Vous êtes sur la page 1sur 360

Stochastic Calculus - Summary of tools

Céline Azizieh

Université Libre de Bruxelles

November 2018

1 / 80
Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

2 / 80
Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

3 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed.

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed. Mathematicians speak about the notion of
“measurable space” ˆΩ, F .

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed. Mathematicians speak about the notion of
“measurable space” ˆΩ, F .
In financial modelling, Ω represents the different possible evolution
scenarios in a given market:

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed. Mathematicians speak about the notion of
“measurable space” ˆΩ, F .
In financial modelling, Ω represents the different possible evolution
scenarios in a given market: each “scenario” ω > Ω typically represents
a possible evolution in time of the prices of a set of financial
instruments.

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed. Mathematicians speak about the notion of
“measurable space” ˆΩ, F .
In financial modelling, Ω represents the different possible evolution
scenarios in a given market: each “scenario” ω > Ω typically represents
a possible evolution in time of the prices of a set of financial
instruments.
Scenarios are hence the “elementary events” in our modelling
framework.

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

We consider the set of “all possible states of the world” Ω, with a


“sigma-algebra” F representing all "admissible" events to which a
probability will be attributed. Mathematicians speak about the notion of
“measurable space” ˆΩ, F .
In financial modelling, Ω represents the different possible evolution
scenarios in a given market: each “scenario” ω > Ω typically represents
a possible evolution in time of the prices of a set of financial
instruments.
Scenarios are hence the “elementary events” in our modelling
framework.

4 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

Let us recall that a probability measure is a mapping:

5 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

Let us recall that a probability measure is a mapping:

PF 0, 1  A ( PˆA
such that:
PˆA > 0, 1 for all A > F ,
P ˆΩ  1,
If Ai ˆi > N0  are 2 by 2 disjoint, then:

P ‰8i >N0 Ai Ž Qª P A
i 1
ˆ i

(sigma-additivity)

5 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

Let us recall that a probability measure is a mapping:

PF 0, 1  A ( PˆA
such that:
PˆA > 0, 1 for all A > F ,
P ˆΩ  1,
If Ai ˆi > N0  are 2 by 2 disjoint, then:

P ‰8i >N0 Ai Ž Qª P A
i 1
ˆ i

(sigma-additivity)
We will always work on a probability space ˆΩ, F , P in which
elementary events represent typically

5 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Financial modelling: Probabilistic framework

Let us recall that a probability measure is a mapping:

PF 0, 1  A ( PˆA
such that:
PˆA > 0, 1 for all A > F ,
P ˆΩ  1,
If Ai ˆi > N0  are 2 by 2 disjoint, then:

P ‰8i >N0 Ai Ž Qª P A
i 1
ˆ i

(sigma-additivity)
We will always work on a probability space ˆΩ, F , P in which
elementary events represent typically the evolution of a given market
during a given time interval.
5 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets,

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc),

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time,

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω


such that for each fixed t, X ˆt, .  Ω R is a random variable.

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω


such that for each fixed t, X ˆt, .  Ω R is a random variable.
Hence:

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω


such that for each fixed t, X ˆt, .  Ω R is a random variable.
Hence:
For a fixed scenario ω, we have a trajectory

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω


such that for each fixed t, X ˆt, .  Ω R is a random variable.
Hence:
For a fixed scenario ω, we have a trajectory
For a fixed instant t, we have a random variable

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

The prices of financial assets, or more generally, of market


variables (e.g.: interest rates, exchange rates, etc), will be
modelled by stochastic processes in continuous time, i.e. the
given of a random variable at each instant t:

X  0, ª  Ω R  ˆt, ω  ( X ˆt, ω


such that for each fixed t, X ˆt, .  Ω R is a random variable.
Hence:
For a fixed scenario ω, we have a trajectory
For a fixed instant t, we have a random variable

6 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time,

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability,

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer quantities seen as stochastic at t=0 are not
stochastic anymore at a future instant t A 0 if their value is revealed in the
mean time, i.e. given the available information at t A 0.

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer quantities seen as stochastic at t=0 are not
stochastic anymore at a future instant t A 0 if their value is revealed in the
mean time, i.e. given the available information at t A 0.

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer quantities seen as stochastic at t=0 are not
stochastic anymore at a future instant t A 0 if their value is revealed in the
mean time, i.e. given the available information at t A 0.
Without entering into details, one can say that a filtration represents the
growing information that becomes available progressively when time passes...

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer quantities seen as stochastic at t=0 are not
stochastic anymore at a future instant t A 0 if their value is revealed in the
mean time, i.e. given the available information at t A 0.
Without entering into details, one can say that a filtration represents the
growing information that becomes available progressively when time passes...
Now, in order to avoid formalism, we will just suppose that market variables
are modelled by stochastic processes whose value at t is known "as soon as
we have reached instant t".

