where the last inequality is due to
E −X1{X≤−40} ]
T CE0.05 (X) = EP [−X|X ≤ −V aR0.05 (X)] = P P[ (X≤−40)
.
= 400·0.01+160p+40q
0.1 = 40 + 1600p + 400q.
Solving system (12.9) we obtain

⎧
⎨ q = 0.09 − p
0 < p ≤ 0.02
. p ≤ 0.04 .
⎩ q = 0.09 − p
p ≤ 0.02
A pair of admissible p and q is, for instance, p = 0.02 and q = 0.07.

Exercise 12.3 Consider two stocks (A and B) whose daily returns are jointly
normally distributed. The correlation of the stock returns is 0.25, the current price
of stock A is 4.2 euros, its daily drift 0.0002 and its daily volatility 1.2%, while the
current price of the stock B is 3.6 euros, its daily drift 0.0008 and its daily volatility
2.8%.
1. Find the V aR (at 1% and on a time interval of 10 days) of the portfolio (PORT1)
composed by 20 shares of A and by 15 shares of B.
2. Consider two other stocks (C and D) having the same features as stocks A and
B, respectively, but both having daily drift equal to zero.
Compute the V aR (at 1% and on a time interval of 10 days) of the portfolio
(PORT2) composed by 20 shares of C and 15 shares of D.
3. Consider now a portfolio (PORT3) composed by 20 options (with Delta ΔC =
0.4) on the underlying C and by 15 options (with Delta ΔD = −0.1) on the
underlying D. Give the Delta-approximation of the V aR (at 1% and on a time
interval of 10 days) of portfolio PORT3.
Solution
1. Applying formula (12.2) for the computation of the V aR of a portfolio of stocks
whose returns are jointly normally distributed, we get immediately that the V aR
(at 1% and on a time interval of 10 days) of portfolio PORT1 is
10gg
V aR0.01 ΔP port1
√
= N −1 (1 − α) · Δt · n
i=1
n i j
j =1 xi xj S0 S0 σi σj ρij
n j
− j =1 xj S0 μj (Δt)
−1
√
.
= N (0.99) · 10 · (20 · 4.2 · 0.012)2 + (15 · 3.6 · 0.028)2
+2 · 0.25 · (20 · 4.2 · 0.012 · 15 · 3.6 · 0.028)]1/2
− [20 · 4.2 · 0.0002 + 15 · 3.6 · 0.0008] · 10
= 14.831 − 0.6 = 14.231.
In other words, the amount of cash to be deposited as a margin for PORT1 is

equal to 14.231 euros.
2. Since the features of stocks C and D are the same as those of A and B, except for
the drifts that are zero for C and D, the V aR of portfolio PORT2 can be obtained
by taking only the first term in the computation above for the V aR of PORT1,
i.e.

√ n n
.V aR
10gg
ΔP port2
= N −1
(1 − α)· Δt j
xi xj S0i S0 σi σj ρij = 14.831.
0.01
i=1 j =1
In other words, the amount of money to be deposited as a margin for PORT2 is

14.831 euros.
3. Applying formula (12.4) for the Delta approximation of the V aR of a portfolio of
options having underlyings with jointly normal returns, we obtain immediately
that the V aR (at 1% and on a time interval of 10 days) of portfolio PORT3 is
10gg
V aR0.01 ΔP port3
√
. = N −1 (1 − α) · Δt · n
i=1
n i j i j
j =1 xi xj Δ0 Δ0 S0 S0 σi σj ρij
= . . . = 2.896.
In other words, the amount of cash to be deposited as a margin for PORT3 is

2.896 euros. Note that this margin is much smaller than the one of PORT2. This
ca be explained by the fact that some options in PORT3 have positive Δ, others
negative Δ. Hence, PORT3 is more diversified than PORT2.
Exercise 12.4 Consider the daily log-return of a stock represented by the random
variable X of mean 0.0002 and standard deviation 0.02.
1. Compute the V aR of X at 5% and on a period of 1 day, once X is assumed to be
normally distributed.
2. Applying the Cornish-Fisher expansion, compute the V aR of X at 5% and on a
period of 1 day when the skewness of X is supposed to be equal to (−0.3).
Solution Denote by m, s and ξ the mean, the standard deviation and the skewness
of X, respectively.
1. If X is distributed as a normal with mean m = 0.0002 and standard deviation
s = 0.02, then the V aR of X at 5% and on a period of 1 day is given by
.V aR0.05 (X) = s · N −1 (0.95) − m = 0.0327.

2. Applying the Cornish-Fisher expansion, the V aR of X at α = 0.05 and on a

period of 1 day can be approximated as follows:

1 2
V aR0.05 (X) ∼
. = −m − s · N −1 (0.05) + ξ N −1 (0.05) − 1 = 0.0344.
6
Exercise 12.5 Consider a financial position that is distributed as a Pareto with

parameters x0 = 100 and a = 2.
1. Compute the V aR at 10% of the position. Is a margin of 400 euros sufficient as
a deposit?
2. What about the V aR at 1%?
3. Compute the CV aR at 10% of the position. Is the margin of 400 euros also
sufficient in this case?
Solution Denote by X the random variable representing the P&L of the initial
position. The cumulative distribution function and the density function of X are,
respectively,
0; if x < x0
FX (x) =
. x0 a
1− x ; if x ≥ x0
0; if x < x0
fY (y) =
ax0a x −a−1 ; if x ≥ x0
First of all, note that E [X] = a

a−1 x0 / L∞ since X may take any
= 200 and that X ∈
value greater or equal to x0 .
1. Because X is a continuous random variable, the α0 -quantile of X solves
FX (x) = α0 .
.
It follows, therefore, that

x0 a
x = 1 − α0
. x0
x= 1/a .
(1−α0 )
For α0 = 0.10, we obtain that q0.10 (X) = √100 = 105.41 and that
0.9
V aR0.10 (X) = −105.41, hence that X is acceptable with respect to V aR0.10
as it is and it would not require any extra margin to be deposited. Furthermore,
V aR0.10 (X + 400) = −505.41, so the margin of 400 euros would be (more
than) enough for X.
2. The V aR at α1 = 0.01 of X can be computed in (at least) two ways. The first
way consists in proceeding as before, so to obtain
x0
V aR0.01 (X) = −
. = −100.50.
(1 − α1 )1/a
The second consists in recalling (see Barucci et al. [4], Fiori et al. [20]) that for
Pareto distributed random variables one has
1/a
1 − α0
. V aRα1 (X) = V aRα0 (X) · ,
1 − α1
hence
1/2
0.9
.V aRα1 (X) = −105.41 · = −100.50.
0.99
As from the item above, X would not require any extra margin to be deposited.
The margin of 400 euros would be therefore sufficient also when the riskiness of
X is evaluated by means of V aR at the level 1%.
3. Since X does not belong
to L∞ but to L1 , it is easy to check that CV aRα0 (X) =
T CEα0 (X) = EP −X|X ≤ −V aRα0 (X) is still true. Hence:

E P X1 X≤−V aRα0 (X)
.CV aRα0 (X) = EP −X|X ≤ −V aRα0 (X) = − .
P X ≤ −V aRα0 (Y )
(12.10)
By item 1., we deduce that V aRα0 (X) = − x0 1/a and that

(1−α0 )
P X ≤ −V aRα0 (X) = α0 . Set now x ∗ = x0
= −V aRα0 (X). We
(1−α0 )1/a
obtain that
x∗ x ∗
a

