UNIVERSITY OF ZAGREB

FACULTY OF ELECTRICAL ENGINEERING AND

COMPUTING

MASTER THESIS num. 2748

Option Pricing and Hedging under

Jump-diffusion model
Ivan Almer
Mentor: Tomislav Burić

Zagreb, June 2022

TABLE OF CONTENTS

1. Introduction 1

2. Stochastic processes 2
2.1. Introduction to Probability Theory . . . . . . . . . . . . . . . . . . . 2
2.2. Brownian motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3. Stochastic Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4. Martingales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5. Stochastic Calculus and Itô’s Lemma . . . . . . . . . . . . . . . . . . 7
2.5.1. Geometric Brownian motion . . . . . . . . . . . . . . . . . . 8

3. Financial concepts 11
3.1. Financial instruments . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2. Arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3. Fundamental theorems of asset pricing . . . . . . . . . . . . . . . . . 15
3.3.1. First fundamental theorem of asset pricing . . . . . . . . . . . 15
3.3.2. Second fundamental theorem of asset pricing . . . . . . . . . 15

4. Pricing an asset 16
4.1. Poisson process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.1.1. Exponential distribution . . . . . . . . . . . . . . . . . . . . 17
4.1.2. Defining a Poisson process . . . . . . . . . . . . . . . . . . . 17
4.2. Pure Jump process . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2.1. Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3. Jump-diffusion process . . . . . . . . . . . . . . . . . . . . . . . . . 21

5. Pricing an option 24
5.1. Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.1.1. European Call and Put Options . . . . . . . . . . . . . . . . . 24
5.2. Itô’s Lemma for the Jump-diffusion Process . . . . . . . . . . . . . . 25

iv
 5.3. Constructing a Hedging Portfolio to price an Option . . . . . . . . . . 27
5.3.1. Delta hedge . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3.2. Delta hedge for the Jump-diffusion process . . . . . . . . . . 29
5.3.3. Employing the Captial Asset Pricing Model . . . . . . . . . . 31
5.4. Obtaining a formula for European Call Option . . . . . . . . . . . . . 34
5.5. Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

6. Hedging 40
6.1. Delta hedging in Black-Scholes world . . . . . . . . . . . . . . . . . 40
6.2. Quadratic hedging of contingent claims . . . . . . . . . . . . . . . . 44

7. Conclusion 49

Bibliography 50

List of Figures 51

A. Tables and derivations 54

A.1. Delta hedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
A.2. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

v
1. Introduction

Stochastic or random process is a mathematical object which is usually defined as

a collection of random variables. It can be seen as a random variable evolving over
time. The weather, for example, is a stochastic process. A lot of similar examples are
found in everyday life. Stochastic processes can be used to model any kind of process
that, in itself, has some kind of uncertainty involved.
Compared to a deterministic function, for which, at any time we know its value, in
case of a stochastic process one cannot for sure know its value, but can maybe have an
estimation or a probability of having a certain value. Although it is not ideal, it is for
sure better than not knowing anything about it whatsoever. Under the assumption that
processes in real world are random, but are to some degree defined, gives us hope that,
without knowing their exact future behavior or position, we can have a good feeling of
how they could move.
Stochastic processes have a very wide application in the field of finance. The main
reason for that is that they can be used to model an asset price as a process where
uncertainty is present. Behavior of such processes can be observed to potentially draw
some conclusions from them.
The goal of this thesis is to utilize stochastic processes and apply them to tackle
some of the problems from mathematical finance. Firstly, I will start with defining nec-
essary mathematical and financial concepts which will be used throughout this thesis.
After the foundation is set, starting with the Brownian motion, step-by-step we will
arrive at the jump-diffusion process which will be used as a process of the price of an
asset. We will then introduce the idea of a financial derivative, in our case an option,
and we will tackle the problem of determining a fair price of the option. Lastly, we will
introduce the idea of risk, how to quantify it and how to hedge the risk of our position
in the asset or the option. It is my hope that, this thesis will not only provide you with
a good overview of pricing and hedging of financial instruments, but also give you the
understanding of such the underlying processes are useful to model the scenarios that
occur every day in the market.

1
2. Stochastic processes

Stochastic processes have a well defined purpose in several fields, and one of those
fields is finance. They seem to be the perfect choice when it comes to modeling as-
sets’ dynamics, because their randomness is used to replicate the uncertainty of asset
prices that can be observed in the market. By utilizing this type of modeling it is pos-
sible to forecast probabilities of different outcomes under different conditions. Many
companies from different industries use stochastic processes to improve their business
workflows and increase profitability. Similarly, portfolio managers, analysts and plan-
ners in financial sector services use stochastic modeling to manage their assets and
liabilities as well as for portfolio optimization.

2.1. Introduction to Probability Theory

A probability space is a mathematical triplet (Ω, Σ, P) providing a formal model of
a random experiment. Such a space consists of three elements:
• Set of all possible outcomes Ω,
• σ-algebra of events Σ, where each event is a subset of Ω,
• a probability measure P.
We have mentioned a term σ-algebra, which a reader may not be familiar with. A σ-
algebra on a set Ω is a collection Σ of subsets of Ω which is closed under complement,
union and intersection. What this means is that if certain subsets A, B ⊆ Ω are ele-
ments of Σ, then simply also A, B, A ∪ B and A ∩ B have to be elements of Σ. This
has to hold for any two elements of Σ.
Another important concept is that of filtration. A filtration {Ft }t≥0 is an increasing
collection of σ-algebras, such that for every 0 ≤ s ≤ t the following holds: Fs ⊆ Ft ⊆
F. Filtration can be thought of as "flow of information". More specifically, Ft can be
considered as the information generated by all observed events up to time t.

2
Definition 1. A continuous time stochastic process {Xt : t ≥ 0} is a collection of
random variables indexed over time, often times simply X(t). The probability space of
these random variables is (Ω, Σ, P), where Ω is the set of all possible outcomes, Σ is
σ-algebra of events and P being the probability measure.

2.2. Brownian motion

The famous Brownian motion is indeed a stochastic process. The goal of this
chapter is to show how Brownian motion can be derived from a random walk process.
This is important to get a good understanding of what Brownian motion in fact is, as
we will use it very extensively throughout the this text.
To describe the process of a random walk we will use a sequence of independent
identically distributed (iid) random variables Xi with the following distribution:
!
h −h
Xi ∼ , (2.1)
p q
where Xi describes the move of the particle in the the i-th step and q = 1 − p. Let us
calculate the expected value and the variance of each step:

E[Xi ] = h(p − q) ,
V ar[Xi ] = E[Xi2 ] − E[Xi ]2 = h2 (p − q) − h2 (p − q)2
= h2 p + h2 q − h2 (p2 − 2p1 + q 2 ) (2.2)
2 2 2
= h p(1 − p) + h q(1 − q) + 2h pq
= 4h2 pq .
Now that we have the building blocks of the random walk, let us define the state of
the random walk after n steps as a random variable X(t) = ni=1 Xi . Where t is the
P

length of the interval [0, t]. This interval is divided into n equal intervals of length ∆t,
so we have:
t
∆t = . (2.3)
n
We calculate the expected value and the variance of this new random variable:

E[X(t)] = nh(p − q),

(2.4)
V ar[X(t)] = ni=1 V ar[Xi ] = 4nh2 pq .
ind. P

According to the Central Limit Theorem with n → ∞ we know that X(t) will have a
normal distribution, namely:
Pn Pn
i=1 Xi − nh(p − q) Xi − nh(p − q)
√ p = i=1 √ √ ∼ N (0, 1)
n V ar[Xi ] 2h n pq

3
 n
X
⇒ Xi − nh(p − q) ∼ N (0, 4nh2 pq)
i=1
n
X
Xi ∼ N (nh(p − q), 4nh2 pq) . (2.5)
i=1
t
We substitute n = ∆t
:
X(t) ∼ N (µt, σ 2 t) , (2.6)
h(p − q)
µ := lim , (2.7)
∆t→0,h→0 ∆t
2 4h2 pq
σ := lim . (2.8)
∆t→0,h→0 ∆t

Here µ is called the drift and σ 2 the diffusion of the process X(t). We can have a look
p √
at the special case where p = q = 0.5 and h = t/n = ∆t and it is easy to see that
then X(t) has the following distribution:

X(t) ∼ N (0, t) . (2.9)

The above mentioned process is called Wiener process and is often denoted by W =
{Wt : t ≥ 0} where Wt ∼ N (0, t). The formal definition is the following:

Definition 2. A Wiener process is the process with the following properties:

1. W0 = 0 ,

2. Wt is continuous in t ,

3. Wt has independent increments, i.e. for every r < s ≤ t < u increments

(Wu − Wt ) and (Ws − Wr ) are independent random variables,

4. W has Gaussian increments, i.e. for every u, t ≥ 0:

(Wt+u − Wt ) ∼ N (0, u) . (2.10)

2.3. Stochastic Integrals

As already stated in the earlier chapters, one of the objectives of this book is to
study asset pricing in financial markets using stochastic processes. We want to model
the price as a continuous time stochastic process and for this case the most complete
and elegant theory is obtained if we use diffusion processes and stochastic differen-
tial equations as main building blocks. Although we already mentioned diffusion a

4
couple of times, we have not really intuitively explained what it is. Loosely speaking,
a stochastic process is a diffusion if its local dynamics can be approximated with the
following stochastic difference equation:

Xt+h − Xt = µ(t, Xt )∆t + σ(t, Xt )∆Wt , (2.11)

where ∆Wt is defined as:

∆Wt = Wt+h − Wt . (2.12)

If we let h tend to 0 we can write the above equation like this:

dXt = µ(t, Xt )dt + σ(t, Xt )dWt , (2.13)

which can be interpreted as a shorthand for the folowing integral equation:

Z t Z t
Xt = X0 + µ(s, Xs )ds + σ(s, Xs )dWs . (2.14)
0 0

The ds integral can be viewed as a regular Riemann integral, whereas for the dWt
integral we have to introduce a new concept - Itô integral. In order to be able to
construct a stochastic integral of form:
Z t
gs dWs , (2.15)
0

we have to impose some kind of integrability condition on another stochastic process

gs .

Definition 3.

(i) We say that the process gs belongs to the class L2 [a, b] if the following holds:
Rb
• a E[gs2 ]ds < ∞ ,
• The process gs is adapted to filtration FtW .

(ii) We say the process g belongs to the class L2 if g belongs to L2 [0, t] for every t .