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic process and filtration

Now, if t represents time, in order to model concepts like the progressive


arrival of information, causality and predictability, we should normally
introduce the concept of filtration.
In a dynamical framework, when time passes, information is progressively
revealed to the observer quantities seen as stochastic at t=0 are not
stochastic anymore at a future instant t A 0 if their value is revealed in the
mean time, i.e. given the available information at t A 0.
Without entering into details, one can say that a filtration represents the
growing information that becomes available progressively when time passes...
Now, in order to avoid formalism, we will just suppose that market variables
are modelled by stochastic processes whose value at t is known "as soon as
we have reached instant t".
This leads to the so-called notion of adapted process : intuitively, a process
ˆXt  is adapted when the value of the r.v. Xt is known as soon as we have
reached instant t.

7 / 80
Financial modelling: Probabilitic framework Financial modelling: Probabilistic framework

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

8 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest:

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales,

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation ,

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.
Suppose that X and Y are two random variables, where Y takes a
finite number of possible values.

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.
Suppose that X and Y are two random variables, where Y takes a
finite number of possible values. The conditional expectation of X
given Y : E X SY  corresponds to a new random variable whose value
is known as soon as the value of Y is known:

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.
Suppose that X and Y are two random variables, where Y takes a
finite number of possible values. The conditional expectation of X
given Y : E X SY  corresponds to a new random variable whose value
is known as soon as the value of Y is known:

E X SY ˆω  E X SY Y ˆω .

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.
Suppose that X and Y are two random variables, where Y takes a
finite number of possible values. The conditional expectation of X
given Y : E X SY  corresponds to a new random variable whose value
is known as soon as the value of Y is known:

E X SY ˆω  E X SY Y ˆω .
The last conditional expectation is taken w.r.t. an event.

9 / 80
Martingales Conditional expectation

Conditional expectation

An important type of processes used in financial modelling are of


particular interest: martingales.
Before defining martingales, let us recall the concept of conditional
expectation , that will play an important role.
Suppose that X and Y are two random variables, where Y takes a
finite number of possible values. The conditional expectation of X
given Y : E X SY  corresponds to a new random variable whose value
is known as soon as the value of Y is known:

E X SY ˆω  E X SY Y ˆω .
The last conditional expectation is taken w.r.t. an event.
This means that it takes value E ˆX SY yi  with probability P Y yi .

9 / 80
Martingales Conditional expectation

Conditional expectation

Remark
1 We can interpret E ˆX SY  as a function of ω

10 / 80
Martingales Conditional expectation

Conditional expectation

Remark
1 We can interpret E ˆX SY  as a function of ω or as a function of y :

10 / 80
Martingales Conditional expectation

Conditional expectation

Remark
1 We can interpret E ˆX SY  as a function of ω or as a function of y :

E ˆX SY ˆω  or E ˆX SY y .

Actually, the value of E ˆX SY  is known as soon as the value of Y


is known , so we can see it as a function of Y : E ˆX SY  f ˆY .

10 / 80
Martingales Conditional expectation

Conditional expectation

Remark
1 We can interpret E ˆX SY  as a function of ω or as a function of y :

E ˆX SY ˆω  or E ˆX SY y .

Actually, the value of E ˆX SY  is known as soon as the value of Y


is known , so we can see it as a function of Y : E ˆX SY  f ˆY .
2 This concept can be extended in a broader context, i.e. if Y is non
necessarily discrete anymore. See for instance [Karr] for more
details.

10 / 80
Martingales Conditional expectation

Conditional expectation

Remark
1 We can interpret E ˆX SY  as a function of ω or as a function of y :

E ˆX SY ˆω  or E ˆX SY y .

Actually, the value of E ˆX SY  is known as soon as the value of Y


is known , so we can see it as a function of Y : E ˆX SY  f ˆY .
2 This concept can be extended in a broader context, i.e. if Y is non
necessarily discrete anymore. See for instance [Karr] for more
details.
3 If Y is replaced by a random vector, or n random variables
Y1 , Y2 , ..., Yn , the definition can be directly extended for defining
E X SY1 , ..., Yn .

10 / 80
Martingales Conditional expectation

Conditional expectation

In practice

11 / 80
Martingales Conditional expectation

Conditional expectation

In practice

1 If X , Y are discrete:

E ˆX SY y Qx P X
i
i ˆ xi SY y Q x P XP Yx , Yy
i
i
ˆ
ˆ
i


11 / 80
Martingales Conditional expectation

Conditional expectation

In practice

1 If X , Y are discrete:

E ˆX SY y Qx P X
i
i ˆ xi SY y Q x P XP Yx , Yy
i
i
ˆ
ˆ
i


2 If ˆX , Y  has a joint density f ˆx, y  (see later) and Y has density


fY ˆy , then one can show:

E ˆX SY y S R
x
f ˆx, y 
f Y ˆy 
dx

11 / 80
Martingales Conditional expectation

Conditional expectation

In practice

1 If X , Y are discrete:

E ˆX SY y Qx P X
i
i ˆ xi SY y Q x P XP Yx , Yy
i
i
ˆ
ˆ
i


2 If ˆX , Y  has a joint density f ˆx, y  (see later) and Y has density


fY ˆy , then one can show:

E ˆX SY y S R
x
f ˆx, y 
f Y ˆy 
dx  SR
xfX SY ˆx, y dx

11 / 80
Martingales Conditional expectation

Conditional expectation

In practice

1 If X , Y are discrete:

E ˆX SY y Qx P X
i
i ˆ xi SY y Q x P XP Yx , Yy
i
i
ˆ
ˆ
i


2 If ˆX , Y  has a joint density f ˆx, y  (see later) and Y has density


fY ˆy , then one can show:

E ˆX SY y S R
x
f ˆx, y 
f Y ˆy 
dx  SR
xfX SY ˆx, y dx

We call fX SY ˆx, y  the conditional density of X given Y y.