.EP X1 X≤−V aR
= ax0a x −a dx = x a x 1−a
α 0 (X)
x0 1−a 0
x0

a 1
= x0 1 −
a−1 (1 − α0 )1/a−1
a
= E [X] + (1 − α0 ) V aRα0 (X) .
a−1
From the argument above and from (12.10) it follows that

CV aRα0 (X) = EP −X|X ≤ −V aRα0 (X)
.
E [X] + a
a−1 (1 − α0 ) V aRα0 (X)
=−
α0
200 + 2 · 0.9 · (−105.41)
=− = −102.62.
0.1
We can then conclude that a margin of 400 euros would be (more than) enough
also in this case.
⎧
⎪
⎪ −1200; on ω1
⎪
⎪
⎪
⎪ −400; on ω2
⎪
⎨
−80; on ω3
.X = ,
⎪
⎪ 0; on ω4
⎪
⎪
⎪
⎪ 16; on ω5
⎪
⎩
x; on ω6
where 16 < x < 4000, P (ω1 ) = 0.01, P (ω2 ) = 0.04, P (ω3 ) = 0.05, P (ω4 ) = 0.7,
P (ω5 ) = 0.1 and P (ω6 ) = 0.1.
Find the smallest x such that, according to the coherent risk measure generated
by the set Q = {P , Q1 } (with Q1 (ω1 ) = Q1 (ω2 ) = Q1 (ω3 ) = Q1 (ω6 ) = 1/8
and Q1 (ω4 ) = Q1 (ω5 ) = 1/4), the margin to be deposited for X is smaller than or
equal to 200.
Solution First of all, recall that the coherent risk measure generated by the set of
generalized scenarios Q = {P , Q1 } is given by ρQ (X) supQ∈Q EQ [−X]. Let
us start by computing EP [−X] and EQ1 [−X]:
.EP [−X] = 1200 · 0.01 + 400 · 0.04 + 80 · 0.05

x
+0 − 16 · 0.1 − x · 0.1 = 30.4 −
10
1200 400 80 16 x x
EQ1 [−X] = + + +0− − = 206 − .
8 8 8 4 8 8
Since 30.4 − 10
x
≥ 206 − x8 holds if and only if x ≥ 7024, in the present case
(16 < x < 4000) we deduce that
x
.ρQ (X) = sup EQ [−X] = sup EP [−X] ; EQ1 [−X] = 206 − .
Q∈Q 8
In order to fulfill the constraint on the margin for X (according to ρQ ), one should
have 206 − x8 ≤ 200, hence x ≥ 48.
Consequently, 48 is the smallest x ∈ (16, 4000) verifying the conditions required
in terms of margin for X.
Exercise 12.7 Consider a sample space with only two elementary events, i.e. Ω =
{ω1 , ω2 }.
Find the coherent risk measure ρA associated to the acceptance set
x
A = (x, y) ∈ R2 : y ≥ −2x;
. y≥− .
2
Establish which of the following positions is preferred according to the risk measure
ρA :
−30; on ω1
X1 =
. ; X2 = 0.
150; on ω2
Solution It is easy to check that the acceptance set A (see Fig. 12.1) satisfies all
the properties (A), hence the risk measure
ρA (X) = inf {m ∈ R : m + X ∈ A }
.
associated to A is coherent.
Since Ω has only two elements, any random variable X on Ω can be identified
with (X(ω1 ), X(ω2 )) ∈ R2 . Note that m + X ∈ A if and only if
m + X (ω2 ) ≥ −2 (m + X (ω1 )) m ≥ − 23 X (ω1 ) − 31 X (ω2 )

. .
m + X (ω2 ) ≥ − m+X(ω
2
1)
m ≥ − 31 X (ω1 ) − 32 X (ω2 )
−4 −2 0 2 4
−2
Fig. 12.1 Acceptance set A

It follows that
2 1 1 2
m ≥ sup − X (ω1 ) − X (ω2 ) ; − X (ω1 ) − X (ω2 ) ,
.
3 3 3 3
hence
2 1 1 2
ρA (X) = sup − X (ω1 ) − X (ω2 ) ; − X (ω1 ) − X (ω2 ) .
.
3 3 3 3
In terms of sets of generalized scenarios, ρA can also be written as
ρA (X) =
. sup EQ [−X]
Q∈{Q1 ,Q2 }
where Q1 (ω1 ) = 23 , Q1 (ω2 ) = 13 and Q2 (ω1 ) = 13 , Q2 (ω2 ) = 23 .

For what concerns positions X1 and X2 , we obtain that
2 1 1 2
ρA (X1 ) = sup − · (−30) − · 150; − · (−30) − · 150 = −30
.
3 3 3 3
ρA (X2 ) = 0,
hence both positions are acceptable even if X1 is preferable over X2 (with respect
to ρA ).
Exercise 12.8 Consider a sample space with only two elementary events, i.e. Ω =
{ω1 , ω2 }.
Find the risk measure ρA associated to the acceptance set
x
.A = (x, y) ∈ R2 : y ≥ ; y ≤ 4x
2
and verify that ρA does not satisfy monotonicity.
Solution It is easy to check that the acceptance set A (see Fig. 12.2) does not
satisfy properties (A) (in particular A R2+ ). Consequently, the risk measure ρA
associated to such a set is not coherent.
Since Ω has only two elements, any random variable X on Ω can be identified
with (X (ω1 ) , X (ω2 )) ∈ R2 . Note that m + X ∈ A if and only if
m + X (ω2 ) ≥ m+X(ω
2
1)
m ≥ X (ω1 ) − 2X (ω2 )
. .
m + X (ω2 ) ≤ 4 (m + X (ω1 )) m ≥ − 43 X (ω1 ) + 13 X (ω2 )
It follows that
4 1
m ≥ sup X (ω1 ) − 2X (ω2 ) ; − X (ω1 ) + X (ω2 ) ,
.
3 3
−4 −2 0 2 4
−2
Fig. 12.2 Acceptance set A
hence
4 1
ρA (X) = sup X (ω1 ) − 2X (ω2 ) ; − X (ω1 ) + X (ω2 ) .
.
3 3
It is immediate to verify that ρA satisfies subadditivity, positive homogeneity

and cash-invariance.
The following counterexamples shows that monotonicity is not guaranteed by
ρA .
Taking, for instance,
0; on ω1
X1 =
. ,
30; on ω2
we obtain that X1 ≥ 0 but ρA (X1 ) = 10 > 0 = ρA (0), so X1 ∈ / A . In other

words, even if X1 will never give a loss, an amount of 10 euros has to be deposited
as a margin.
Furthermore, take the two positions:
−30, on ω1 −6, on ω1
Y1 =
. ; Y2 = .
60, on ω2 180, on ω2
We obtain that ρA (Y2 ) > ρA (X1 ) even if Y2 (ω) ≥ Y1 (ω) for any ω ∈ Ω. In fact,
4 1
. ρA (Y1 ) = sup −30 − 120; · 30 + · 60 = 60
3 3
4 1
ρA (Y2 ) = sup −6 − 360; · 6 + · 180 = 68.
3 3
It is also straightforward to check that
24, on ω1
Y2 − Y1 =
. ≥0
120, on ω2
and ρA (Y2 − Y1 ) = 8 > 0. Consequently, Y2 − Y1 ∈