We will show how the define the stochastic integral for the case of gs which is simple,
i.e. such gs that there exist deterministic points in time a = t0 < t1 < ... < tn = b,
such that g is constant on each subinterval. To put this formally, gs = gtk where
s ∈ [tk , tk+1 ). In that case, the stochastic integral can be defined as follows:
Z b n
X t
gs dWs = lim gtk −1 (Wtk − Wtk−1 ) , where tk = k .
a n→∞
k=1
n

5
The process gs is evaluated in the summation on the left-hand point (tk−1 ) which is
known as a non-anticipatory integration. This is a natural choice in finance ensuring
that we use no information from the future for our present actions.
For the case of gs which is not simple (as described above), we would have to approx-
imate the process gs in a certain way, but this is outside of the scope of this thesis and
will not be considered because we will not encounter such integrals.

2.4. Martingales
Theory of stochastic integration is closely related to the theory of martingales.
Moreover, it is a foundation of the modern theory of financial derivatives. It would
be unreasonable to not mention this topic, because of its extreme importance in the
field. In discrete time, we say a stochastic process is a martingale if for any time n the
following properties hold:

• E[|Xn |] < ∞ ,
• E[Xn |X0 , ..., Xn−1 ] = Xn−1 .

Where the first condition is more of a technical condition and the second one is the
most important one. The second condition states that if at any time t = n we look
at the expected value of the process at any time t > n, it is equal to the value of the
process at time t = n. A process that is a martingale can be viewed as a model of a
fair game.
In order to extend the definition of a martingale to a continuous case, let us first con-
sider a concept of filtration Ft≥0 , that we have already mentioned earlier. As before, a
filtration can be thought of as a flow of information and Ft as the information generated
by all observed events up to time t. For any random variable X let the symbol

E[X|Ft ]

represent the "expected value of X given the information available up to time t". It is
important to note here that for a fixed t, the object E[X|Ft ] is a random variable. Let,
for example, a filtration be generated by the process Y , then the information available
up to time t will, of course, be dependent on the behavior of the process Y in the time
interval [0, t]. Before defining a continous time martingale, let us first consider the
following proposition:

6
 • If Y and Z are random variables, and Z is Ft -measurable (known at time t),
then
E[Z · Y |Ft ] = Z · E[Y |Ft ] .

• If Y is a random variable and s < t, then

E[E[Y |Ft ]|Fs ] = E[Y |Fs ] .

Definition 4. A stochastic process X is an (Ft )-martingale if the following conditions

hold:
• X is adapted to filtration {Ft }t≥0 ,
• For all t we have
E[|Xt |] < ∞ ,

• For all s and t, where s ≤ t we have

E[Xt |Fs ] = Xs .

The first condition above is a condition saying that the process Xt is observable (known)
at time t. The second one, like in discrete case, is just a technical condition. Also like
in discrete case, the third condition is the important one, which states that the expected
future value of the X, given the information available today, is equal to today’s ob-
served value of X. In other words, a martingale has an absence of drift.

2.5. Stochastic Calculus and Itô’s Lemma

When talking about stochastic processes, one can see that they are different than
the usual functions. If we look at a infinitesimally small time interval ∆t and observe
the function behaviorin that time interval, the function is a straight line. This is unfor-
tunately not the case with the stochastic processes, where even if we zoom-in on the
trajectory, the trajectory is not smooth. Therefore, a japanese mathematician Kiyoshi
Itô started and created the field of stochastic calculus which is able to operate on
stochastic processes. We come to the very important theorem in the field of stochastic
calculus and that is Itô’s lemma.

Theorem 1 (Itô’s Lemma). Assume that the process X has a stochastic differential
given by
dXt = µt Xt dt + σt Xt dWt ,

7
where µ and σ are adapted processes, and let f be a C 1,2 -function. Define the process
Z by Z(t) = f (t, Xt ). Then Z has a stochastic differential given by

1 2 2 ∂ 2f
 
∂f ∂f ∂f
df (t, Xt ) = + µt Xt + σt Xt 2 dt + σXt dWt .
∂t ∂x 2 ∂x ∂x

A proof of this theorem is outside the scope of this thesis and it is rather cumbersome.
The power and importance of this theorem is that we can define the differential of a
function of a stochastic process. This will prove to be an indispensible tool when we
will be pricing an option, as the price of the option depends on the price of an asset,
which means it is a function of the asset price.
We also introduce an alternative, more general, form of Itô’s lemma which we will
use in the following section to obtain an expression for the process governed by the
geometric Brownian motion:

∂f ∂f 1 ∂ 2f
df = dt + dXt + (dXt )2 . (2.16)
∂t ∂x 2 ∂x2

2.5.1. Geometric Brownian motion

Let us now show how to apply Itô’s formula for the case of geometric Brownian
motion. Before we start, it is crucial to first describe what kind of process that is.

Definition 5. A process X is said to be a geometric Brownian motion if it satisfies the

following stochastic differential equation:

dXt = µt Xt dt + σt Xt dWt . (2.17)

In the equation above µt and σt are drift and diffusion (respectively) of the process X.
Wt is the basic Brownian motion, that is, Wt ∼ N (0, t). Let us now consider a function
f (t, Xt ) = ln(Xt ) which, of course, yields another stochastic process. We will employ
an alternative form of Itô’s lemma found in the equation 2.16. If we expand the (dXt )2
we get:

(dXt )2 = (µt Xt dt + σt Xt dWt )2 = µ2t Xt2 (dt)2 + 2µt σt Xt2 dtdWt + σt2 Xt2 (dWt )2 ,
(2.18)
where we use the formal multiplication table:



 (dt)2 = 0,
dt · dWt = 0, (2.19)

 (dW )2 = dt

t

8
so we have:
(dXt )2 = σt2 Xt2 dt . (2.20)

Before we apply the Itô’s formula to this problem, let us first calculate the partial
derivatives that appear in the formula:

∂ ln(Xt ) ∂ ln(Xt ) 1 ∂ 2 ln(Xt ) 1

=0 , = and 2
=− 2 . (2.21)
∂t ∂x x ∂x x
Now we are finally ready to apply the formula. We will step by step derive the formula
for Xt :
1 1 1
df = 0 · dt + dXt − (dXt )2 ,
Xt 2 Xt2
1 1 1 2 2
df = (µt Xt dt + σt Xt dWt ) − σ X dt ,
Xt 2 Xt2 t t
1
df = (µt − σt2 )dt + σt dWt , (2.22)
2
and if we integrate it over the time interval [0, t] we get
Z t Z t Z t
1 2
df (s) = (µs − σs )ds + σs dWs . (2.23)
0 0 2 0

For simplicity purposes, let us assume that µ and σ are constants. Then they can come
outside of the integral and the resulting formula is
1
ln(Xt ) − ln(X) = (µ − σ 2 )t + σ(Wt − W0 ) . (2.24)
2
Remembering that the W0 = 0 we get the following result:
Xt 1
ln = (µ − σ 2 )t + σWt ,
X0 2
 
1 2
Xt = X0 exp (µ − σ )t + σWt . (2.25)
2
This model is in finance also known as log-normal asset return model, because we
are using logarithmic prices. It is important to note that this model only has positive
values of stock prices, which is in that sense in line with the real world, where stock
prices can not become negative. Trajectories of four different realisations of such a
model can be observed on the Figure 2.1. The parameters in this case were set to be
µ = 0.05 and σ = 0.4.

9
 Geometric Brownian motion (µ=0.05, σ = 0.4)

200

180

160

140
Price

120

100

80

60

40
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t

Figure 2.1: Geometric Brownian motion paths for parameters µ=0.05 and σ = 0.4

10
3. Financial concepts

Let us start by defining what a market is. Market is a place where parties can
meet to engage in an economic transaction. Usually, while only two parties are needed
to make a trade, at minimum one more party is needed to introduce competition and
bring balance to the market. A market in a state of perfect competition is charac-
terized by a high number of active buyers and sellers. Markets vary for a number of
reasons, including the types of products sold, size, location etc. One market which is
of particular interest to us is the financial market where currencies, bonds and various
types of securities are traded. They form capital and provide liquidity for businesses.
Stock exchanges like New York Stock Exchange (NYSE) or Nasdaq are one type of
financial markets. Other types of finincial markets include, for example bond markets
and foreign exchange (FX) markets.
As already mentioned, markets in perfect competition have a high number of par-
ticipants (buyers and sellers) and market determines the prices of goods and other
servides traded there. Prices are determined by supply and demand, where supply is
created by the sellers and demand by the buyers. The law of supply and demand is a
theory that explains the interaction between the sellers of a resource and the buyers for
that resource. In general, if the price for a good decreases, more people are willing to
buy it and less people are willing to sell. That is because the opportunity cost for the
buyers decreases, as they can obtain the good at a lower price than earlier, whereas the
opportunity cost for the sellers increases, as they earn less by selling the same amount
of goods as before. Although it is one of the most basic economic laws, it is a part of
almost all economic principles. The willingness of people to buy or sell a good deter-
mines the market equilibrium price at which the quantity of goods people are willing
to sell equals the quantity of goods people are willing to buy.

11
3.1. Financial instruments
In this section we will cover a few basic financial instruments. As discussed earlier,
different types of markets offer different types of goods. A bond market (also called
fixed-income markets) is a collective name attributed to all issues and trades of debt
securities. Stock exchange is another type of market where, among many other things,
stocks and derivatives can be traded.

A bond is a fixed-income financial instrument typically issued by governments or

corporations in order to raise money needed to fund a certain project, for example.
When a corporation (or a government) needs to borrow money, it issues bonds that
include the terms of the loan, the time at which the loaned funds (bond principal) must
be paid back (maturity date) and the interest payments that will be paid. They can
essentially be thought of as a contract issued by the borrower promising to pay back
the loan plus some interest on it at some time in the future. An interest is a premium
for the person loaning the money, because by buying a bond (loaning money) the buyer
takes on a risk that the issuer will not be able to pay back at maturity (risk of default).
Governments are typically less likely to fail than corporations, therefore the interest
paid on government bonds is generally lower than the interest paid on corporate bonds.
There are also bonds that pay additional coupons between the issuance and maturity,
but they will not be of particular interest to us in the scope of this thesis.

A stock (also called equity) is a type of a security issued by the corporations and
represents the ownership of a fraction of a company. The units of stock are called
shares. They are issued for a company to raise funds to operate their business. It is
important to point out that corporations are treated as legal persons, so a shareholder
does not own a corporation, but own shares issued by the corporations and have a claim
on its assets and earnings.

Having explained the characteristics of bonds and stocks we arrive at securities

that are not completely basic. Derivatives are a type of security which derives its
value from another asset (e.g. stock). There are many types of derivatives, but we will
focus on the ones used in this thesis and those are options. Option is a contract that
gives the buyer a right, but not an obligation, to buy or sell (depending on the type of
the contract they hold) the underlying asset at a certain time in the future.