11 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")
E E X SY  E X  : in average, the conditional expectation of a
random variable given another rv preserves the expectation.

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")
E E X SY  E X  : in average, the conditional expectation of a
random variable given another rv preserves the expectation.
E g ˆY SY  g ˆY : the conditional expectation of a function of Y,
given Y, is simply that function. This is an easy consequence of
the definition.

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")
E E X SY  E X  : in average, the conditional expectation of a
random variable given another rv preserves the expectation.
E g ˆY SY  g ˆY : the conditional expectation of a function of Y,
given Y, is simply that function. This is an easy consequence of
the definition.
E Xg ˆY SY  g ˆY E X SY 

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")
E E X SY  E X  : in average, the conditional expectation of a
random variable given another rv preserves the expectation.
E g ˆY SY  g ˆY : the conditional expectation of a function of Y,
given Y, is simply that function. This is an easy consequence of
the definition.
E Xg ˆY SY  g ˆY E X SY 
E aX  bZ SY  aE X SY   bE Z SY  where X , Y , Z are rvs and
a, b > R (linearity of the conditional expectation).

12 / 80
Martingales Conditional expectation

Conditional expectation

One recall the following properties:


E X SY  is a function of Y (it is "σ ˆY -measurable")
E E X SY  E X  : in average, the conditional expectation of a
random variable given another rv preserves the expectation.
E g ˆY SY  g ˆY : the conditional expectation of a function of Y,
given Y, is simply that function. This is an easy consequence of
the definition.
E Xg ˆY SY  g ˆY E X SY 
E aX  bZ SY  aE X SY   bE Z SY  where X , Y , Z are rvs and
a, b > R (linearity of the conditional expectation).
E X SY  E X  if X , Y are independent (straightforward
consequence of the definition).

12 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)


The SP ˆYn n>N0 is a martingale w.r.t. the SP ˆXn n>N0


13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)


The SP ˆYn n>N0 is a martingale w.r.t. the SP ˆXn n>N0

(i) Yn is σ ˆX1 , . . . , Xn -measurable for all n;

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)


The SP ˆYn n>N0 is a martingale w.r.t. the SP ˆXn n>N0

(i) Yn is σ ˆX1 , . . . , Xn -measurable for all n;
(ii) E SYn S @ ª for all n;

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)


The SP ˆYn n>N0 is a martingale w.r.t. the SP ˆXn n>N0

(i) Yn is σ ˆX1 , . . . , Xn -measurable for all n;
(ii) E SYn S @ ª for all n;
(iii) E Yn1 SX1 , . . . , Xn  Yn a.s. for all n;

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

We are now ready to define a martingale.


We can introduce this concept in a discrete or in a continuous time
environment.

Definition (Discrete time martingale w.r.t. a stochastic process)


The SP ˆYn n>N0 is a martingale w.r.t. the SP ˆXn n>N0

(i) Yn is σ ˆX1 , . . . , Xn -measurable for all n;
(ii) E SYn S @ ª for all n;
(iii) E Yn1 SX1 , . . . , Xn  Yn a.s. for all n;

13 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn 

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 
E Ynk 1 SX1 , ..., Xn 

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 
E Ynk 1 SX1 , ..., Xn 
...

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 
E Ynk 1 SX1 , ..., Xn 
... E Yn1 SX1 , ..., Xn 

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 
E Ynk 1 SX1 , ..., Xn 
... E Yn1 SX1 , ..., Xn 
Yn a.s. for all n.

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Remarks:
1 (i) just states that "Yn is a function of X1 , . . . , Xn only" (i.e. Yn is
known once X1 , ..., Xn are known).
2 (iii) shows that a martingale can be thought of as the fortune of a
gambler betting on a fair game.
3 (iii)  E Yn  E Y0  for all n (mean-stationarity).
4 Using (iii), we also have (for k 2, 3, . . .)

E Ynk SX1 , ..., Xn  EE Ynk SX1 , ..., Xnk TX1 , ..., Xn 
E Ynk 1 SX1 , ..., Xn 
... E Yn1 SX1 , ..., Xn 
Yn a.s. for all n.

14 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time.

15 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time. Martingales can also be defined in this framework.
Definition (Continuous time martingale)
A stochastic process ˆMt t C0 is a martingale w.r.t. a process ˆXt t if for
all t:
1 Mt is σ ˜Xs , s B t -measurable

15 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time. Martingales can also be defined in this framework.
Definition (Continuous time martingale)
A stochastic process ˆMt t C0 is a martingale w.r.t. a process ˆXt t if for
all t:
1 Mt is σ ˜Xs , s B t -measurable
2 E SMt S @ ª

15 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time. Martingales can also be defined in this framework.
Definition (Continuous time martingale)
A stochastic process ˆMt t C0 is a martingale w.r.t. a process ˆXt t if for
all t:
1 Mt is σ ˜Xs , s B t -measurable
2 E SMt S @ ª
3 E Mt SXu , u B s  Ms for all s @ t.