/ A.
Exercise 12.9 Consider a portfolio formed by two stocks whose returns are jointly
distributed as bivariate normal. Suppose that the mean returns of the stocks are,
respectively, 10% and 6% per year, their standard deviations 40% and 30% per year
and their correlation amounts to 0.5.
1. Supposing short-selling is not allowed, find the optimal portfolio so to minimize
the Conditional Value at Risk of the portfolio return [minus its mean] at level
α = 0.01 and so to have an average return of 6.5% at least. What about the
optimal portfolio for an average return of at least 8%?
2. Show that the optimal portfolio of the previous item is the same when the
riskiness is measured by Value at Risk, Conditional Value at Risk or standard
deviation.
Solution Our goal is to construct a portfolio by investing a percentage x1 in one
stock (with return R1 ) and a percentage x2 in the other stock (with return R2 ). Since
short-selling is not allowed, x1 and x2 have to satisfy:
x1 , x2 ≥ 0,
.
x1 + x2 = 1,
hence x1 = x ∈ [0, 1] and √

x2 = 1 − x.
Let μi = E[Ri ], σi = V (Ri ) (for i = 1, 2) and denote with ρ the correlation
between R1 and R2 . Since R1 and R2 are jointly normal, then also the portfolio
return
Rp = x1 R1 + x2 R2 = xR1 + (1 − x)R2
.
is distributed as a normal with mean E[Rp ] = xE[R1 ] + (1 − x)E[R2 ] =

x(μ1 − μ2 ) + μ2 and with variance
V (Rp ) = V (xR1 +(1−x)R2 ) = x 2 σ12 +(1−x)2 σ22 +2ρσ1 σ2 x(1−x).

. (12.11)
1. When Conditional Value at Risk is used as a criterion for risk minimization, the
optimization problem can be written as

. min CV aRα Rp − E[Rp ] , (12.12)
x ∈ [0, 1];
E[Rp ] ≥ l
where l stands for the target on the portfolio return.

For μ1 > μ2 , the constraint in the minimization becomes x ∈ [0, 1] such that
x ≥ μl−μ 2
1 −μ2
. Hence, the minimization problem above reduces to

. min CV aRα Rp − E[Rp ] . (12.13)
x ∈ [0, 1];
x ≥ μl−μ 2
1 −μ2
−1
Since Rp − E[Rp ] ∼ N(0; V (Rp )) and CV aRα (Y ) = −m + s N (Nα (α)) for
Y ∼ N (m; s 2 ) (see, for instance, Barucci et al. [3]) with N (·) denoting the
density function of a standard normal, problem (12.13) becomes

N (N −1 (α))
. min x 2 (σ12 + σ22 − 2ρσ1 σ2 ) − 2x(σ22 − ρσ1 σ2 ) + σ22 .
x ∈ [0, 1]; α
x ≥ μl−μ 2
1 −μ2
(12.14)
It is then easy to check that the unconstrained minimum is attained at x̃ =

σ22 −ρσ1 σ2
. In our case, x̃ = 3
13 .
σ12 +σ22 −2ρσ1 σ2
Consider now l1 = 0.065. Since the constraints on x become x ∈ [0, 1] and
x ≥ μl11−μ2 1 ∗
−μ2 = 8 , the constrained minimization is obtained for x = 13 . The
3
3
optimal portfolio consists thus in investing in 13 shares of the first stock and in
10
13 shares of the second stock. The corresponding value of the variance of the
portfolio is V (Rp ) = 0.127 and, consequently, the Conditional Value at Risk is
CV aR0.01 (Rp − E[Rp ]) = 0.950.
When l2 = 0.08, instead, the constraints on x become x ∈ [0, 1] and x ≥
l2 −μ2 ∗
μ1 −μ2 = 0.5 and the constrained minimization is obtained for x = 0.5. The
optimal portfolio consists thus in investing in 0.5 shares of the first stock and in
0.5 shares of the second stock. The corresponding value of the variance of the
portfolio is V (Rp ) = 0.155 and, consequently, the Conditional Value at Risk is
CV aR0.01 (Rp − E[Rp ]) = 1.049. As one could expect, the minimal riskiness of
this last portfolio is greater than the other (we have indeed required an average
portfolio return greater than before).
2. Consider now the optimization problem studied above, where the criterion of risk
minimization of Conditional Value at Risk is replaced by that of Value at Risk or
of standard deviation.
Because of the normality assumption on the joint distribution of (R1 , R2 )
(hence, normality of Rp ), the V aR minimization reduces to

. min N −1 (1 − α) x 2 (σ12 + σ22 − 2ρσ1 σ2 ) − 2x(σ22 − ρσ1 σ2 ) + σ22 ,
x ∈ [0, 1];
x ≥ μl−μ 2
1 −μ2
(12.15)
while the minimization with respect to the standard deviation becomes

. min x 2 (σ12 + σ22 − 2ρσ1 σ2 ) − 2x(σ22 − ρσ1 σ2 ) + σ22 . (12.16)
x ∈ [0, 1];
x ≥ μl−μ 2
1 −μ2
By the arguments above and since N −1 (1 − α) > 0 for α = 0.01, in the present
case the optimal composition of the portfolio is the same with respect to the
CV aR criterion, to the V aR criterion and to the standard deviation criterion.
Indeed, the minimum in problems (12.13), (12.15) and (12.16) is always attained
at the same x ∗ . The minimal riskiness of the portfolio is, respectively, equal to
CV aR0.01 (Rp ) = 0.950

.
V aR0.01 (Rp ) = 0.828

σ (Rp ) = 0.356
for l1 = 0.065, while for l2 = 0.08
.CV aR0.01 (Rp ) = 1.049

V aR0.01 (Rp ) = 0.916
σ (Rp ) = 0.394.
Exercise 12.10 Consider a financial investment whose Profit and Loss in 2 years is
represented by the random variable:
⎧
⎪
⎪ 100, on ω1
⎪
⎪
⎪
⎨ 20, on ω2
.X = 0, on ω3 ,
⎪
⎪
⎪ −10,
⎪ on ω4
⎪
⎩ −40, on ω5
where the filtrated probability space (Ω, F , (Ft )t=0,1,2 , P ) is given by
Ω = {ω1 , ω2 , ω3 , ω4 , ω5 }
.
U = {ω1 , ω2 }
D = {ω3 , ω4 , ω5 }
F0 = {∅, Ω}
F1 = {∅, U, D, Ω}
F2 = P(Ω)
and
P (U ) = 0.8; P (D) = 0.2

. P (ω1 |U ) = 0.4; P (ω2 |U ) = 0.6
P (ω3 |D) = 0.6; P (ω4 |D) = 0.3; P (ω5 |D) = 0.1
(0) (0) (1)
1. Verify whether V aR0.05 (X) = V aR0.05 (−V aR0.05 (X)), where V aRα(t) (X)
stands for the Value at Risk of X at time t.1
(0) (0) (1)
2. Verify whether CV aR0.05 (X) = CV aR0.05 (−CV aR0.05 (X)), where
(t)
CV aRα (X) stands for the Conditional Value at Risk of X at time t.
3. Consider now the dynamic risk measure (ρt )t=0,1,2 defined as
ρt (X) ess sup EQ [ −X| Ft ],

. (12.17)
Q∈Q
1 We recall (see, among many others, [2]) that a dynamic risk measure is a family (ρt )t=0,1,2
defined on a space X of random variables such that ρt (X) is Ft -measurable, i.e. it takes into
account all the information available until time t. In particular, ρ0 (X) ∈ R. A risk measure is then
said to be time-consistent if ρ0 (−ρt (X)) = ρ0 (X) for any X ∈ X and t ∈ [0, T ].
where Q = {P , Q1 , Q2 , Q3 } and
Q1 (U ) = Q2 (U ) = Q3 (U ) = 0.8; Q1 (D) = Q2 (D) = Q3 (D) = 0.2

Q1 (ω1 |U ) = Q2 (ω1 |U ) = 0.5; Q1 (ω2 |U ) = Q2 (ω2 |U ) = 0.5
. Q3 (ω1 |U ) = 0.4; Q3 (ω2 |U ) = 0.6; Q1 (ω3 |D) = Q3 (ω3 |D) = 0.4
Q1 (ω4 |D) = Q3 (ω4 |D) = 0.4; Q1 (ω5 |D) = Q3 (ω5 |D) = 0.2
Q2 (ω3 |D) = 0.6; Q2 (ω4 |D) = 0.3; Q2 (ω5 |D) = 0.1
Verify whether ρ0 (X) = ρ0 (−ρ1 (X)) or not.