12
 European options (also called vanilla options) are the most basic form of an op-
tion. We differentiate two different types of European options: call and put option. A
call option gives a buyer a right to buy the underlying security at some predetermined
price (strike price) at a certain time in the future (maturity). In case the price of the un-
derlying asset is greater than the strike price at the time of maturity, the buyer can buy
the underlying asset at the strike price and immediately sell the asset at its market price
which is now higher than the strike price, thus making a profit. However, if the price
of the underlying asset is lower than the strike price at maturity, it does not make sense
for the buyer to exercise the option and buy the asset at strike price, because she/he
can simply buy the asset at the market price. A put option works in an opposite way,
as it gives a buyer a right to sell the underlying security at the strike price at maturity.
Put option expires worthless in case the price of the underlying asset is greater than
the strike price as the buyer can than simply sell the security at its market price. In
this case an option expires without exercising, the only loss made by the buyer is the
amount paid for the option in the first place. To illustrate this very important concept
of European options let us look at the following example:

Suppose at time t = 0 the price of an asset is S0 = 100 and that we can buy a put
option with maturity T and strike price K = 90. The price of such option contract at
time t = 0 is P = 5. Let us assume that at maturity the price of the asset is ST = 93.
This means that the buyer will simply let the contract expire because the asset can
be sold at a market price which is higher than the strike price of the put option. If,
however, the price of an asset dips below S0 to value of ST = 80, the buyer can buy the
asset in the market for the price of 80 and immediately sell it in the market for K = 90
by exercising the option. Thus, making a profit of (K − ST ) − P = (90 − 80) − 5 = 5.
Note that in this example, the time value of money was disregarded for simplicity
purposes.

3.2. Arbitrage
In finance and economics, arbitrage is the practice to take advantage of market
inefficiencies to make a risk-free profit. One such inefficiency can be, for example,
the difference in prices of the same asset on two different markets. In other words,
an arbitrage-free market is a market where there does not exist such trade (or a set of
trades) which would allow a trader with zero capital at time t = 0 to make a positive
profit at some time in future t = T with positive probability. In practice, arbitrage

13
opportunities exist, but are very hard to spot, they get almost immediately exploited
and the opportunity is gone, often in the matter of seconds.
Here are two examples to better understand this very important principle:

Example 1

Suppose that the stock XYZ is trading for $50 on the NYSE and the same stock
is trading at $49.5 at the LSE (London Stock Exchange). The trader can then buy the
stock at LSE for $49.5 and immediately sell it at NYSE for $50, thus realizing profit
of $0.5.

Example 2

Let’s see a more complex example. Suppose we have a stock XYZ with spot price
S0 = 100 and in one year period there are only two possibilities for the stock move-
ment: it can go up to S1 = u ∗ S0 where u = 1.15, or it can go down to S1 = d ∗ S0
where d = 0.9. The risk-free rate in the market is constant during this one-year period
and it is r = 1%. The probability that the stock will go up in that period is p = 0.4.
The inconsistency is not easy to spot here. If the market is arbitrage-free, the
expected value of the future payoff should be zero. By looking more closely, and con-
sidering the time value of money one can find the following arbitrage strategy: We
decide to short-sell the stock today and put that money in the bank to be remunerated
at the risk-free rate of 1%. After one year we take out our money from the bank and
now we own the value of (1 + r)S0 = 101. We have to buy back the stock that we
shorted one year earlier to cover this short position. There are two possible scenarios:

• The stock went up and our profit in that case is: (1 + r)S0 − uS0 = -14
• The stock went down and our profit in that case is: (1 + r)S0 − dS0 = 11

Although at the first glance this does not look like a good deal, we have to take into
account the probabilities of these events happening. More specifically, let us look at
the expecte payoff at time t = 1:

E[payoff at t = 1] = 0.4 · (−14) + 0.6 · 11 = 1

14
3.3. Fundamental theorems of asset pricing
Whenever there is an arbitrage opportunity there is one side that is losing money.
It is logical to expect that a party will try avoid creating an arbitrage opportunity for
others to exploit. Modeling a market to be arbitrage free is thus a natural way forward.
The following theorems will provide conditions for a market to be free of arbitrage and
for a market to be complete. They form a sort of a bridge between the economical and
mathematical understanding of the market.

3.3.1. First fundamental theorem of asset pricing

Theorem 2. A discrete market, on a discrete probability space (Ω, Σ, P), is arbitrage-
free if, and only if, there exists at least one risk neutral probability measure that is
equivalent to the original probability measure, P.

This first theorem is important in that it ensures a fundamental property of market

models. It will allow for some nice manipulations with discounted price processes we
will mention later in this thesis.

3.3.2. Second fundamental theorem of asset pricing

Theorem 3. An arbitrage-free is said to be complete (every contingent claim replica-
ble) if and only if there exists a unique risk-neutral measure equivalent to the physical
one.

This theorem can be reformulated in an equivalent way, that is, the market is complete
if and only if there exists only one source of randomness in asset dynamic. Although
this property is common in many models it is sometimes not always considered desir-
able or realistic.

15
4. Pricing an asset

The famous Black-Scholes model assumes that the price of the asset follows the
geometric Brownian motion we have just covered. Geometric Brownian motion satis-
fies a local Markov property, that is, the change in the value of the process in a very
small amount of time can only be marginal. The distribution of returns of the geomet-
ric Brownian motion is not completely comparable to the real world scenario. To be
more specific, in real world from time to time there is a shock hitting the market and
the prices move rapidly in some direction. Such movements are called jumps. The
problem with the geometric Brownian motion is that it does not allow for such events
to happen, therefore yielding a return distribution with thinner tails than what would
be the case in real markets. It is of interest to us to try and model such events. The idea
was previously introduced in [8]. In this section we will describe the building blocks
of such process and arrive to the dynamics which have to be satisfied for the process to
be a Jump-diffusion process.

4.1. Poisson process

Poisson process and exponential distribution are very closely linked so let us have a
look at exponential distribution first. Poisson process has some convenient mathemat-
ical properties, which has led to it being used as a mathematical model for seemingly
random processes in various disciplines, including, of course, economics. It is named
after French mathematician Simeon Denis Poisson despite the fact that he has never
studied this process. The reason why it is called a Poisson process is that if a collection
of points in some space forms a Poisson process than the number of those points in a
finite size region is a random variable with a Poisson distribution. This will be step by
step explained in next sections.

16
4.1.1. Exponential distribution
As mentioned earlier, exponential distribution is very closely linked to the Poisson
process as the time between events in Poisson process has exponential distribution. It
is a continuous analogue of the geometric distribution.

Definition 6. We say that a random variable X is exponentially distributed if its prob-

ability density function (PDF) is:
(
λe−λx , x ≥ 0
f (x) = , (4.1)
0 ,x < 0

where λ > 0 is the parameter of the distribution, often called the rate parameter. The
notation to indicate that a random variable is exponentially distributed is

X ∼ ε(λ) .

The expected value of a random variable X which is exponentially distributed is

1
E[X] = . (4.2)
λ
The most famous property of the exponential distribution is memorylessness. Memo-
rylessness is a property of certain probability distributions which refers to the fact that
the waiting time until the occurrence of a certain event does not depend on how much
time has already elapsed. Memorylessness of an exponential distribution is formally
expressed as:
P (X > t + s|X > s) = e−λt = P (X > t) . (4.3)

If X is a random variable modeling a waiting time, this would mean that the probability
that we wait more than t + s if we already know that at least s time has elapsed, is the
same as the probability that we will wait more than t.

4.1.2. Defining a Poisson process

A Poisson process registers the occurrences of the certain event in a finite time
interval. It notes the number of occurrences as well as times when the event happened.

Definition 7. Poisson process (Nt , t ≥ 0) is given by the following conditions:

1. N0 = 0

2. N has independent increments

17
 3. Random variable N (s, t) = Nt −Ns where 0 ≤ s ≤ t, has a Poisson distribution
with parameter λ(t − s), that is,
[λ(t − s)]n
P (Nt − Ns = n) = e−λ(t−s) . (4.4)
n!

If we would like to show the trajectory of such a process we have to simulate it. One
way to simulate it is through simulating the waiting times between arrivals as we know
that the waiting time is exponentially distributed with parameter λ. Trajectory simula-
tion of one Poisson process is shown on the Figure 4.1.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure 4.1: Poisson process path with deterministic increment of 1 and parameter λ = 4

There is an alternative definition of a Poisson process which we will use when we

will be defining a jump-diffusion process. As we already mentioned before, Poisson
process registers occurrences of a certain event on a finite time interval. If we let that
interval to become very small and denote it by h, we denote the probabilities of the
number occurrences by:

P (N (h) ≥ 2) = o(h) ,
P (N (h) = 1) = λh , (4.5)
P (N (h) = 0) = 1 − λh + o(h) ,

18
Where o(h) is infinitesimally small amount, that is, some function with with the prop-
erty:
o(h)
lim =0. (4.6)
h→0h
Then we can let dN denote an increment in the Poisson process on an infinitesimally
small time interval dt. dN is the given by
(
1 with probability λdt
dN = . (4.7)
0 with probability (1 − λdt)

4.2. Pure Jump process

4.2.1. Definition
Now that the Poisson process has been explained, it will be easy to understand
the Jump process. Like we stated earlier, the Poisson process registers occurrences of
certain events. When talking about a jump process those events that occur will be the
jumps. On each such event the value of our Jump process will be increased by a ceratin
value (it will "jump"). The initial value of the Jump process is J(0) = 0. The size of
the i-th jump will be a random variable Ui independant of the Poisson process, that
is, N (t)⊥Ui , ∀i. All Ui are independent and identically distributed random variables
(iid).

Definition 8. We define a Jump process as

N (t)
X
J(t) = Ui , (4.8)
i=1

assuming J(0) = 0. The Jump process is right-continuous.

On the Figure 4.2 we can observe trajectories of different pure jump process realiza-
tions. The process we are observing has λ = 6 and a jump amplitude drawn from a
normal distribution U ∼ N (µ = 0, σ 2 = 0.04).
We will use Qj and Qu to indicate the probability measure governing the jump
arrival times and the probability measure governing the jump sizes, respectively. We
assume independence of the mentioned probability measures. Let us now have a look
at the expected value of the Jump process we have just defined:

N (t)
X
Qj Qu
E[J(t)] = E E Ui . (4.9)
i=1

19
 0.5

0.4

0.3

0.2

0.1

−0.1

−0.2

−0.3

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

Figure 4.2: Pure Jump Process with λ = 6 and jump size U ∼ N (µ = 0, σ 2 = 0.04)

When performing the calculation we have to have in mind that EQu [Ui ] = k:

∞
" n
# ∞ X
n
X X X
Qu
E[J(t)] = E Ui P (N (t) = n) = EQu [Ui ] P (N (t) = n)
n=0 i=1 n=0 i=1
∞ n ∞ n
X (λt) −λt X (λt) −λt
= nk e = nk e
n=0
n! n=1
n! (4.10)
∞ n−1 ∞ n
X (λt) (λt) X
= ke−λt λt = kλte−λt
n=1
(n − 1)! n=1
n!
= kλt .