15 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time. Martingales can also be defined in this framework.
Definition (Continuous time martingale)
A stochastic process ˆMt t C0 is a martingale w.r.t. a process ˆXt t if for
all t:
1 Mt is σ ˜Xs , s B t -measurable
2 E SMt S @ ª
3 E Mt SXu , u B s  Ms for all s @ t.

Interpretation: accumulated gains of a player in an equilibrated game

15 / 80
Martingales Martingales in discrete time

Martingales: Definitions and basic comments

Often in finance, we will consider stochastic processes in continuous


time. Martingales can also be defined in this framework.
Definition (Continuous time martingale)
A stochastic process ˆMt t C0 is a martingale w.r.t. a process ˆXt t if for
all t:
1 Mt is σ ˜Xs , s B t -measurable
2 E SMt S @ ª
3 E Mt SXu , u B s  Ms for all s @ t.

Interpretation: accumulated gains of a player in an equilibrated game

In particular, E Mt  M0 .

15 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn P n
i 1 Xi .

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case.

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn 

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn 

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn 

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn  E Yn SX1 , ..., Xn E Xn1 SX1 , ..., Xn 

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi  1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn  E Yn SX1 , ..., Xn E Xn1 SX1 , ..., Xn 

Yn  E Xn1 

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi 1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn  E Yn SX1 , ..., Xn E Xn1 SX1 , ..., Xn 

Yn  E Xn1  Yn  0

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi 1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn  E Yn SX1 , ..., Xn E Xn1 SX1 , ..., Xn 

Yn  E Xn1  Yn  0

16 / 80
Martingales Martingales in discrete time

Example of martingale: random walk

Let X1 , X2 , . . . be i.i.d. with

P Xi 1 p and P Xi 1 q 1  p.

Let Yn  Sn n
P
i 1 Xi .
; The SP ˆYn  is called a random walk.It represents the fortune after
n games of a player tossing a coin again and again.
If p q 1~2, the RW is said to be symmetric.
One can easily see that in the symmetric case Y is a martingale
w.r.t. ˆXn  in the symmetric case. Indeed, if we focus on condition
(iii):
E Yn1 SX1 , ..., Xn  E Yn Xn1 SX1 , ..., Xn  E Yn SX1 , ..., Xn E Xn1 SX1 , ..., Xn 

Yn  E Xn1  Yn  0 Yn a.s. for all n,


where we used that Yn is a function of X1 , ..., Xn and that Xn1 X1 , ... ÙÙ
16 / 80
Martingales Martingales in discrete time

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

17 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

18 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

S ˆn  S ˆ0 M1n

k 1
ˆ  ik  S ˆ0  Me
n

k 1
δk

18 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

S ˆn  S ˆ0 M1
n

k 1
ˆ  ik  S ˆ0  Me
n

k 1
δk

Stochastic model :
S ˆn  S ˆ0 Me
n

k 1
Yk

où Y1 , ..., Yn are i.i.d. :

18 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

S ˆn  S ˆ0 M1n

k 1
ˆ  ik  S ˆ0  Me
n

k 1
δk

Stochastic model :
S ˆn  S ˆ0 Me
n

k 1
Yk

où Y1 , ..., Yn are i.i.d. :


δσ 1~2
Yi œ
δσ 1~2

18 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

S ˆn  S ˆ0 M1n

k 1
ˆ  ik  S ˆ0  Me
n

k 1
δk

Stochastic model :
S ˆn  S ˆ0 Me
n

k 1
Yk

où Y1 , ..., Yn are i.i.d. :


δσ 1~2
Yi œ
δσ 1~2

We can rewrite: Yi δ  σXi

18 / 80
Brownian motion

Brownian Motion: Motivation

Binomial model on several periods (discrete time model):


Deterministic model (not constant return ik on period k ):

S ˆn  S ˆ0 M1n

k 1
ˆ  ik  S ˆ0  Me
n

k 1
δk

Stochastic model :
S ˆn  S ˆ0 Me
n

k 1
Yk

où Y1 , ..., Yn are i.i.d. :


δσ 1~2
Yi œ
δσ 1~2

1 1~2
We can rewrite: Yi δ  σXi where Xi œ .
 1 1~2
18 / 80
Brownian motion

Brownian Motion: Motivation

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯


determ. trend

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯
 σ QXn

i 1
i
determ. trend
´¹¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹¶
random walk

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯
 σ QXn

i 1
i
determ. trend
´¹¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹¶
random walk

In particular,

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯
 σ QXn

i 1
i
determ. trend
´¹¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹¶
random walk

In particular, the moments of cumulative log-returns are given by:

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯
 σ QXn

i 1
i
determ. trend
´¹¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹¶
random walk