Solution By the evolution of information in time, we obtain the following tree:
ω1

U = {ω1 , ω2 }

ω2
·
. ω3

D = {ω3 , ω4 , ω5 } → ω4

ω5
−−− − − − − − − −− − − −−
0 1 year 2 years
Furthermore, it also follows that
P (ω1 ) = 0.32; P (ω2 ) = 0.48;

. P (ω3 ) = 0.12; P (ω4 ) = 0.06;
P (ω5 ) = 0.02.
1. In order to compute the Value at Risk of X at the initial time 0 we can just
consider the last nodes of the tree and the corresponding probabilities (evaluated
by means of P ). We deduce that
(0)
V aR0.05 (X) = 10.
. (12.18)
(1)
Let us compute now V aR0.05 (X), a random variable taking two values: one
(1,U )
at the node U (denoted by V aR0.05 (X)), the other at the node D (denoted by
(1,D)
V aR0.05 (X)).
(1,U )
Note that V aR0.05 (X) is the V aR of the random variable X on the sub-tree
starting from time t = 1 and node U . More precisely:
ω1

U = {ω1 , ω2 }
.
ω2
− − − − −− − − −−
1 year 2 years
or, in terms of X and P ,
X(ω1 ) = 100
prob.0.4
U
. prob.0.6
X(ω2 ) = 20
− − −− − − − − − − −−
1 year 2 years
(1,U )
It follows that V aR0.05 (X) = −20.
(1,D)
In order to compute V aR0.05 (X) we can proceed as previously by consider-
ing the sub-tree starting from time t = 1 and node D. More precisely:
ω3

D = {ω3 , ω4 , ω5 } → ω4
.
ω5
− − − − − − −− − − −−
1 year 2 years
X(ω3 ) = 0
prob.0.6
D →prob.0.3 X(ω4 ) = −10
. prob.0.1
X(ω5 ) = −40
− − −− − − − − − − −−
1 year 2 years
(1,D)
It follows that V aR0.05 (X) = 40.
(0) (1)
Finally, to find V aR0.05 (−V aR0.05 (X)) we proceed backwards once again by
considering the sub-tree starting from 0 and ending at nodes U, D:
U = {ω1 , ω2 }

·
.
D = {ω3 , ω4 , ω5 }
−−− − − − − − − −−
0 1 year
(1)
or, in terms of V aR0.05 (X) and P ,
(1,U )
−V aR0.05 (X) = 20 (with prob. 0.8)

·
.
(1,D)
−V aR0.05 (X) = −40 (with prob. 0.2)
−−− − − − − − − − − −−
0 1 year
Hence
(0) (1) (0)
V aR0.05
. (−V aR0.05 (X)) = 40 = 10 = V aR0.05 (X),
i.e. the dynamic version of V aR is not time-consistent.

2. Proceeding as above, we deduce that
(0) E[(−10 − X)+ ]

.CV aR0.05 (X) = + 10 = 22. (12.19)
0.05
(1) (1,U )
CV aR0.05 (X) is then a random variable taking two values: CV aR0.05 (X) at
(1,D)
node U, CV aR0.05 (X) at node D.
(1,U )
Note that CV aR0.05 (X) is the CV aR of the random variable X on the sub-
tree starting from time t = 1 and node U . More precisely:
ω1

U = {ω1 , ω2 }
.
ω2
− − − − −− − − −−
1 year 2 years
X(ω1 ) = 100
prob.0.4
U
. prob.0.6
X(ω2 ) = 20
− − −− − − − − −−
1 year 2 years
+ + ·0.6
(1,U )
It follows that CV aR0.05 (X) = (20−100) ·0.4+(20−20)
0.05 − 20 = −20.
(1,D)
Similarly, we obtain that CV aR0.05 (X) = 40 and
(0) (1) (0)

CV aR0.05 (−CV aR0.05 (X)) = 40 = 22 = CV aR0.05 (X),
. (12.20)
i.e. the dynamic version of CV aR is not time-consistent.

3. We compute now ρ0 (X) = supQ∈{P ,Q1 ,Q2 ,Q3 } EQ [−X]. It is easy to check what
follows:
ωi P Q1 Q2 Q3
ω1 0.32 0.4 0.4 0.32
ω2 0.48 0.4 0.4 0.48
ω3 0.12 0.08 0.12 0.08
ω4 0.06 0.08 0.06 0.08
ω5 0.02 0.04 0.02 0.04
Hence, we obtain
EP [−X] = −100 · 0.32 − 20 · 0.48 + 0 + 10 · 0.06 + 40 · 0.02 = −40.2

EQ1 [−X] = −100 · 0.4 − 20 · 0.4 + 0 + 10 · 0.08 + 40 · 0.04 = −45.6
.
EQ2 [−X] = −100 · 0.4 − 20 · 0.4 + 0 + 10 · 0.06 + 40 · 0.02 = −46.6
EQ3 [−X] = −100 · 0.32 − 20 · 0.48 + 0 + 10 · 0.08 + 40 · 0.04 = −39.2,
so
ρ0 (X) =
. sup EQ [−X] = −39.2. (12.21)
Q∈{P ,Q1 ,Q2 ,Q3 }
In order to find ρ1 (X), we proceed as in items 1.–2. so to obtain that ρ1 (X)

takes the following values at U and D, respectively,
ρ1U (X) = sup{−100 · 0.4 − 20 · 0.6;

−100 · 0.5 − 20 · 0.5; −100 · 0.5 − 20 · 0.5; −100 · 0.4 − 20 · 0.6}
= −52;
.
ρ1D (X) = sup{0 + 10 · 0.3 + 40 · 0.1; 0 + 10 · 0.4
+40 · 0.2; 0 + 10 · 0.3 + 40 · 0.1; 0 + 10 · 0.4 + 40 · 0.2}
= 12.
Proceeding backwards once more, we deduce that
ρ0 (−ρ1 (X)) = −39.2.

. (12.22)
By the arguments above and by (12.21), it follows that
ρ0 (−ρ1 (X)) = ρ0 (X) = −39.2.