As we can see the the expected value of the Jump process at time t is different than 0
which will subsequently lead to the conclusion that the process J(t) is not a martingale.
Sometimes it is useful to work with a Jump process which, in turn, is a martingale.
Such process is called a compensated Jump process.

20
 ˜ and let it be
Proposition 1. Let us denote a compensated Jump process with J(t)
defined by:
˜ := J(t) − kλt .
J(t) (4.11)

Such a process is a martingale.

Proof. Let us assume we that we have 0 ≤ s ≤ t and we determine the expected value
˜ given that we have information up to time s:
of the J(t)

˜
E[J(t)|F ˜ ˜ ˜
s ] = E[J(t) + J(s) − J(s)|Fs ]

˜ − J(s)|F
= E[J(t) ˜ ˜
s ] + E[J(s)|Fs ]

˜ − s)] + J(s)
= E[J(t ˜ (independent increments)
˜
= E[J(t − s) − kλ(t − s)] + J(s)
˜
= kλ(t − s) − kλ(t − s) + J(s)
˜ .
= J(s)

Before we continue further and combine the knowledge we have on geometric Brow-
nian motion and Jump processes to obtain a Jump-diffusion process, let us first denote
with dJ˜ the change in the compensated Jump process in an infinitesimally small time
interval dt: (
Ui − kλdt with probability λdt
dJ˜ = . (4.12)
−kλdt with probability (1 − λdt)

4.3. Jump-diffusion process

This has led us to the final part where we will obtain the final process of our asset
price. This process will be the combination of the geometric Brownian motion and a
(compensated) Jump process. The reason to use a Jump-diffusion process as a process
which will govern the price of an asset is because it better mimics the real-world sit-
uation where unexpected jumps can happen. In the previously defined jump process
the jump sizes were iid random variables Ui . We wish to make the jumps proportional
to the value of the geometric Brownian motion at that point in time. Formally the
differential form of a jump-diffusion process will be:

dXt = µt Xt dt + σt Xt dWt + Xt U dNt , (4.13)

21
or also,
dXt
= µt dt + σt dWt + U dNt . (4.14)
Xt
We will assume, for simplicity purposes, that the µ, σ and λ are constant in time. Fur-
ther more, if we observe the behavior of the process we can differentiate two different
types of segments in the process:

1. between jumps it is driven by a pure diffusion process

2. at jump times (τi ) the value of the process after a jump depends on the value
right before the jump (τi− )

For the first case when the process is driven by a pure diffusion process we have
dXt
= µdt + σdWt , (4.15)
Xt
which can be written also as
1
d(ln Xt ) = (µ − σ 2 )dt + σdWt . (4.16)
2
After a jump at time τi we have that

X(τi ) − X(τi− ) = Ui X(τi− ) ,

or equivalently
X(τi ) = X(τi− )(1 + Ui ) . (4.17)

In the part 2.5.1 we have shown that the formula for the value of the diffusion process
at time t is  
1 2
X(t) = X(0) exp (µ − σ )t + σWt ,
2
which enables us to denote the value of the process right after the i-th jump as
  
1 2
X(τi ) = X(τi−1 ) exp (µ − σ )∆τ + σW∆τ 1 + Ui ,
2
where ∆τ = τi − τi−1 . If we would iterate this process, starting at t = 0 and assuming
there were no jumps at that time, we would get the final equation of the resulting asset
price process:
 
  N (t)
1 Y
X(t) = X(0) exp (µ − σ 2 )t + σWt  1 + Ui  . (4.18)
2 i=1

We have, however, not taken into account that the expected change in the resulting
process is not anymore Xt µdt, because we have incorporated jumps, and we have not

22
compensated for it. In order to fix that we would have to compensate for that like
we did earlier with the compensated Jump process. Doing that and then following the
same logic as above we would obtain the following resulting equation for the value of
the compensated jump-diffusion process at time t:
 
  N (t)
1 Y
X(t) = X(0) exp (µ − kλ − σ 2 )t + σWt  1 + Ui  . (4.19)
2 i=1

With the equation 4.19 we have obtained an expression needed to simulate the jump-
diffusion path. On Figure 4.3 we can observe a trajectory of a jump diffusion process
with parameters (µ = 0.05, σ = 0.4, λ = 4) and jump size U ∼ N (0, 0.04).

Asset price
200
Jump times

180

160
Price

140

120

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure 4.3: Jump-diffusion process (µ = 0.05, σ = 0.4, λ = 4) with jump size U ∼

N (0, 0.04)

23
5. Pricing an option

5.1. Options
An option is a form of financial derivative, which gives the buyer a right, but not
an obligation, to buy or sell the underlying asset at a certain time in the future. Option
contract is characterized by the following parameters:
• option price
• maturity - contract expiration
• strike price
The first two bullet points are self-explanatory, but the third one may be unclear. A
strike price is a set price at which the underlying security can be bought or sold when
the option is exercised. Options range from the most simplest ones (European options)
to exotic path dependent options (Asian options). In this thesis we will be mainly
concerned with European options.

5.1.1. European Call and Put Options

European options are often called vanilla options, because of their simplicity. A
European Call option gives the buyer a right to buy the underlying asset at maturity T
and pay for it the value of strike price K. Similarly, a European Put option gives the
buyer a right to sell the underlying asset at maturity and get for it the value of the strike
price.

Let us assume that we can trade an asset with the current price (spot price) of
X0 = 100. At this moment in time, we are convinced that the price of the asset will
rise in the future and as an alternative to the direct purchase of the asset, we decided
to buy a call option on that asset. The call option with maturity of one month and a
strike price K = 102 is trading at C = 6, but considering the fact that one call option

24
contract refers to 100 units of the underlying stock, the price of one such contract is
then 100 · 6 = 600. For simplicity, we can assume that after one month there are only
two possible scenarios:
• XT = 120
• XT = 90
In the second scenario, our contract looses its value and expires unexercised. This
means that we lose our initially invested capital of 600. If, however, the second sce-
nario happens, we exercise our right to buy 100 units of the underlying asset for the
price of 102. Considering the face that one unit of that asset is currently trading at 120
we immediately sell them in the market for the market price and earn a premium of
(120 − 102) · 100 − 600 = 1200.
Now that we have a proper understanding of how options work we can put that into
practice and dive into the pricing of the option. Although we have described both call
and put options, in the following chapters we will focus on pricing a call option, but
the process is analogous for the put option.

5.2. Itô’s Lemma for the Jump-diffusion Process

Although Ito’s lemma could have already been introduced and used to obtain the
formulas 4.18 and 4.19 we chose to use the iterative process to obtain them. Now, when
talking about pricing an option, which is a financial derivative and therefore derives its
value from the underlying price process, it comes naturally. The term derives means
that price of an option can be defined as a function of the underlying price process.
Let us denote with Xt the underlying price process, which will be a jump-diffusion
process. We are interested in obtaining a stochastic process driving the f where:

f = f (Xt ) .

At times between the jumps, the process Xt follows the simple diffusion and the change
in f in intervals between jumps can be expressed using Itô’s lemma:
∂f ∂f 1 ∂ 2f
df = dt + dXt + 2
(dXt )2
∂t ∂x 2 ∂x
(5.1)
1 2 2 ∂ 2f
 
∂f ∂f ∂f
= + Xt (µ − λk) + Xt σ dt + X t σ dWt .
∂t ∂x 2 ∂x2 ∂x
Whereas at jump times t = τi the change in f is given by
f (Xτi ) − f (Xτi− ) = f (Xτi− + Xτi− Ui ) − f (Xτi− )
(5.2)
= f (Xτi− (1 + Ui )) − f (Xτi− ) .

25
The above equation can be rewritten for any time t which is in that case

f (Xt ) − f (Xt− ) = [f (Xt− (1 + U )) − f (Xt− )]dNt . (5.3)

Let us now define the expected change of f at jump times. This will be useful further
on. If we donote the expected change in f at jump times as kf it is
Z
kf = E [f (Xt ) − f (Xt− )] = [f (Xt− (1 + u)) − f (Xt− )]g(u)du ,
Qu
(5.4)

where g(u) is the probability density function of the jump size. If we were to look at
the change df in an infinitesimally small interval dt conceptually we could define it as
something like this:
(
DP with probability (1 − λdt)
df = ,
DP + JP with probability λdt

where DP is the change due to diffusion part (equation 5.1), and JP is the change due
to the jump property of the underlying price process (equation 5.3). Let us now exam-
ine the expected value of the change in f with respect to both jump size distribution
and jump time distribution:

1 2 2 ∂ 2f
 
∂f ∂f
Qj Qu P
E[df ] = E E E [df ] = + Xt (µ − λk) + X σ dt(1 − λdt)+
∂t ∂x 2 t ∂x2
1 2 2 ∂ 2f
  
∂f ∂f
+ Xt (µ − λk) + X σ dt + kf λdt λdt
∂t ∂x 2 t ∂x2
1 2 2 ∂ 2f
 
∂f ∂f
= + Xt (µ − λk) + X σ dt + kf λdt
∂t ∂x 2 t ∂x2
1 2 2 ∂ 2f
 
∂f ∂f
= + Xt (µ − λk) + X σ + kf λdt dt .
∂t ∂x 2 t ∂x2

We can denote with µf drift of the process f

∂f ∂f 1 ∂ 2f
µf = + Xt (µ − λk) + Xt2 σ 2 2 + kf λdt , (5.5)
∂t ∂x 2 ∂x
and the expectation of the change in f is then simply

E[df ] = µf dt .

This allows us to specify df more precisely than our conceptual picture of it:
(
(µf − λkf )dt + σf dWt with probability (1 − λdt)
df = ,
(µf − λkf )dt + σf dWt + [f (Xt ) − f (Xt− )] with probability λdt

26
where
∂f
σf = σXt .
∂x
The final part now is to rewrite it in the form of Itô’s lemma:

df = (µf − λkf )dt + σf dWt + [f (Xt ) − f (Xt− )]dNt . (5.6)

As we can see the resulting process is again a Jump-diffusion process, but now the
drift µf of the process is not constant as before. We can see that µf as well as kf
are now dependent also on Xt , but this is not an issue as the definition of a diffusion
process allows that as well. For simplicity purposes earlier we decided to keep the
µ and σ constant. Now we have a general form of Itô’s lemma for a function of our
Jump-diffusion driven price process and this will help us in the next chapter where we
will show that jumps add a non-systematic risk in play which is not possible to hedge
using just the famous delta hedge.