In particular, the moments of cumulative log-returns are given by:


E log ˆS ˆn~S ˆ0 δn

19 / 80
Brownian motion

Brownian Motion: Motivation

We hence have :

S ˆn  S ˆ0 Me 
n

k 1
δ σXk
S ˆ0enδσ Pk
n
1 Xk

We hence obtain the cumulative log-return until n:

logˆS ˆn~S ˆ0 QY


n

i 1
i δn
¯
 σ QXn

i 1
i
determ. trend
´¹¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹¶
random walk

In particular, the moments of cumulative log-returns are given by:


E log ˆS ˆn~S ˆ0 δn
var log ˆS ˆn~S ˆ0 σ 2 n
(variance proportional to time)
19 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t,

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

W ˆn  ∆x. QX ,
m

i 1
i with m n~∆t

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

W ˆn  ∆x. QX ,
m

i 1
i with m n~∆t

and in terms of moments:

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

W ˆn  ∆x. QX ,
m

i 1
i with m n~∆t

and in terms of moments:


E W ˆn 0

20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

W ˆn  ∆x. QX ,
m

i 1
i with m n~∆t

and in terms of moments:


E W ˆn 0
2
Var W ˆn mˆ∆x 2 n ˆ∆x
∆t


20 / 80
Brownian motion

Brownian Motion: Motivation

We will now consider passing to a continuous time model.


This will be done in two steps:
1 Modification of the (space and time) scale of the random
walk:
The time periods of 1 are replaced by ∆t, and jumps of +1 and -1
are replaced by jumps of ∆x and ∆x on each period of time.
The random walk then becomes:

W ˆn  ∆x. QX ,
m

i 1
i with m n~∆t

and in terms of moments:


E W ˆn 0
2
Var W ˆn mˆ∆x 2 n ˆ∆x
∆t


20 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk with ∆xk , ∆tk 0

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk with ∆xk , ∆tk 0 in


such a way that the limiting process is non-trivial (i.e. non zero
variance).

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk with ∆xk , ∆tk 0 in


such a way that the limiting process is non-trivial (i.e. non zero
variance).
For this purpose, we will require:

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk with ∆xk , ∆tk 0 in


such a way that the limiting process is non-trivial (i.e. non zero
variance).
For this purpose, we will require:

ˆ∆xk 2
1
∆tk

21 / 80
Brownian motion

Brownian Motion: Motivation

2 Passing to the limit ∆x, ∆t 0

We will now consider a sequence of ∆xk , ∆tk with ∆xk , ∆tk 0 in


such a way that the limiting process is non-trivial (i.e. non zero
variance).
For this purpose, we will require:

ˆ∆xk 2
1
∆tk
∆xk
(and not ∆tk 1 !)

21 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

22 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

23 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

24 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

25 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

26 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

27 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

28 / 80
Brownian motion

Brownian Motion: Motivation

Construction of BM as limit of a random walk: illustration


3

1
Xk(t)

-1

-2

-3
0 1 2 3 4 5 6
Time t

29 / 80
Brownian motion

Brownian Motion: Motivation

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process
is called:

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process
is called:
Standard Brownian motion, or

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process
is called:
Standard Brownian motion, or
Wiener process, or

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process
is called:
Standard Brownian motion, or
Wiener process, or
"infinitesimal random walk"

30 / 80
Brownian motion

Brownian Motion: Motivation

One can show that the sequence of processes ˆW ˆk  ˆt  defined


above converges (in distribution) to a process with continuous
trajectories called Brownian motion:

QX
t ~∆tk
W ˆk  ˆ t  ∆xk i W ˆt 
i 1

where the limit process W ˆt  :


is a continuous time process
is called:
Standard Brownian motion, or
Wiener process, or
"infinitesimal random walk"

30 / 80
Brownian motion

Brownian Motion: Motivation

This convergence is actually the dynamic equivalent of the central limit


theorem in the static case:

31 / 80
Brownian motion

Brownian Motion: Motivation

This convergence is actually the dynamic equivalent of the central limit


theorem in the static case:

P
If ˆXn  is a sequence of i.i.d. random variables of mean µ and finite
n
variance σ 2 , then we note Sn i 1 Xi , and we have:

31 / 80
Brownian motion

Brownian Motion: Motivation

This convergence is actually the dynamic equivalent of the central limit


theorem in the static case:

P
If ˆXn  is a sequence of i.i.d. random variables of mean µ and finite
n
variance σ 2 , then we note Sn i 1 Xi , and we have:

Sn  nµ d
º N ˆ0, 1
σ n

31 / 80
Brownian motion

Brownian Motion: Motivation

By construction, this process W ˆt  has the following properties:

32 / 80
Brownian motion

Brownian Motion: Motivation

By construction, this process W ˆt  has the following properties:


(i) E W ˆk  ˆt  0 Ð E W ˆt  0

32 / 80
Brownian motion

Brownian Motion: Motivation

By construction, this process W ˆt  has the following properties:


(i) E W ˆk  ˆt  0 Ð E W ˆt  0
(ii) var W ˆk 
ˆt  t
ˆ∆Xk 2
∆tk Ð var W ˆt  t

32 / 80
Brownian motion

Brownian Motion: Motivation

By construction, this process W ˆt  has the following properties:


(i) E W ˆk  ˆt  0 Ð E W ˆt  0
(ii) var W ˆk 
ˆt  t
ˆ∆Xk 2
∆tk Ð var W ˆt  t
ˆk 
(iii) W ˆt  = sum of i.i.d. rvs Ð W ˆt   N ˆ0, t  (Central Limit
Theorem)

32 / 80
Brownian motion

Brownian Motion: Motivation

By construction, this process W ˆt  has the following properties:


(i) E W ˆk  ˆt  0 Ð E W ˆt  0
(ii) var W ˆk 
ˆt  t
ˆ∆Xk 2
∆tk Ð var W ˆt  t
ˆk 
(iii) W ˆt  = sum of i.i.d. rvs Ð W ˆt   N ˆ0, t  (Central Limit
Theorem)
(iv) W is a process with independent and stationary increments

32 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent,

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:0 B t1 @ t2 @ ... @ tn B T , the random variables
Bt2  Bt1 ,

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:0 B t1 @ t2 @ ... @ tn B T , the random variables
Bt2  Bt1 , Bt3  Bt2 , ...

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:0 B t1 @ t2 @ ... @ tn B T , the random variables
Bt2  Bt1 , Bt3  Bt2 , ..., Btn  Btn1

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:0 B t1 @ t2 @ ... @ tn B T , the random variables
Bt2  Bt1 , Bt3  Bt2 , ..., Btn  Btn1
are independent.

33 / 80
Brownian motion

Brownian Motion: Definition

All these motivate the following definition:


Definition (Standard Brownian Motion)
A stochastic process ˜Bt  0 B t B T  in continuous time is standard
Brownian motion on 0, T  if
1 B0 0.
2 The increments of Bt are independent, i.e. for all finite set of
instants:0 B t1 @ t2 @ ... @ tn B T , the random variables
Bt2  Bt1 , Bt3  Bt2 , ..., Btn  Btn1
are independent.
3 For all 0 B s @ t B T , the increments Bt  Bs have a Gaussian
distribution of mean 0 and variance t  s.
4 Bt ˆω  is a continuous function of t for almost all ω.

33 / 80
Brownian motion

Brownian Motion: Illustration

A typical sample path:


1.5

0.5

0
X(t)

-0.5

-1

-1.5

-2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t

34 / 80
Brownian motion

Brownian Motion: Illustration

A typical sample path:


1.5

0.5

0
X(t)

-0.5

-1

-1.5

-2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t

It can be shown that the sample paths (a.s.) are nowhere differentiable,
although continuous everywhere. 34 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 
where ˜Bt  is a standard BM

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 
where ˜Bt  is a standard BM

Brownian motion from level a, with drift µ and volatility σ

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 
where ˜Bt  is a standard BM

Brownian motion from level a, with drift µ and volatility σ

Z ˆt   a  µt  σBt

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 
where ˜Bt  is a standard BM

Brownian motion from level a, with drift µ and volatility σ

Z ˆt   a  µt  σBt
In particular,
E Z ˆt  a  µt

35 / 80
Brownian motion

Non standard Brownian Motion

Brownian motion from level a (instead of 0):

Z ˆ t   a  B ˆt 
where ˜Bt  is a standard BM

Brownian motion from level a, with drift µ and volatility σ

Z ˆt   a  µt  σBt
In particular,
E Z ˆt  a  µt

var Z ˆt  σ2t

35 / 80
Brownian motion

Non standard Brownian Motion: Illustration

Brownian motion with drift 0.07 volatility 0.2


1

0.8

0.6

0.4

0.2
X(t)

−0.2

−0.4

−0.6

−0.8

−1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t

36 / 80
Brownian motion

Non standard Brownian Motion: Illustration

Brownian motion with drift 0 volatility 0.35


1

0.8

0.6

0.4

0.2
X(t)

−0.2

−0.4

−0.6

−0.8

−1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time t

37 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T
2 ™Bt2  t  0 B t B Tž

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T
2 ™Bt2  t  0 B t B Tž
3 ™exp ‰αBt  α2 t ~2Ž  0 B t B T ž, where α is a constant in R

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T
2 ™Bt2  t  0 B t B Tž
3 ™exp ‰αBt  α2 t ~2Ž  0 B t B T ž, where α is a constant in R
are martingales

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T
2 ™Bt2  t  0 B t B Tž
3 ™exp ‰αBt  α2 t ~2Ž  0 B t B T ž, where α is a constant in R
are martingales w.r.t. process ˜Bt  .

38 / 80
Brownian motion

Martingales and Brownian motion

Theorem
Let ˜Bt  0 B t B T  be a standard Brownian motion. Then the
processes:
1 ˜Bt  0 B t B T
2 ™Bt2  t  0 B t B Tž
3 ™exp ‰αBt  α2 t ~2Ž  0 B t B T ž, where α is a constant in R
are martingales w.r.t. process ˜Bt  .