.
Note that one can also prove that the dynamic risk measure (ρt )t=0,1,2 defined
in (12.17) is time-consistent. In such a case, indeed, the set Q contains any
probability measure obtained by “pasting” other probability measures of Q. Note
that stability under pasting is not always satisfied by the set of generalized scenarios
in the representation of CV aR.
Exercise 12.11 Consider a portfolio whose Profit & Loss is represented by a
random variable X that is distributed as a Uniform on the interval [−200, 300].
Compute the the V aR at 10% of the position. Furthermore, if there exists, find
the level α ∈ (0, 1) (and the corresponding scaling factor c with α = c · 0.1) for
which V aR at 10% and CV aR at level α of X coincide.
Note the scaling factor c is known as PELVE (Probability Equivalent Level of
VaR and ES) and studied in Li and Wang [29].
Solution It is well known that a Uniform random variable on the interval

[−200, 300] has the following cumulative distribution function
⎧
⎨ 0; if x < −200
FX (x) =
.
1
500 (x + 200); if − 200 ≤ x < 300 .
⎩
1 if x ≥ 300
The quantile at 10% of X then solves
. FX (q0.1 ) = 0.1
1
(q0.1 + 200) = 0.1,
500
implying that V aR0.1 (X) = −q0.1 (X) = 150 and, more in general, that
V aRα (X) = 200 − 500α.
It remains to deduce the level α such that CV aRα (X) = V aR0.1 (X). Since
X is a continuous random variable, CV aRα (X) coincides with T CEα (X) and,
consequently,

EP −X1{X≤−V aRα (X)}
CV aRα (X) = T CEα (X) =
.
P (X ≤ −V aRα (X))
−V aRα (X)
1 x
= − dx
α −200 500
2002 − (−V aRα (X))2
= .
1000α
Assuming that CV aRα (X) = V aR0.1 (X), α ∈ (0, 1) should then solve the
following equation
2002 − (−V aRα (X))2

. = 150
1000α
40,000 − (500α − 200)2 = 150,000α
250,000α 2 − 50,000α = 0
α = 0.2
The scaling factor c (or, better, the PELVE at 10% for X) for which V aR at 10%
and CV aR at level α = c · 0.1 of X coincide is then given by c = 0.1
α
= 2.2
2 This
result can be found in Li and Wang [29] who proved that any Uniform random variable has
PELVE equal to 2.
Exercise 12.12 Consider three stocks (or sub-portfolios) whose Profit & Loss (per
year) are represented, respectively, by the following random variables:
⎧
⎪
⎪ −200; on ω1 ⎧
⎪
⎪ ⎪ 400; on ω1
⎪
⎨ −400; on ω2 ⎪
⎨
0; on {ω2 ; ω3 }
.X1 = 0; on ω3 ; X2 = ;
⎪
⎪ ⎪
⎪ −200; on ω4
⎪ 800;
⎪ on ω4 ⎩
⎪
⎩ −400; −400; on ω5
on ω5
⎧
⎪
⎪ 0; on {ω1 ; ω2 }
⎨
−200; on ω3
X3 =
⎪ 2000;
⎪ on ω4
⎩
−800; on ω5
with P (ω1 ) = 0.01, P (ω2 ) = 0.09, P (ω3 ) = 0.8 and P (ω4 ) = P (ω5 ) = 0.05.
Consider now the whole portfolio X = X1 + X2 + X3 .
1. Compute the V aR at 5% of X.
2. Assume now we need to share the margin given by V aR0.05 (X) among the
different sub-portfolios. Compute the capital to be allocated to each sub-portfolio
by means of the marginal capital allocation, given by ρ(X) − ρ(X − Xi ) for sub-
portfolio Xi .
3. Compute the capital to be allocated to each sub-portfolio by means of the haircut
capital allocation at the level p = 5%, given by
V aRp (Xi )
.K · 3
i=1 V aRp (Xi )
for sub-portfolio Xi , where K denotes the total margin of the position

(V aR0.05 (X) in this case). Note that the haircut allocation is a particular case of
“proportional” capital allocations.
capital allocation at the level p = 1%.
Solution First of all, the Profit & Loss of the whole portfolio X is given by
⎧
⎪
⎪ −1600; on ω5
⎪
⎪
⎪
⎨ −400; on ω2
.X = X1 + X2 + X3 = −200; on ω3
⎪
⎪
⎪ 200;
⎪ on ω1
⎪
⎩ 2600; on ω4
1. It easily follows that V aR0.05 (X) = 400.

2. In order to compute the marginal allocations we need to evaluate the impact

(better, marginal contribution) of any sub-portfolio Xi on the whole portfolio X,
given by ρ(X) − ρ(X − Xi ).
Since
⎧
⎪
⎪ 400; on ω1
⎪
⎪
⎪
⎨ 0; on ω2
.X − X1 = X2 + X3 = −200; on ω3 ;
⎪
⎪
⎪
⎪ 1800; on ω4
⎪
⎩
−1200; on ω5
⎧
⎪
⎪ −200; on ω1
⎪
⎪
⎪
⎨ −400; on ω2
X − X2 = X1 + X3 = −200; on ω3 ;
⎪
⎪
⎪
⎪ 2800; on ω4
⎪
⎩ −1200; on ω
5
⎧
⎪
⎪ 200; on ω1
⎪
⎪
⎪
⎨ −400; on ω2
X − X3 = X1 + X2 = 0; on ω3 ,
⎪
⎪
⎪
⎪ 600; on ω4
⎪
⎩ −800; on ω
5
it follows that the capital to be allocated to each sub-portfolio with respect to the
marginal allocation is given, respectively, by
K̄1 = V aR0.05 (X) − V aR0.05 (X − X1 ) = 400 − 200 = 200

. for X1
K̄2 = V aR0.05 (X) − V aR0.05 (X − X2 ) = 400 − 400 = 0 for X2
K̄3 = V aR0.05 (X) − V aR0.05 (X − X3 ) = 400 − 400 = 0 for X3 .
3. Since V aR0.05 (X1 ) = 400, V aR0.05 (X2 ) = 200, V aR0.05 (X3 ) = 200 and the
total margin K = V aR0.05 (X) = 400, we deduce that the capital to be allocated
to each sub-portfolio with respect to the haircut allocation with p = 5% is given,
respectively, by
V aR0.05 (X1 ) 400

.K1 = K · 3 = 400 · = 200 for X1
i=1 V aR0.05 (Xi )
800
V aR0.05 (X1 ) 200
K2 = K · 3 = 400 · = 100 for X2
i=1 V aR0.05 (Xi )
800
V aR0.05 (X1 ) 200
K3 = K · 3 = 400 · = 100 for X3 .
i=1 V aR 0.05 (Xi ) 800
Note that the sum of Ki ’s as above is equal to the total margin V aR0.05 (X) = 400
to be deposited for X. This property is known as full allocation.
4. Proceeding as above, we obtain that V aR0.01 (X1 ) = 400, V aR0.01 (X2 ) =
400, V aR0.01 (X3 ) = 800. Since the total margin to be allocated is K =
V aR0.05 (X) = 400, the capital to be allocated to each sub-portfolio with respect
to the haircut allocation with p = 1% is then given, respectively, by
V aR0.01 (X1 ) 400

K1∗ = K · 3
. = 400 · = 100 for X1
i=1 V aR0.01 (Xi )
1600
V aR0.01 (X2 ) 400
K2∗ = K · 3 = 400 · = 100 for X2
i=1 V aR0.01 (Xi )
1600
V aR0.01 (X3 ) 800
K3∗ = K · 3 = 400 · = 200 for X3 .
i=1 V aR0.01 (Xi )
1600
Exercise 12.13 Consider two stocks (A and B) whose daily returns are jointly
normal. Assume that the correlation between their returns is −0.8, that stock A
has current price 2 euros, negligible daily drift and daily volatility of 1.2%, and that
stock B has current price 1 euro, negligible daily drift and daily volatility of 1.6%.
Rank the following financial positions, based on the riskiness evaluated with
respect to V aR at 2% and on a period of 1 day:
• 40 shares of stock A;
• 10 shares of stock B;
• a portfolio composed by 40 shares of A and by 10 shares of B.
Is diversification of risk encouraged in the present setting?
⎧
⎪
⎪ −4000; on ω1
⎨
−48; on ω2
.X =
⎪
⎪ 0; on ω3
⎩
640; on ω4
where P (ω1 ) = 0.02, P (ω2 ) = 0.08, P (ω3 ) = 0.8 and P (ω4 ) = 0.1.
1. Compute the Value at Risk at 5% and the Tail Conditional Expectation at 5%