5.3. Constructing a Hedging Portfolio to price an Op-

tion
It is important to understand the idea behind the hedging portfolio before we dive
into any details. Let us suppose we have two assets that generate the same cash flows,
then the price of those two assets must be the same, otherwise the arbitrage is possi-
ble and the first fundamental theorem of asset pricing is not satisfied. It is, however,
possible to mimic the cash flows of one asset by finding a portfolio of different assets
that generate the same cash flow. Such portfolio is called a replicating portfolio. The
main idea behind creating a replicating portfolio to price and option is, that if we find
a portfolio that mimics the payoff of the option, we know that the price of the option
and the portfolio have to be the same.
If we short an option, a hedging portfolio is a portfolio constructed from the short
position in an option and a long position in its replicating portfolio, or vice versa. The
value of such portfolio is expected to remain unchanged over time. In order to hedge it,
we must consider which assets will fall into consideration when choosing the weights
of our portfolio. In our simple world, we assume that we can trade only in:
• a derivative security - the option we want to price
• an underlying asset - a stock
• bonds - riskless asset

27
In the following sections we go into details of how to create a such portfolio for an
option written on an asset, when the price of the asset is governed by the jump-diffusion
process.

5.3.1. Delta hedge

The value of an option clearly depends on the underlying price process. This is,
of course, also the case for European call and put options. For simplicity purposes
we move for a bit in the Black-Scholes world where an underlying asset price follows
a simple geometric Brownian motion without jumps. This will allow us to get an
intuition on the topic of hedging, and we will then expand it to a jump-diffusion model.
Let us consider a situation where we are short one option, that is, we wrote an option
and sold it to somebody. The delta hedging strategy would tell us how many units of an
underlying asset we would have to buy if we want to offset the risk of our portfolio for a
small time interval dt. In other words, if we buy a correct amount of an underlying asset
to hedge our position in an option the change in value of our portfolio in a small time
interval dt should be zero. The downside of a delta hedge is that it requires frequent
readjustments of portfolio weights, which could, in real world, lead to high transaction
costs. Although this is not really our most important concern it was important to flag
it. Before actually applying it to our Jump-diffusion process and observing the issue
let us define the delta hedge more precisely.
We take a portfolio composed of stock, bond and a derivative security (option).
The value of the portfolio is given by

V (Xt ) = sXt + yB(t) + zf (Xt ) ,

where s is the number of units of stock in the portfolio, y is the number of units of
a riskless asset (bond) in the portfolio and z is the number of units of the derivative
security. The change in the portfolio value in an infinitesimally small time interval can
be expressed as
d
dV (Xt ) = V (Xt ) · dXt .
dXt
We consider the case we illustrated above where we were short one derivative security
or z = −1. The price of the bond is not dependend on the price of out stock so the
derivative of portfolio value with respect to stock price is then
d d
V (Xt ) = s − f (Xt ) .
dXt dXt

28
The goal of the delta hedge is to obtain such a portfolio that the change in the portfolio
value with respect to change in price of the underlying asset is zero. More precisely the
dV
equation dXt
= 0 must hold, which gives us the solution for the amount of stock that
we have to buy to offset the portfolio risk we took over when we shorted one option:

d
s= f (Xt ) . (5.7)
dXt
In the next section we will see how using this technique, although in another form, does
not eliminate all the risk from a portfolio when the underlying asset price is governed
by a jump-diffusion process.

5.3.2. Delta hedge for the Jump-diffusion process

As we have already announced we will consider a case when our underlying asset
follows the jump-diffusion process. Before we dive in into an examination of such a
hedge we will briefly give a description of the processes involved. Firstly, the price
process will be satisfying the stochastic differential equation for a jump-diffusion pro-
cess (4.14), namely
dXt
= (µ − λk)dt + σdWt + U dNt . (5.8)
Xt
For the option price we will slightly modify the notation of certain parts of the equation
5.6 to use µ1 , k1 and σ1 instead of µf , kf and σf

df = (µ1 − λk1 )dt + σ1 dWt + [f (Xt ) − f (Xt− )]dNt ,

where, of course, they stay the same:

Z
k1 = E [f (Xt ) − f (Xt− )] =
Qu
[f (Xt− (1 + u)) − f (Xt− )]g(u)du ,

∂f ∂f 1 ∂ 2f
µ1 = + Xt (µ − λk) + Xt2 σ 2 2 + k1 λdt ,
∂t ∂x 2 ∂x
∂f
σ1 = σXt .
∂x
The reason for that is to have a nice notation when we divide the equation by f to
obtain the relative change in f :

df
= (µf − λkf )dt + σf dWt + Uf dNt , (5.9)
f

29
 µ1 k1 σ1
where now we have µf = f
, kf = f
and f
. One new notation detail that we
introduced here is the Uf which is defined as

f (Xt ) − f (Xt− )
Uf = . (5.10)
f (Xt )
One last thing we still have to define is a portfolio and its value. A portofolio is a set of
weights [π1 , π2 , ...] which tells us the proportion of our money invested into financial
instruments 1,2,.. etc.

n
X
πi = 1 πi ∈ R, ∀i .
i=1

We can see that the weights can be negative, which would mean that we are in short
position for that instrument, that is, we sold some amount of that instrument and will
at some point have to buy back the same amount to cover. In our case we will have just
a stock, a derivative security (option) and a riskless asset which can be for example a
bond. Such defined portfolio then looks like this

π = [πx , πf , πr ] .

The relative change in the portfolio value V can be expressed as the following
dV dXt df
= πx + πf + πr dr . (5.11)
V Xt f
Now we will incorporate the equations 5.8 and 5.9 into the above equation and regroup
it a bit:

dV
=πx [(µ − λk)dt + σdWt + U dNt ]
V
+ πf [(µf − λkf )dt + σf dWt + Uf dNt ] + πr rdt
=(πx (µ − λk) + πf (µf − λkf ) + πr r)dt
+ (πx σ + πf σf )dWt + (πx U + πf Uf )dNt
=(µV − λkV )dt + σV dWt + UV dNt .

With this we obtained yet another jump diffusion process which now has new param-
eters µV , kV , σV and UV . Let us explicitly state the value of each one of them as we
will see they will come in handy soon.

30
 µV =πx (µ − r) + πf (µf − r) + r ,
kV =πx k + πf kf = EQU [UV ] ,
(5.12)
σV =πx σ + πf σf ,
UV =πx U + πf Uf .

If a process follows a simple diffusion process like in Black-Scholes world we would

obtain a delta neutral portfolio by setting πx , πf such that

σV = πx σ + πf σf = 0 . (5.13)

However, this choice of portfolio weights will not be able to eliminate the risk caused
by the jump property of the resulting process. Actually, no choice of weights will be
able to eliminate that risk. If we were to set the weights πx , πf that will satisfy the
equation 5.13 and denote them with πx∗ , πf∗ we would obtain a process defining the
relative change in value V ∗ of such portfolio
dV ∗
= (µ∗V − λkV∗ ) + UV∗ dNt . (5.14)
V∗
Employing only the delta hedge we see that we did not offset all the risk and the
problem of pricing our option is still left unsolved. The risk we still have is the jump
risk and it represents non-systematic risk. The ideas of systematic and non-systematic
risk will be discussed in the following section together with a new idea that has to be
employed in order to find the price of our option.

5.3.3. Employing the Captial Asset Pricing Model

Capital Asset Pricing model (CAPM)

The capital asset pricing model is a model that describes the relationship between
the expected return from an asset and the risk that comes with it. It assumes that the
return should only be dependent on the systematic risk. Systematic risk is the risk that
is related to the market as a whole and it cannot be diversified away, whereas non-
systematic risk is unique to the asset and can be diversified away by choosing a large
enough portfolio of different assets. The capital asset pricing formula is as follows:

E[Ri ] = Rf + βi (E[Rm ] − Rf ) , (5.15)

where Ri denotes the return of the investment, Rf is the return on the risk-free invest-
ment (i.e. risk-free rate), Rm denotes the return of the market and β is the factor that

31
measures the systematic risk of the investment. When we say return of the market we
think of it as a return of some well-diversified stock index like NASDAQ or S&P. The
expression (E[Rm ] − Rf ) is called market risk premium, that is, a premium received
for taking over the risk by being exposed to the market. We can think of the β factor of
some potential investment as a measure of how sensitive the return of the investment
is to the return of the market.
To put this into perspective, if we encounter a factor β = 0.5 this means that, on
average, the excess return of the investment is half the excess return of the market.
When β = 0 that means that the investment is not sensitive to the movements in the
market and therefore has no systematic risk. This may sound trivial, but this will be
crucial in the next section when we will use this idea to obtain a fair price of the option.

Obtaining the expression of the Option price

From modern portfolio theory we know that portfolios containing only non-systematic
risk have a β factor equal to zero. From the CAPM formula for the expected return
of investment it follows that a portfolio return must equal the riskless rate which we
denote with r:
µV = r ,
or if we incorporate that in 5.12 we get

πx (µ − r) + πf (µf − r) = 0 . (5.16)

If we would also take the equation 5.13 which ensures that we have a delta-neutral
portfolio, we then obtain

µ−r µf − r
= =φ. (5.17)
σ σf
The above ratio is called a market price of risk or Sharpe ratio. We will denote it
with φ. There is a way to obtain a martingale representation of the price, although we
will have to change to probability measure. We now incorporate the above equation
into equations 5.8 and 5.9 and they can now be written as

dXt
= (r − λk)dt + σ(dWt + φdt) + U dNt ,
Xt
df
= (r − λkf )dt + σf (dWt + φdt) + Uf dNt .
f
If we additionally denote with dW̃t a process

dW̃t = dWt + φdt , (5.18)

32
or equivalently Z t
W̃ (t) = W (t) + φ(s)ds , (5.19)
0
it allows us to write down our final equations as

dXt
= (r − λk)dt + σdW̃t + U dNt , (5.20)
Xt
df
= (r − λkf )dt + σf dW̃t + Uf dNt . (5.21)
f
Let us stop for a moment here and examine the equations. It can be shown that under
the original probability measure P the process W̃ will not be a Wiener process:
Z t Z t
P P
E [W̃ (t)] = E [W (t) + φ(s)ds] = φ(s)ds .
0 0
However, by applying the Girsanov theorem for processes involving jumps it allows us
to assert that there exists and equivalent probability measure P̃ under which the process
W̃ is a Wiener process and N remains a jump process with intensity λ. This is a very
important step as it will us to easily obtain an expression which, not only tells us a lot,
but also will allow as to price our option. There is only one ingredient missing in order
to obtain an expression which will tell us how to calculate the price of our option. We
are interested in the discounted price of the option, so we will introduce a function of
the option price and call it g,
g(f ) = f e−rt .