38 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2  t is a function of ˆBs sBt ,

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s 

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

E ˆBt s  SBu , u B s   t  Bs2  2Bs2


2

t  s  t  Bs2 Bs2  s.

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

E ˆBt s  SBu , u B s   t  Bs2  2Bs2


2

t  s  t  Bs2 Bs2  s.
3) E exp ‰αBt  α2 t ~2Ž SBu , u B s 

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

E ˆBt s  SBu , u B s   t  Bs2  2Bs2


2

t  s  t  Bs2 Bs2  s.
3) E exp ‰αBt  α2 t ~2Ž SBu , u B s 
α2 α2 s
E exp Šα ˆBt  Bs   2
ˆt  s SBu , u B s  exp ŠαBs  2


39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

E ˆBt s  SBu , u B s   t  Bs2  2Bs2


2

t  s  t  Bs2 Bs2  s.
3) E exp ‰αBt  α2 t ~2Ž SBu , u B s 
α2 α2 s
E exp Šα ˆBt 
2
ˆt  s  SBu , u B s  exp ŠαBs
Bs   
2

α2 2
E exp ˆαBt s  exp Š 2 ˆt  s exp ŠαBs  α2 s 

39 / 80
Brownian motion

Martingales and Brownian motion

Proof
1) Bt is a function of ˆBs sBt ,
E Bt2  t @ ª,
E Bt  Bs SBu , u B s  0 since this is equal to E Bt  Bs , which is 0.
2) Bt2 t is a function of ˆBs s Bt ,

E TBt2  t T B E Bt2   t 2t,


E Bt2  t SBu , u B s  Bs2  2Bt Bs  t SBu , u B s 
2
E ˆBt  Bs  

E ˆBt s  SBu , u B s   t  Bs2  2Bs2


2

t  s  t  Bs2 Bs2  s.
3) E exp ‰αBt  α2 t ~2Ž SBu , u B s 
α2 α2 s
E exp Šα ˆBt 
2
ˆt  s  SBu , u B s  exp ŠαBs
Bs   
2

α2 2
E exp ˆαBt s  exp Š 2 ˆt  s exp ŠαBs  α2 s 
2
exp ŠαBs  α2 s  .

39 / 80
Brownian motion

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

40 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy can be modelled by a vector φt describing the
quantities invested in each of the assets at instant t: φt ˆφ1t , ..., φdt 

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy can be modelled by a vector φt describing the
quantities invested in each of the assets at instant t: φt ˆφ1t , ..., φdt 
The value at t of the portfolio obtained by following this strategy is then
given by:

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy can be modelled by a vector φt describing the
quantities invested in each of the assets at instant t: φt ˆφ1t , ..., φdt 
The value at t of the portfolio obtained by following this strategy is then
given by:

Qφ S
d
Vt ˆφ φ1t St1  φ2t St2  ...  φdt Std k
t
k
t φt .St
k 1

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy can be modelled by a vector φt describing the
quantities invested in each of the assets at instant t: φt ˆφ1t , ..., φdt 
The value at t of the portfolio obtained by following this strategy is then
given by:

Qφ S
d
Vt ˆφ φ1t St1  φ2t St2  ...  φdt Std k
t
k
t φt .St
k 1

We will denote by 0 T0 , T1 , ..., Tn1 T the rebalancing instants:

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Let us consider a market composed of d assets of price at t :


St ˆSt1 , ..., Std , modeled like a vector process
A trading strategy can be modelled by a vector φt describing the
quantities invested in each of the assets at instant t: φt ˆφ1t , ..., φdt 
The value at t of the portfolio obtained by following this strategy is then
given by:

Qφ S
d
Vt ˆφ φ1t St1  φ2t St2  ...  φdt Std k
t
k
t φt .St
k 1

We will denote by 0 T0 , T1 , ..., Tn1 T the rebalancing instants:


between 2 such instants, the composition of the portfolio is supposed
unchanged.

41 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1 d


i , ..., φi  the composition of the portfolio
between Ti and Ti 1 ,

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

Generally, the rebalancing instants Ti are random (depending e.g. from


the levels reached by some assets in the market...).

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

Generally, the rebalancing instants Ti are random (depending e.g. from


the levels reached by some assets in the market...). The strategy φt
appears as a stochastic process.

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

Generally, the rebalancing instants Ti are random (depending e.g. from


the levels reached by some assets in the market...). The strategy φt
appears as a stochastic process.

Definition (Simple predictable process)


A process φt (t > 0, T ) that can be represented by ˆ‡

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

Generally, the rebalancing instants Ti are random (depending e.g. from


the levels reached by some assets in the market...). The strategy φt
appears as a stochastic process.

Definition (Simple predictable process)


A process φt (t > 0, T ) that can be represented by ˆ‡ is called a simple
predictable process

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

We can then denote by φi ˆφ1i , ..., φdi  the composition of the portfolio
between Ti and Ti 1 , and rewrite φt as a step function:

φt φ0 I t 0  φ1 IT0 ,T1   φ2 IT1 ,T2   ...  φn ITn ,T  ˆ‡

Generally, the rebalancing instants Ti are random (depending e.g. from


the levels reached by some assets in the market...). The strategy φt
appears as a stochastic process.