of X.
2. In X, replace the value −4000 with x. Find the largest possible loss x for which
the margin to be deposited according to T CE at 5% does not exceed twice the
corresponding V aR.
Exercise 12.15 Consider two portfolios whose Profit & Loss are represented by the
following random variables:
⎧ ⎧
⎪
⎪ −160; on ω1 ⎪
⎪ −240; on ω1
⎨ ⎨
−20; on ω2 0; on ω2
.X = Y =
⎪
⎪ 0; on ω3 ⎪
⎪ 160; on ω3
⎩ ⎩
120; on ω4 −280; on ω4
where P (ω1 ) = 0.01, P (ω2 ) = 0.02, P (ω3 ) = 0.57 and P (ω4 ) = 0.40.
1. Compute the Value at Risk at 2% of X, of Y and of X + Y . Is diversification of
risk encouraged or not?
2. Compute the Conditional Value at Risk at 2% of X, of Y and of X + Y . Is
diversification of risk encouraged or not?
⎧
⎪
⎪ −800; on ω1
⎪
⎪
⎪
⎪ −200; on ω2
⎪
⎨
0; on ω3
.X =
⎪
⎪ 120; on ω4
⎪
⎪
⎪
⎪ 400; on ω5
⎪
⎩
1200; on ω6
with P (ω1 ) = 0.02, P (ω2 ) = 0.08, P (ω3 ) = 0.64, P (ω4 ) = 0.16, P (ω5 ) = 0.06
and P (ω6 ) = 0.04.
Let ρQ be the coherent risk measure generated by the set Q = {P , Q1 , Q2 } of
generalized scenarios, where
Q1 (ω1 ) = Q1 (ω2 ) = Q1 (ω3 ) = Q1 (ω4 ) = Q1 (ω5 ) = Q1 (ω6 ) = 1/6

.
Q2 (ω1 ) = Q2 (ω2 ) = 0.2; Q2 (ω3 ) = Q2 (ω4 ) = Q2 (ω5 ) = Q2 (ω6 ) = 0.15
Compute ρQ (X) and establish which between ρQ and the expected value of losses
is the stronger (or more conservative) risk measure.
⎧
⎪
⎪ −480; with prob. 0.02
⎪
⎪
⎪
⎨ −20; with prob. 0.04
.X = 0; with prob. 0.64
⎪
⎪
⎪ 40;
⎪ with prob. 0.2
⎪
⎩ 800; with prob. 0.1
1. Compute the Value at Risk at 10% and the Conditional Value at Risk at 10%
of X.
2. What is the sign of CV aR and V aR of a position having as P & L the one above
increased by an amount of 80 euros? Is it possible to obtain CV aR and V aR of
such a new position with one direct computation only?
Exercise 12.18 Consider the filtrated probability space (Ω, F , (Ft )t=0,1,2 , P )
given by
Ω = {ω1 , ω2 , ω3 , ω4 }
.
U = {ω1 , ω2 }
D = {ω3 , ω4 }
F0 = {∅, Ω}
F1 = {∅, U, D, Ω}
F2 = P(Ω)
and a financial investment whose Profit and Loss in 2 years is represented by the
random variable
⎧
⎪
⎪ 20, on ω1
⎨
10, on ω2
.X = ,
⎪
⎪ 0, on ω3
⎩
−30, on ω4
defined on the space above. Consider now the dynamic risk measures (ρtR )t=0,1,2
and (ρtS )t=0,1,2 defined as
ρtR (X) ess supQ∈R EQ [ −X| Ft ].

. (12.23)
ρtS (X) ess supQ∈S EQ [ −X| Ft ], (12.24)

where R = {P , Q1 , Q2 }, S = {P , Q1 , Q2 , Q3 } and the probability measures are