As we will utilize Itô’s lemma in order to obtain a differential of g, we first have to find
the expressions for partial derivatives used in the lemma, namely

∂g ∂g ∂ 2g
= −f re−rt , = e−rt and =0.
∂t ∂f ∂f 2
We now apply the lemma to obtain the expression for the term dg

∂g ∂g 1 ∂ 2g
d(f e−rt ) = dt + df + 2
(df )2
∂t ∂f 2 ∂f
= −re−rt f dt + e−rt df + 0
= −re−rt f dt + f e−rt [rdt + σf dW̃t + (Uf dNt − λkf dt)]
= f e−rt σf dW̃t + f e−rt (Uf dNt − λkf dt) .

The expected value of the change in the discounted option value process is 0 which
leads us to the conclusion that the process g is a martingale under P̃. This means that

33
the following holds:

f (t)e−rt =EP̃ [f (T )e−rT |Ft ]

f (t)e−rt =e−rT EP̃ [f (T )|Ft ] ,

which yields the final equation that we will use to price our option

f (t) = e−r(T −t) EP̃ [f (T )|Ft ] . (5.22)

When pricing our option we will focus on the options that have a signle fixed exercise
date, which is the case for European options. If we think about it, the price of the
European option at maturity T has to be equal to the payoff of the option and the payoff
is easy to calculate if we know the strike price and current price of the underlying
security at T . The way we will utilize the expression obtained above is that we will
simulate the paths of the process Xt (5.20) and we will be able to empirically obtain
the expected payoff of the option and with it, the price of the option.

5.4. Obtaining a formula for European Call Option

In the previous section we have obtained an expression (5.22) which tells us that
a premium requested by the writer of the option is equal to the discounted expected
future price of the option under the equivalent probability measure P̃. When it comes
to European options they can be exercised only at maturity T , which is different from
American options that can be exercised at any time before maturity. In this section we
will focus on European options.
As already stated, the price of the option at maturity will be equal to its payoff, because
otherwise market is not arbitrage-free. Putting together the equation 5.22 and this idea,
we obtain the following expression:

f (t) = e−r(T −t) EP̃ [P [f (T )]|Ft ] , (5.23)

where P denotes the payoff of the option. In case of the European call option this
payoff is
P (f (T )) = (X(T ) − K)+ .

If we further incorporate the price process X inside and denote the time to maturity
with τ = T − t, we get:

34
    + 
  N (T )
1 Y
f (t) = e−rτ EP̃ X(0) exp (r − kλ − σ 2 )T + σ W̃T  1 + Ui  − K  Ft 
2 i=1
   + 
  NY(τ )
1
= e−rτ EP̃ X(t) exp (r − kλ − σ 2 )τ + σ W̃τ  1 + Ui  − K  Ft  .
2 i=1

We know that under the risk-neutral probability measure P̃ the process W̃ is a regular
Wiener process and N is a Poisson process with the same intensity λ. Furthermore,
X(t) is adapted to filtration Ft , that is, it is known at time t. This allows us to write
the above expression as follows

   + 
  N (τ )
1 Y
f (t) = e−rτ E X(t) exp (r − kλ − σ 2 )τ + σWτ  1 + Ui  − K  Ft  .
2 i=1
(5.24)
From Black-Scholes theory we know that the call option price cBS is expressed as
"   + #
1
cBS (x, t) = e−r(T −t) E x exp (r − σ 2 )(T − t) + σWT −t − K Ft .
2

If we regroup the elements in our process we get

 
  N (T −t)
1 Y
X(t) exp (r − kλ − σ 2 )(T − t) + σWT −t  1 + Ui 
2 i=1
 
  N (T −t)
1 Y
= X(t) exp (r − σ 2 )(T − t) + σWT −t exp [−kλ(T − t)]  1 + Ui  .
2 i=1

This means that we can rewrite the equation 5.24 in terms of Black-Scholes formula
    
N (T −t)
Y
f (t) = E cBS X(t) exp [−kλ(T − t)]  1 + Ui  , t . (5.25)
i=1

Now because of the independence of the Poisson process N and the jump size Ui we
can calculate the above expectation as a sum of expectations for every possible number
of jumps and every possible jump size:
∞
" n
! !#
X Y [λ(T − t)]n
f (t) = E cBS X(t)e−kλ(T −t) 1 + Ui ,t e−λ(T −t) .
n=0 i=1
n!
(5.26)

35
In our case we will use a normally distributed jump size and the formula above is the
farthest we can go. We will still have to simulate the jumps and obtain Black-Scholes
price like that, but this result is more specific than the formula we proposed initially
(5.22). In the following section there are the results obtained using this formula for
different values of parameters.

5.5. Simulations
In this section we will now utilize our newly developed formula to simulate option
European call option price curves. The call option in question matures in 1 year from
now and a strike price K = 100. We want to examine how the price of such option
today depends on the spot price, that is, the current (observed) price of the underlying
asset. We simulate 5 paths, one for each λ, where λ ∈ [0, 1, 2, 4, 8]. We observe this
on the Figure 5.1.

70
λ=0
λ=1
60 λ=2
λ=4
50 λ=8

40
Call price

30

20

10

0 20 40 60 80 100 120 140

Spot price

Figure 5.1: European Call Option price with strike price K = 100 and maturity of 1 year

We notice that the higher the value of λ the higher the price of the option. This
makes sense, because the for a process with more frequent jumps the asset price pro-
cess has more probability to end up in the territory above the strike price K and have

36
a positive payoff. We remind that a European option cannot have a negative payoff,
we just simply do not exercise it in case the asset price process yields our option to be
unprofitable.
Next we examine how the price of the same European call option develops through
time if it matures at time of 1 year from now. As we approach the maturity of the
option the price of the option will be closer and closer to its payoff. Let us observe
the following 2 plots depicting the price of the Euorpean call option for 2 simulated
process. First of the processes (Figure 5.2) remains almost exclusively in the territory
north of the strike price K, while other process (Figure 5.3) spends some time both
above and under the strike price. What we examine here is, of course, expected, but it
is very interesting to observe. Namely, for the first case we see that the sensitivity of the
option price to the movements of the asset price is fairly high. It may be immediately
clear why this is the case, but let us support it with one example:

If we are at the time maturity t = T , we know that the price of the option will be
equal to its payoff which is now known - (ST − K)+ . Now, if we were to move back in
time for some small time interval ∆t, the option has still ∆t time left to maturity and
in this time interval the asset price process can move in either direction. Since there
is not much time left to maturity, we expect the price of the option to be close to the
payoff if the option would mature now, that is, (ST −∆t − K)+ . If we continue this
pattern we may, loosely speaking, interpret the option price to be determined by the
payoff if the option would mature at that moment, but adding to it a part determined
by the uncertainty of where the asset price process will end up at maturity. If we take
some time t, let us say t = 0.6, the price of the asset is somewhere between 160 and
180 which means that it has a higher probability of staying in the profitable territory
and therefore the option moves in high correlation with the asset price. For the same
reason we see on the second plot that as we move closer and closer to maturity, while
the asset price stays below the strike price, the price of the call option is less and less
responsive to the movements in the price of the underlying asset.
The behaviors we have just described can be summed up in the Figure 5.4 which de-
picts the surface of the option price depending on the spot price and time left to matu-
rity. One can clearly see that the surface is leaning towards the value of the payoff as
the time to maturity τ approaches 0.

37
 200 Asset price
Call price
180 Strike price

160

140

120
Price

100

80

60

40

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure 5.2: Price of the European Call Option that matures at T = 1 with K = 100, r = 2%

140
Asset price
Call price
120 Strike price

100

80
Price

60

40

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure 5.3: Price of the European Call Option that matures at T = 1 with K = 100, r = 2%

38
 60
Call price

40

20

100
0
0
0.2 50
0.4
0.6 Spot price
0.8
1
Time to maturity

Figure 5.4: Call price surface K = 95, λ = 4, µ = 0.05, σ = 0.5, U ∼ N (0.05, 0.09)

39
6. Hedging

Hedging is an investment made with the incentive of reducing the risk of adverse
price movements in an asset. Usually, this is done by taking an offsetting (opposite)
position in a related security. Indeed, we have already done that when we were con-
structing a hedging portfolio to price our option. Investors may consider hedging in
order to manage their exposure to risk. Hedging is, of course, no free-lunch and it
comes with a cost. While the benefit of hedging is that it allows us to reduce the risk,
it also reduces the potential gains made by the investment. If an investor is facing
uncertain market conditions and his positions are not hedged, the potential gains are
high, but so are the potential losses. By hedging his positions to mitigate the risk, an
investor has to accept the face the fact that the potential gains will scale down as well.
In section 5.3.1 we have already illustrated how to determine the quantity of the stock
that we have to hold to hedge a position in an option written on that stock. The method
we used there is called a delta hedge (∆-hedge).

6.1. Delta hedging in Black-Scholes world

Before diving into the process of hedging our option written on the stock which is
governed by the jump-diffusion process, let us first take a step into the Black-Scholes
world where we have ideal market conditions:

(i) the risk free interest rate is known and is constant through time,

(ii) the stock price is governed by the geometric Brownian motion

dXt = µt Xt dt + σt Xt dWt ,

(iii) the stock pays no dividend,

(iv) the market is frictionless, that is, there are no transaction costs,

40
 (v) it is possible to borrow or deposit any amount of money at the risk free interest
rate,

(vi) there are no penalties for shot selling.

Our starting point was the assumption that the option price will depend on the under-
lying asset price, that is:
ft = f (Xt ) .

Later we have arrived derived that in order to hedge a short position of one option we
would have to hold the amount s of the underlying asset, where s is
d
s= f (Xt ) . (6.1)
dXt
This means that if we would hold any time t this amount of the underlying, our position
would be perfectly hedged. In practice, however, continuous hedging is not possible.
The best we can do is to choose specific points in time at which we would rebalance
our portfolio. The more rebalancing times we have, the closer we get to the continuous
hedging. The effect of hedging on the value of the portfolio will be investigated soon.
Let us now examine on Figure 6.1 a generated asset price process, together with the
price process of its call option that matures at time T = 1 and has a strike price
K = 120. As we are still in the Black-Scholes world there is an absence of jumps and
the underlying process is a pure geometric Brownian motion.
Suppose we wrote (shorted) one call option at time t = 0, and we want to hedge our
position as we are unsure of the future price movements. In order for our position to
be perfectly hedged we want to have s units of the underlying asset, but to calculate
s we have to be able to differentiate the option price f with respect to the process X.
Luckily, this was already done in the original [2] paper. Namely, the expression for
obtaining the price of a call option is:

Ct = Xt N (d1 ) − N (d2 )Ke−rτ , (6.2)

41
 Asset price
120
Call option price
Strike price
100

80
Price

60

40

20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t

Figure 6.1: Asset price governed by the geometric Brownian motion and call option price with
K = 120

where

Ct − call price at time t

Xt − price of the underlying asset at time t
τ − time to maturity
K − strike price
r − risk-free rate
N − cumulative distribution function of N (0, 1)

and the terms d1 and d2 are

 
σ2
ln XKt + r + 2
τ √
d1 = √ and d2 = d1 − σ τ . (6.3)
σ τ
As the underlying asset price follows geometric Brownian motion we can use the ex-
pression 6.2 to obtain the call option price at each time t. The result can be observed

42
in the Figure 6.1. If we were to differentiate the above expression for C with respect
to X we would obtain (for details refer to A.1):
∂C
= N (d1 ) . (6.4)
∂X
Combining this expression with the expression 6.1 we obtain that the number of assets
we have to hold at each time t is N (d1 ). Now we have all ingredients to be able to
simulate and show hedging results. To illustrate the results, we set the initial portfolio
value to be 50, and we already know that we are short one call option, that means that
our position in that option is −1. We are considering the process shown at Figure 6.1
and we will perform delta hedging with different number of rebalancing times. On
Figure 6.2 one can observe the effect of hedging with different number of rebalancing
times. The first rebalancing time occurs at time t = 0.

56
54
52
50
48
Portfolio value

46
44
42
40
38 n=1
n=2
36 n=5
n = 10
34
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t

Figure 6.2: Development of the portfolio value with initial value of 50 for different number of
rebalancing times n

By frequently rebalancing our portfolio we ensure the stability in the value of our
portfolio. This can be further examined if we were to plot the distributions of portfolio
value at time t = T for different numbers of rebalancing times. This is shown on the
Figure (6.3).

43
 0.8 n=4
n = 10
0.7 n = 50
n = 100
0.6 n = 500

0.5

0.4

0.3

0.2

0.1

38 40 42 44 46 48 50 52 54 56 58 60 62 64
Portfolio value at t = T

Figure 6.3: Portfolio value distributions at time t = T for different numbers of rebalancing
times n

6.2. Quadratic hedging of contingent claims

When we were showing that a regular Black-Scholes delta hedge in case of a jump-
diffusion model cannot eliminate jump risk we stopped there. In fact in [3] the autors
demonstrate that in case we have a finite number n of jump amplitudes, we can hedge
away the jump risk by using n additional derivative securities with different strike
prices and maturities. In our case the jump amplitude is normally distributed, that is,
there is an infinite number of different jump sizes, and we would need an infinite num-
ber of different derivative securities to hedge away the jump risk. This is not possible
and that means that a contingent claim cannot be perfectly replicated and according to
the Theorem 3 we have an incomplete market situation. Since there will always exist
some amount of hedging error the best we can do is to try and minimize that error, this
leads us to investigate other hedging techniques, one of which is quadratic hedging.
Moreover, in this section we will also have a look at how to mathematically define a
problem of hedging. For that we will need to introduce a notion of a trading strategy.
Although we have not defined it explicitly, we have already used it. A trading strat-

44
egy is simply our selection of a covering portfolio in the underlying asset. [1] define it
as

Definition 9. Denote by L(X) the linear space of all Rd -valued predictable X-integrable
processes φ. A self-financing trading strategy is any pair (V0 , φ) such that V0 is and
F0 -measurable random variable and φ ∈ L(X).

A value φi (t) can be interpreted of as a number of shares of asset i held at time t and
V0 is the initial capital of an investor. By choosing a covering portfolio, we generate a
value process, which is used to cover our exposure. Its value is given by
Z t
Vt (φ) := V0 + φ(u)dXu , (6.5)
0
and the gains made from the trading process up to time t is denoted by
Z t
Gt (φ) := φ(u)dXu . (6.6)
0

When we choose a trading strategy and the initial capital we can then think about how
to evaluate the trading strategy. At time T the contingent claim will have a payoff of
H and our carefully chosen portfolio will have a value of VT (φ). To asses the quality
of the trading strategy we will use the difference of those 2 quantities. To be more
specific, we will calculate the hedging error as the square of that difference - quadratic
loss function. [1] give in their book the definition of mean-variance strategy to hedge
H (Definition 10). So far, we have mentioned a term risk multiple times and it is
mostly clear what it means, but it may be unclear how to measure it. There are various
ways of quantifying risk. One such is to quantify it represent it through the variance
of returns. Logically, if we have a process with low variance, this means that the
probability of a process steering far away from its expected value is low. Analogously,
for a high-variance process we can with less certainty predict its movements and future
paths. This is the core idea of hedging methods that aim to reduce the risk of the future
payoff.

Definition 10. Let Φ be a the space of all φ ∈ L(X) for which GT (φ) is in L2 (P)
and the gains from the trading process Gt (φ) is a martingale under the probability
measure P̃. A pair (V0 , φ) such that V0 ∈ R and φ ∈ Φ is called a mean-variance
optimal strategy for H, if it solves the optimization problem

min E (H − V0 − GT (φ))2 .
 
(6.7)
φ

45
In case there is an absence of jumps, that is, we are in the Black-Scholes world, the
solution to this optimization problem is simply sensitivity of the option price with
respect to the underlying asset price
∂C
= N (d1 ) .
∂X
In case there is an absence of diffusion, and we are left with only jump process with
only 1 possible jump amplitude, the solution is:

C(Xt + k) − C(Xt )
φ∗ = . (6.8)
k
To illustrate why, we look at the time t and observe an asset price Xt and a price of the
call option C(Xt ) on that asset. We know that at time t + dt a process can either stay
the same or jump, where the size of the jump has a single possible value k. We denote
the value of our portfolio at time t with

Vt = φ∗ Xt − C(Xt ) .

If the jump arrives inside the time interval dt the value of out portfolio will be

Vt+dt = φ∗ (Xt + k) − C(Xt + k) ,

and if the jump does not arrive the asset price will stay the same and the price of the call
option will change by only marginal amount (because we are closer to the maturity),
which can be ignored. This means that the value of our portfolio, in this case, has
virtually stayed unchanged no matter the value of φ∗ . If we want the value of our
portfolio to stay unchanged also in presence of jump inside the time interval dt we
have to choose φ∗ wisely:

Vt = Vt+dt
Xt φ∗ − C(Xt ) = Xt φ∗ + kφ∗ − C(Xt + k)
kφ∗ = C(Xt + k) − C(Xt ) ,

and we arrive to the final formula (6.8).

The derivation of the solution to the optimization problem 6.7 goes beyond the scope
of this thesis and is rather cumbersome. In turn, we will present the solution for the
problem obtained in [4]:
∂C 1
R
∗ σ 2 ∂X + X
z[C(t, X(1 + z)) − C(t, X)]νU (z)dz
φ = R , (6.9)
σ 2 + z 2 νU (z)dz

46
 ∂C
where C(t, X) is the Black-Scholes price for the call option and accordingly the ∂X
is the sensitivity of that price with respect to the price of the underlying asset. The
function νU is the probability density function of the jump size U . This result can
be interpreted as the average of the simple ∆-hedge in the absence of jumps and the
∆-hedge in case there is no diffusion part (6.8). To check the effectiveness of the
quadratic hedge, we will first use the jump-diffusion process on figure 5.2 and compare
the performance of the quadratic hedge and the regular ∆-hedge, where we pretend that
the underlying process has only diffusion component without jumps.

70
∆-hedge
65 Quadratic hedge
Jump times
60

55
Portfolio value

50

45

40

35

30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t

Figure 6.4: Development of the portfolio value with initial value of 50 for two different hedg-
ing strategies

On Figure 6.4, in terms of stability, both processes seem to be similar, and we even
observe a lower final value of the portfolio with quadratic hedging. Let us remember
that profit is not the goal of hedging, but rather a method to preserve the value of
the portfolio over time. With this in mind, one possible way to measure success of a
hedging method is the absolute difference from the initial portfolio value remunerated
at the risk free rate r. This means, if the portfolio value is initially set to V , the expected
value at maturity is V · erT . Continuing further, we will now compare the performance
of both hedging methods on 30 different jump-diffusion processes. The characteristics

47
 Symbol Description Value
X0 initial asset price 100
µ asset price drift 0.05
σ asset price volatility 0.5
λ jump intensity 4
r risk free rate 0.02
K call option strike price 100
U jump size distribution N (µ = 0.05, σ 2 = 0.222 )
T option maturity 1

Table 6.1: Jump-diffusion process characteristics

of the processes are outlined in Table 6.1.

A table with final values of the portfolio can be seen in Appendix A.2. To obtain
a numerical comparison between the two hedging methods we perform a paired t-test,
as the two methods are not independent. They are related through the generated pro-
cesses. We quantify the success of each portfolio as the absolute distance from final
value of the risk-free portfolio with the same initial value. We want to test whether
the value absolute difference of the quadratically hedged portfolio is lower than the
absolute difference of the ∆-hedged portfolio. By setting up this one-sided alternative,
we obtain a p-value of 0.7366. In other words, there is no evidence that the quadratic
hedge does a better job than a classic ∆-hedge. It is, in fact, even more leaned towards
the side of the ∆-hedge doing a better job. One could argue that by increasing the jump
intensity our quadratic hedge could really shine. The reason for that is that the path of
a process with lower intensity has a higher probability of being closer to the path gen-
erated by a regular geometric Brownian motion. Further improvements could be made
to tune the jump intensity parameter and jump size distribution to mimic the real world
as closely as possible, and perform this analysis again. Nevertheless, by introducing
the quadratic hedging we have obtained a new, very capable, tool which could in future
help us hedge our portfolios in presence of processes with jump characteristics.