as follows:
event P Q1 Q2 Q3
ω1 |U 0.5 0.5 0.5 0.5
ω2 |U 0.5 0.5 0.5 0.5
ω3 |D 0.5 0.8 0.8 0.5
ω4 |D 0.5 0.2 0.2 0.5
U 0.5 0.4 0.5 0.4
D 0.5 0.6 0.5 0.6
Verify if ρ0R (X) = ρ0R (−ρ1R (X)) and if ρ0S (X) = ρ0S (−ρ1S (X)). Explain and
discuss the results obtained.
Exercise 12.19 Consider the P & L of a stock represented by the random variable
X distributed as a Uniform on the interval [−100, 400].
1. Compute the V aR of X at 10%.
2. Applying the Cornish-Fisher expansion, approximate the V aR of X at 10% by
using the skewness of X of a Uniform.
Exercise 12.20 Consider four stocks (or sub-portfolios) whose Profit & Loss (per
year) are represented, respectively, by the following random variables:
⎧
⎪
⎪ −200; on ω1 ⎧
⎪
⎪ ⎪ 400; on ω1
⎪
⎨ −400; on ω2 ⎪
⎨
0; on {ω2 ; ω3 }
.X1 = 0; on ω3 ; X2 = ;
⎪
⎪ ⎪
⎪ 200; on ω4
⎪
⎪ 800; on ω4 ⎩
⎪
⎩ 400; −400; on ω5
on ω5
⎧
⎪ 0; on {ω1 ; ω2 } ⎧
⎪
⎨ ⎨ 200; on {ω1 , ω3 }
−200; on ω3
X3 = ; X4 = 0; on {ω2 ; ω4 }
⎪
⎪ 1000; on ω4 ⎩
⎩ −400; on ω5
200; on ω5
with P (ω1 ) = 0.05, P (ω2 ) = 0.15, P (ω3 ) = 0.7 and P (ω4 ) = P (ω5 ) = 0.05.
Consider now the whole portfolio X = X1 + X2 + X3 + X4 .
1. Compute the V aR at 5% of X.
2. Assume now we need to share the margin given by V aR0.05 (X) among the
different sub-portfolios. Compute the capital to be allocated to each sub-portfolio
by means of the marginal capital allocation, given by ρ(X) − ρ(X − Xi ) for sub-
portfolio Xi .
capital allocation at the level p = 5%.
References
1. Artzner, P., Delbaen, F., Eber, J.M., Heath, D.: Coherent measures of risk. Math. Financ. 4,
203–228 (1999)
2. Artzner, P., Delbaen, F., Eber, J.M., Heath, D., Ku, H.: Coherent multiperiod risk adjusted
values and Bellman s principle. Ann. Oper. Res. 152, 5–22 (2007)
3. Barucci, E. : Financial Markets Theory. Springer-Verlag, London (2003)
4. Barucci, E., Marsala, C., Nencini, M., Sgarra, C.: Ingegneria finanziaria. Un’introduzione
quantitativa. Egea, Milano (2009)
5. Bates, D.: Jumps and stochastic volatility: the exchange rate processes implicit in
Deutschemark options. Rev. Financ. Stud. 9, 69–107 (1996)
6. Björk, T. : Arbitrage Theory in Continuous Time, 3rd edn. Oxford University Press, Oxford
(2009)
7. Black, F.: Fact and fantasy in the use of options. Financ. Anal. J. 31, 36–41, (1975)
8. Bollerslev, T.: General autoregressive conditional heteroskedasticity. J. Econ. 31, 307–327
(1986)
9. Brigo, D., Mercurio, F.: Interest rate models. Theory and Practice, 2nd edn. Springer-Verlag,
Berlin Heidelberg (2006)
10. Capiński, M., Zastawniak, T.: Mathematics for Finance, 2nd edn. Springer-Verlag, London
(2011)
11. Carr, P., Madan, D.: Option valuation using the fast Fourier transform. J. Comput. Financ. 2,
61–73 (1998)
12. Centrone, F., Rosazza Gianin, E.: Capital allocation à la Aumann–Shapley for non-
differentiable risk measures. Eur. J. Oper. Res. 267, 667–675 (2018)
13. Cheridito, P., Stadje, M.: Time-inconsistency of VaR and time-consistent alternatives. Financ.
Res. Lett. 6, 40–46 (2009)
14. Christoffersen, P., Elkamhi, R., Feunou, B., Jacobs, K.: Option valuation with conditional
heteroskedasticity and non-normality. Rev. Financ. Stud. 23, 2139–2183 (2010)
15. Delbaen, F. : Coherent risk measures on general probability spaces. In: Sandmann, K.,
Schönbucher, P.J. (eds.) Advances in Finance and Stochastics, pp. 1–37. Springer-Verlag,
Berlin/Heidelberg (2002)
16. Dhaene, J., Tsanakas, A., Valdez, E.A., Vanduffel, S.: Optimal capital allocation principles. J.
Risk Insur. 79(1), 1–28 (2012)
17. Duan, J.C.: The GARCH option pricing model. Math. Financ. 5, 13–32 (1995)
18. El Karoui, N.: Dynamic Pricing and Hedging of Derivatives: Stochastic and Numerical
Methods, 1st edn. Chapman and Hall/CRC, London/New York (2013)
La Matematica per il 3+2 149, https://doi.org/10.1007/978-3-031-28378-9
302 References
19. Engle, R.: Autoregressive conditional heteroskedasticity with estimates of the variance of the
UK inflation. Econometrica 50, 987–1008 (1982)
20. Fiori, A.M., Rosazza Gianin, E., Spasova, A.: Risk measures and Pareto style tails. In: Perna,
C., Sibillo, M. (eds.) Mathematical and Statistical Methods for Actuarial Sciences and Finance,
pp. 183–191. Springer-Verlag, Milano (2012)
21. Föllmer, H., Schied, A.: Stochastic finance. An Introduction in Discrete Time, 2nd edn. De
Gruyter, Berlin/New York (2004)
22. Frittelli, M., Rosazza Gianin, E.: Dynamic convex risk measures. In: Szegö, G. (ed.) Risk
Measures for the 21st Century, pp. 227–248. Wiley, New York (2004)
23. Gaivoronski, A., Pflug, G.C.: Value-at-risk in portfolio optimization: properties and computa-
tional approach. J. Risk 7, 1–31 (2005)
24. Heston, S.L.: A closed-form solution for options with stochastic volatility with applications to
bond and currency options. Rev. Financ. Stud. 6, 327–343 (1993)
25. Hull, J.: Options, Futures and other Derivatives, 8th edn. Pearson, London (2011)
26. Hull, J., White, A.: The pricing of options on assets with stochastic volatility. J. Financ. 42,
281–300 (1987)
27. Jorion, P.: Value at Risk. Mc Graw Hill, New York (2000)
28. Kalkbrener, M.: An axiomatic approach to capital allocation. Math. Financ. 15(3), 425–437
(2005)
29. Li, H., Wang, R.: PELVE: Probability equivalent level of VaR and ES. J. Econ. 234(1), 353–370
(2023).
30. Luenberger, D.G.: Investment Science. Oxford University Press, New York (1997)
31. Merton, R.: Option pricing when underlying stock returns are discontinuous. J. Financ. Econ.
3, 125–144 (1976)
32. Mikosch, T.: Elementary Stochastic Calculus with Finance in View. World Scientific, Singa-
pore (1998)
33. Musiela, M., Rutkowski, M.: Martingale Methods in Financial Modelling. Springer-Verlag,
Berlin (1997)
34. Øksendal, B.: Stochastic Differential Equations, 6th edn. Springer-Verlag, Berlin Heidelberg
(2003)
35. Pascucci, A.: PDE and Martingale Methods in Option Pricing. Springer, Bocconi University
Press, Milano (2011)
36. Pascucci, A. Runggaldier, W.J.: Financial mathematics. Theory and Problems for Multi-Period
Models. Springer, Milano (2011)
37. Pliska, S.R.: Introduction to Mathematical Finance: Discrete Time Models. Blackwell Publish-
ers, Oxford (1997)
38. Rockafellar, R.T., Uryasev, S.: Conditional value-at-risk for general loss distributions. J. Bank.
Financ. 26, 1443–1471 (2002)
39. Ross, S.M.: An Elementary Introduction to Mathematical Finance, 3rd edn. Cambridge
University Press, New York (2011)
40. Salsa, S.: Equazioni alle Derivate Parziali. Springer-Verlag, Milano (2004)
41. Stein, E., Stein, J.: Stock price distributions with stochastic volatility: an analytic approach.
Rev. Financ. Stud. 4, 727–752 (1991)
42. Tasche, D.: Allocating portfolio economic capital to sub-portfolios. In: Dev, A. (ed.) Economic
Capital: A Practitioner Guide, pp.275–302. Risk Books, London (2004)
43. Wilmott, P., Howison, S., Dewynne, J.: Option Pricing: Mathematical Models and Computa-
tion. Oxford Financial Press, Oxford (1993)
44. Wilmott, P., Howison, S., Dewynne, J.: The Mathematics of Financial Derivatives: A Student
Introduction. Cambridge University Press, Cambridge (1995)
Index
A Compound Poisson process, 256

Affine-term structure, 222 Conditional Value at Risk (CVaR), 271
Aggregational Gaussianity, 255 Contingent claim, 31
Arbitrage, 31 Coupon bond, 221
free of, 31, 65 option on, 227
opportunity, 60, 64
ARCH model, 257
Asset/nothing, 191 D
Attainable, 33, 64 Delta, 135
hedging, 32, 135
B neutral, 135
Beta, 20 Derivative
Binomial, 15 price, 32, 69
distribution, 2 Diffusion equation, 107
model, 31 fundamental solution, 120
process, 2 Diffusion parameter, 3
Black formula, 226 Dirac Delta distribution, 92
Black-Scholes Distribution
equation, 106 Bernoulli, 1
formula, 130, 132 binomial, 2
model, 131 Dirac Delta, 92
partial differential equation, 106 exponential, 2
Bond, 31 normal, 3
Brownian Pareto, 3
local time, 91 Poisson, 2
Brownian motion, 3 Dividend, 132
geometric, 5 continuous, 133
standard, 3 discrete, 134
rate, 133
Drift, 3, 131
C
Capital Asset Pricing Model (CAPM), 20
Caplet, 226 E
Central Limit Theorem, 3 Early exercise, 169
Characteristic function, 260 premium, 186
Completeness, 33
La Matematica per il 3+2 149, https://doi.org/10.1007/978-3-031-28378-9
304 Index
Efficient L
portfolio, 18 Leverage effect, 255
Efficient frontier, 20 Local time, 91
Entropy, 78 London InterBank Offer Rate (LIBOR), 226
criterion, 83 forward rate, 226
spot, 253
spot rate, 226
F
Fat tails, 255
Feynman-Kac Representation Theorem, 106 M
Filtration, 1 Market
generated by, 1 complete, 33, 64
Formula free of arbitrage, 31
of Black, 226 incomplete, 64
of Black-Scholes, 130, 132 portfolio, 20
of Itô, 90 Martingale, 5
of Tanaka, 91 measure, 32, 64
Forward measure, 226 Model
Fourier transform, 256 binomial, 31
Free boundary problem, 170 Black-Scholes, 131
Free lunch, 64 Ho-Lee, 223
Fundamental Theorem of Asset Pricing, 34, 64 Hull-White, 224
jump-diffusion, 256
with jumps, 256
stochastic volatility, 256
G
Vasiček, 223
Gamma, 135
Mutual Funds Theorem, 20
hedging, 135
neutral, 135
GARCH model, 257
N
No-arbitrage interval, 65
Numéraire, 225
H
change of, 225
Hedging strategy, 134
Ho-Lee model, 223
Hull-White model, 224
O
Occupation time, 91
Optimal stopping, 169
I Option, 34
Instantaneous forward rate, 221 American, 34, 169
Interest rate, 221 Asian, 192
instantaneous forward rate, 221 average rate, 192
short rate, 221 average strike, 192
Itô Barrier, 192
formula, 90 binary, 191
lemma, 90 cash/nothing, 191
stochastic integral, 89 chooser, 192
on a coupon bond, 227
European, 34, 132
J exotic, 191
Jump, 256 Lookback, 194
-size distribution, 257 path-dependent, 191
Jump-Diffusion model, 256 perpetual, 184
Index 305
price, 32, 65 Hull-White model, 224