48
7. Conclusion

We have undergone a journey starting from the basic financial concepts and estab-
lishing a process that will model a price of an asset, through pricing an option that
derives its value from this asset and finally we have shown how to hedge such a contin-
gent claim. The theory of derivatives market is more concerned about the relationship
between the underlying and its derivative and much less with economic fundamentals
that determine the price of the underlying. However, some basic economic reasoning
and concepts have to be incorporated, but the focus is on the mathematical formulation.
We have constructed a mathematical model of real markets, which is of course
still a simplification. Nevertheless, it allowed us to draw some useful conclusions,
for example, why the premium requested by the writer of the option is equal to the
expected value of the discounted future payoff of the option. Furthermore, we have
seen that by increasing the average number of jumps (parameter λ) the price of the
option increased as a consequence of the fact that an option has a higher probability of
ending up in-the-money. Like we have already mentioned, the Black-Scholes model of
stock prices does not take into account the potential jumps that, in real world, occur due
to some political or economical events and are not predictable. What this entails is if
we were to use Black-Scholes model to price options, we may end up with underpriced
options and there may exist an arbitrage opportunity. The similar goes for hedging.
Further improvements of this work are naturally possible. For example, one could
incorporate stochastic processes that will model volatility and risk-free rate, both of
which have been kept constant in our work so far. Nonetheless, we have created a
decent model of the market with less constraints than the one proposed by [2]. By
doing this, we have taken a serious step into derivative pricing, a field that has a lot of
ideas still left to explore. Hopefully, this thesis has given a reader a thorough overview
of the main ideas in mathematical finance and derivative pricing which forms a stable
foundation for further exploration in the field.

49
B IBLIOGRAPHY

[1] N. H. Bingham and R. Kiesel. Risk-Neutral Valuation: Pricing and Hedging of

Financial Derivatives. 2004.

[2] F. Black and M. Scholes. The Pricing of Options and Corporate Liabilities. 1,
May 1972.

[3] C. Chiarella, X.-Z. He, and C. Sklibosios Nikitopoulos. Derivative Security Pric-
ing, volume 21 of Dynamic Modeling and Econometrics in Economics and Fi-
nance. Springer Berlin Heidelberg, Berlin, Heidelberg, 2015.

[4] R. Cont and P. Tankov. Financial modelling with jump processes. Chapman &
Hall/CRC financial mathematics series. Chapman & Hall/CRC, 2004.

[5] E. Eberlein and J. Kallsen. Mathematical Finance. Springer Finance. Springer

International Publishing, 2019.

[6] R. A. Jarrow. Continuous-Time Asset Pricing Theory: A Martingale-Based Ap-

proach. Springer Finance. Springer International Publishing, 2021.

[7] D. Mandel. Understanding Market Price of Risk.

[8] R. C. Merton. Option pricing when underlying stock returns are discontinuous.
Journal of Financial Economics, 3(1-2):125–144, Jan. 1976.

[9] S. E. Shreve. Stochastic Calculus for Finance II: Continuous-time models.

Springer finance textbook. Springer, New York, NY, 2008.

[10] P. Whitt. IEOR 4701: Stochastic Models in FE.

50
L IST OF F IGURES

2.1. Geometric Brownian motion paths for parameters µ=0.05 and σ = 0.4 10

4.1. Poisson process path with deterministic increment of 1 and parameter

λ=4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2. Pure Jump Process with λ = 6 and jump size U ∼ N (µ = 0, σ 2 = 0.04) 20
4.3. Jump-diffusion process (µ = 0.05, σ = 0.4, λ = 4) with jump size
U ∼ N (0, 0.04) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5.1. European Call Option price with strike price K = 100 and maturity of
1 year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2. Price of the European Call Option that matures at T = 1 with K =
100, r = 2% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.3. Price of the European Call Option that matures at T = 1 with K =
100, r = 2% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.4. Call price surface K = 95, λ = 4, µ = 0.05, σ = 0.5, U ∼
N (0.05, 0.09) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6.1. Asset price governed by the geometric Brownian motion and call op-
tion price with K = 120 . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2. Development of the portfolio value with initial value of 50 for different
number of rebalancing times n . . . . . . . . . . . . . . . . . . . . . 43
6.3. Portfolio value distributions at time t = T for different numbers of
rebalancing times n . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.4. Development of the portfolio value with initial value of 50 for two
different hedging strategies . . . . . . . . . . . . . . . . . . . . . . . 47

51
 Option Pricing and Hedging under Jump-diffusion model

Abstract

Stochastic processes have a widespread application in the computer science and

in this thesis their application in the field of finance will be studied. The focus of this
thesis will be to mathematically define a price of an asset using a jump-diffusion model
which is then used to price a derivative. By programming simulations of prices of an
asset and its derivative, their behavior in time will be shown. After the prices of an asset
and its derivative have been defined, we want to show how to mathematically define
a problem of hedging a position. This also has to be accompanied with implemented
simulations to illustrate the effect of hedging.

Keywords: stochastic processes, finance, pricing, options, hedging

 Odred̄ivanje cijene i premošćivanja rizika u slučaju difuzijskog procesa sa
skokovima

Sažetak

Stohastički procesi imaju široku primjenu u raznim područjima računarske znanosti,

a u sklopu ovog rada istražit će se njihova primjena u području financija. Fokus rada
je matematički definirati proces cijene nekog sredstva u obliku difuzijskog procesa
sa skokovima, na temelju kojeg se naknadno odred̄uje cijena njegovog financijskog
derivata. Potrebno je programskim simulacijama cijena sredstva i derivata pokazati
njihovo ponašanje u vremenu. Nakon što su spomenuti procesi cijena definirani, že-
limo pokazati kako se matematički može predstaviti i riješiti problem premošćivanja
rizika, te navedeno popratiti programskim simulacijama scenarija s ciljem ilustracije
efekta premošćivanja rizika.

Ključne riječi: stohastički procesi, financije, odred̄ivanje cijene, opcije, premošći-

vanje rizika
Appendix A
Tables and derivations

A.1. Delta hedge

Starting from 6.2 and 6.3 we want to obtain the expression 6.4. Of course we just
need to differentiate the call price by X, but also d1 and d2 are functions of X and it is
maybe not immediately straighforward how to obtain the expression 6.4:
. ∂
Ct = Xt N (d1 ) − N (d2 )Ke−rτ
∂X
∂C ∂ ∂ 
N (d2 )Ke−rτ

= [XN (d1 )] −
∂X ∂X ∂X
∂C ∂N (d1 ) ∂N (d2 )
= N (d1 ) + X − Ke−rτ . (A.1)
∂X ∂X ∂X
∂N (d1 ) ∂N (d2 )
Now we will decompose the terms ∂X
and ∂X
, but first let us take a look at what
N (z) represents. N is a unit normal cumulative distribution function, that is:
Z z
1 u2
N (z) = √ e− 2 du , (A.2)
2π −∞
∂N
which means that the ∂z
is the value of the unit normal probability density function at
the point z or in our case this would mean:
∂N 1 d2
1 ∂N 1 d2
2
= √ e− 2 and = √ e− 2 . (A.3)
∂d1 2π ∂d2 2π
We also repeat expressions for d1 and d2 :
 
σ2
ln XKt + r + 2
τ √
d1 = √ , d2 = d1 − σ τ .
σ τ
∂N (d1 ) ∂N (d2 ) ∂N (d1 )
Let us now go back to ∂X
and ∂X
we obtain the following expression for ∂X
:
∂N (d1 ) ∂N ∂d1 1 d2
1 1 K 1
= = √ e− 2 √
∂X ∂d1 ∂X 2π σ τXK

54
 ∂N (d1 ) 1 d2
1
= √ e− 2 , (A.4)
∂X Xσ 2πτ
and analogously for d2
√
∂N (d2 ) 1 d2
2 1 (d1 −σ τ )2
= √ e− 2 = √ e− 2
∂X Xσ 2πτ Xσ 2πτ
∂N (d2 ) 1 d2
1
√ σ2
= √ e− 2 · ed1 σ τ · e−τ 2 . (A.5)
∂X Xσ 2πτ
Now we will plug it all back in into the equation A.1 we obtain

d1 2

∂C e− 2 h X
−rτ ln K
2
+(r+ σ2 )τ −τ σ2
2i
= N (d1 ) + √ X − Ke e ·e
∂X Xσ 2πτ
d2
1
e− 2 h X
ln K
i
= N (d1 ) + √ X − Ke
Xσ 2πτ
d2
1
e− 2
= N (d1 ) + √ [X − X] .
Xσ 2πτ
Finally we have
∂C
= N (d1 ) . (A.6)
∂X

A.2. Tables

n ∆ [u] Quadratic [v] d1 = |u − V erT | d2 = |v − V erT | d = d1 − d2

1 51.2364 50.7511 0.2263 0.259 -0.0327
2 49.4292 49.5088 1.5809 1.5013 0.0796
3 46.6714 46.2663 4.3386 4.7437 -0.4051
4 52.8528 52.7247 1.8428 1.7146 0.1282
5 54.5897 54.1004 3.5796 3.0903 0.4893
6 46.9578 47.0606 4.0523 3.9494 0.1029
7 46.6498 46.2147 4.3602 4.7954 -0.4351
8 46.9598 47.1167 4.0503 3.8933 0.157
9 53.3914 53.8338 2.3813 2.8238 -0.4425
10 57.397 57.3993 6.387 6.3892 -0.0023
11 47.0675 47.4386 3.9426 3.5715 0.3711
12 55.3093 55.424 4.2992 4.414 -0.1147
13 41.0318 40.7874 9.9783 10.2227 -0.2444

55
 14 54.4601 54.1055 3.4501 3.0954 0.3546
15 45.0729 45.0625 5.9372 5.9476 -0.0104
16 51.4265 50.6558 0.4164 0.3543 0.0621
17 55.0025 54.9509 3.9924 3.9408 0.0516
18 55.5128 56.0583 4.5027 5.0482 -0.5455
19 35.3353 35.6649 15.6747 15.3452 0.3296
20 31.3557 31.7701 19.6543 19.24 0.4144
21 51.1031 50.3832 0.0931 0.6269 -0.5338
22 49.2211 48.6752 1.7889 2.3348 -0.5459
23 53.4757 53.4791 2.4657 2.469 -0.0033
24 47.7635 47.6825 3.2466 3.3275 -0.0809
25 45.6697 44.8682 5.3404 6.1418 -0.8014
26 53.6436 53.8819 2.6335 2.8718 -0.2383
27 47.9611 48.018 3.049 2.9921 0.0569
28 55.7981 55.8174 4.7881 4.8074 -0.0193
29 51.9937 51.2039 0.9836 0.1939 0.7897
30 51.9582 52.1369 0.9481 1.1269 -0.1787

Table A.1: Portfolio value at time of maturity T for 2 hedging methods where number of
rebalancings was n = 100

56