on a zero-coupon bond, 224 Vasiček model, 223
Ordinary differential equation, 105 Similarity
method, 107
solution, 112
P Spread, 149
Partial differential equation, 105 bear, 150
backward parabolic, 106 bull, 166
elliptic, 106 butterfly, 150
forward parabolic, 106 Stationary, 266
hyperbolic, 106 Stochastic
parabolic, 106 differential equation, 90
semilinear, 105 integral, 89
solution, 105 process, 89
Portfolio, 17 volatility model, 256
efficient, 19 Stock, 31
of minimum variance, 19 Stop-loss strategy, 165
optimal, 25 Strategy
optimization, 18 hedging, 182
Price replicating, 32, 65
of a derivative, 32, 65 self-financing, 66
Process stop-loss, 165
adapted, 1 trading, 64
binomial, 2
Compound Poisson, 256
continuous-time, 1 T
discrete-time, 1 Tail Conditional Expectation, 271
log-normal, 4 Tanaka formula, 91
martingale, 5 Theta, 136
Poisson, 2 neutral, 136
stochastic, 1 Time series, 266
Put-Call Parity, 34, 133
U
R
Utility, 18
Random variable, 1
maximization, 18
Replicating strategy, 32, 65
Rho, 136
neutral, 136
V
Risk measures, 269
Value at Risk, 269
acceptance set, 272
Delta approximation, 270
coherent, 271
Vasiček model, 223
Conditional Value at Risk, 271
Vega, 136
dynamic, 287
neutral, 136
Tail Conditional Expectation, 271
Volatility, 131
time-consistent, 290
clustering, 255
Value at Risk, 269
implied, 255
Risk-neutral measure, 32, 64
model, 256
smile, 255
S
Self-financing, 66
Short rate, 221 Z
dynamics, 222 Zero-coupon bond, 221
Ho-Lee model, 223 option on, 224

Mathematical Finance: Emanuela Rosazza Gianin Carlo Sgarra

Uploaded by

Copyright:

Available Formats

Mathematical Finance: Emanuela Rosazza Gianin Carlo Sgarra

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Finance: Emanuela Rosazza Gianin Carlo Sgarra

Uploaded by

Copyright:

Available Formats

UNITEXT 149

Emanuela Rosazza Gianin

La Matematica per il 3+2

ISSN 2038-5714 ISSN 2532-3318 (electronic)

Milano, Italy Emanuela Rosazza Gianin

This exercise textbook is the result of several years of experience in teaching

Milano, Italy Emanuela Rosazza Gianin

1 Short Review of Probability and of Stochastic Processes . . . . . . . . . . . . . . 1

8 American Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

1.1 Review of Theory

Given a probability space .(Ω, F , P ), where .Ω denotes a non-empty set, .F a .σ -

The stochastic process .(Xt )t≥0 is said to be a discrete-time stochastic process

A filtration can be interpreted as the evolution of the information available up to

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

Setting .λ = νt (for .t ≥ 0), .ν can be understood as the rate or average number of

fT (t) = νe−νt , ∀t > 0.

The expected value and the variance of T are then equal to .E [T ] = 1

Pareto Random Variable

For .a > 1, X has finite expected value .E [X] = a

As a consequence, its expected value and variance are equal, respectively, to

Brownian motion is called standard and denoted by .(Wt )t≥0 .

By definition of Brownian motion it follows that, for any .0 ≤ s ≤ t,

For a standard Brownian motion we have

A standard Brownian motion can be obtained as limit of a suitable binomial

with mean 0 and variance .σ 2 t. In order to obtain a standard Brownian motion, it is

where .(Xt )t≥0 is a Brownian motion, is said to be log-normal. It follows, therefore,

1.2 Solved Exercises

Exercise 1.1 Let (Wt )t≥0 be a standard Brownian motion.

P (W6 − W2 < 0; W1 > 0) = P (W6 − W2 < 0) · P (W1 > 0) .

P (W6 − W2 < 0; W1 > 0) = P (W6 − W2 < 0) · P (W1 > 0)

and because of the properties of standard Brownian motions we get

2. We have to establish if there exists μ > 0 such that

P (X6 − X2 < 0; X1 > 0) ≥ 0.2.

P (X6 − X2 < 0; X1 > 0) = P (6μ + W6 − (2μ + W2 ) < 0; μ + W1 > 0)

= P (W6 − W2 < −4μ; W1 > −μ)

Recall that Wt ∼ N (0; t) for any t > 0. It follows that σ Wt is distributed as

E [Yt ] = E [μt + σ Wt ] = μt + σ E [Wt ] = μt

Hence, for any t > 0, Yt ∼ N μt; σ 2 t either for μ = 0 or μ = 0.

P (Y6 − Y2 < 0; Y1 > 0) = P (σ (W6 − W2 ) < −4μ; σ W1 > −μ)

P (Y6 − Y2 < 4σ ) = P (σ (W6 − W2 ) < −4μ + 4σ )

By the properties of Brownian motions, we also find

2. By substituting μ1 = 4μ2 and t = 4 in the expression of P St1 ≥ St2 just

Ti+1 − Ti ∼ Exp (λ) .

.E [T40 − T0 ] = E [T40 − T39 ] + E [T39 − T38 ] + . . . + E [T1 − T0 ]

Since n∗ = 4 (by item 1.), it follows that

P (Xn∗ + Yn∗ > 72) = 1 − P (X4 + Y4 ≤ 72)

It is also easy to check that

P (S1 = 9.6) = 0.4; P (S1 = 6.4) = 0.6

E [ S1 | S0 ] = S1u p + S1d (1 − p) = 9.6 · 0.4 + 6.4 · 0.6

it follows immediately that (Sn )n=0,1,2 is not a martingale with respect to P .

The first equality is equivalent to

hence S̃1u = S̃1d = 8. This implies that

Q (S1 = 9.6) = q; Q (S1 = 6.4) = 1 − q

We look now for q ∈ (0, 1) such that

By Eq. (1.4) it follows that

There remains therefore to check if such a q also satisfies equations (1.5)

Proceeding as above, we need to check

Since the previous system is equivalent to ûp + d̂ (1 − p) = 1, we find

Let X1 and X2 be two random variables defined as

Proceeding as above, we need to check