Definition (Simple predictable process)


A process φt (t > 0, T ) that can be represented by ˆ‡ is called a simple
predictable process if moreover the stochastic instants Ti are stopping
times, and if each φi is a bounded random variables whose value is revealed
at Ti (φi > FTi ).

42 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Example of simple predictable process (one of its component):

43 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”),

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi 

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

Remark that here we assumed implicitly that there was no external


money injected in the process, and no consumption: the origin of
the gains is only due to market fluctuations.

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

Remark that here we assumed implicitly that there was no external


money injected in the process, and no consumption: the origin of
the gains is only due to market fluctuations. We say that the
strategy is self-financing .

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

Remark that here we assumed implicitly that there was no external


money injected in the process, and no consumption: the origin of
the gains is only due to market fluctuations. We say that the
strategy is self-financing .
Hence the accumulated capital on 0, T  of the trader beginning
with an initial composition given by φ0 ,

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

Remark that here we assumed implicitly that there was no external


money injected in the process, and no consumption: the origin of
the gains is only due to market fluctuations. We say that the
strategy is self-financing .
Hence the accumulated capital on 0, T  of the trader beginning
with an initial composition given by φ0 , and due to fluctuations of
the market, can be written as:

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Gain process for a simple trading strategy


If ˆφt  is simple (we speak about a “simple trading strategy”), then
the gain realized by the trader between Ti and Ti 1 is equal to:
φ1i ˆST1i 1  ST1i   φ2i ˆST2i 1  ST2i   ...  φdi ˆSTdi 1  STdi  φi .ˆSTi 1  STi 

Remark that here we assumed implicitly that there was no external


money injected in the process, and no consumption: the origin of
the gains is only due to market fluctuations. We say that the
strategy is self-financing .
Hence the accumulated capital on 0, T  of the trader beginning
with an initial composition given by φ0 , and due to fluctuations of
the market, can be written as:

GT ˆφ φ0 .S0  Qφ . S
n

i 0
i ˆ Ti  1  STi 

44 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Definition (Stochastic integral of a simple predictable process)

45 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Definition (Stochastic integral of a simple predictable process)


By definition, GT ˆφ is called the

45 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Definition (Stochastic integral of a simple predictable process)


By definition, GT ˆφ is called the stochastic integral of the simple
predictable process φ ˆφt  w.r.t. the price process ˆSt , and will be
denoted by :

45 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Definition (Stochastic integral of a simple predictable process)


By definition, GT ˆφ is called the stochastic integral of the simple
predictable process φ ˆφt  w.r.t. the price process ˆSt , and will be
denoted by :
S 0
T
φˆu .dS ˆu 

45 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Definition (Stochastic integral of a simple predictable process)


By definition, GT ˆφ is called the stochastic integral of the simple
predictable process φ ˆφt  w.r.t. the price process ˆSt , and will be
denoted by :
S 0
T
φˆu .dS ˆu 

In finance, the accumulated capital of a trader following a


strategy lead naturally to a new concept of integral: the
stochastic integral

45 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Now, in view of getting some results concerning option prices in a


given stochastic model, we cannot limit ourselves to the case
where φ is a simple strategy.

46 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Now, in view of getting some results concerning option prices in a


given stochastic model, we cannot limit ourselves to the case
where φ is a simple strategy.
Hence we need to give a sense to this stochastic integral / gain
process, also in the case where the composition ˆφt  continuously
changes with time.

46 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Now, in view of getting some results concerning option prices in a


given stochastic model, we cannot limit ourselves to the case
where φ is a simple strategy.
Hence we need to give a sense to this stochastic integral / gain
process, also in the case where the composition ˆφt  continuously
changes with time.
We will first give a sense to this integral in the case where ˆSt  is
a Brownian motion of dimension 1, and where φt is a more
general process (not necessarily simple):

46 / 80
Stochastic Integral Stochastic integral - Motivation

Stochastic integral - Motivation

Now, in view of getting some results concerning option prices in a


given stochastic model, we cannot limit ourselves to the case
where φ is a simple strategy.
Hence we need to give a sense to this stochastic integral / gain
process, also in the case where the composition ˆφt  continuously
changes with time.
We will first give a sense to this integral in the case where ˆSt  is
a Brownian motion of dimension 1, and where φt is a more
general process (not necessarily simple):

S 0
T
φˆsdBs

46 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

ΠS
0
T
φˆsdBs ‘ ˆω 

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

ΠS
0
T
φˆsdBs ‘ ˆω  lim
maxˆti ti 1  0 i 1

n
ti 1 ˆω  Bti ˆω   Bti 1 ˆω 

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

ΠS
0
T
φˆsdBs ‘ ˆω  lim
maxˆti ti 1  0 i 1

n
ti 1 ˆω  Bti ˆω   Bti 1 ˆω 

The problem is that (almost all) trajectories of a Brownian motion


are very irregular ...

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

ΠS
0
T
φˆsdBs ‘ ˆω  lim
maxˆti ti 1  0 i 1

n
ti 1 ˆω  Bti ˆω   Bti 1 ˆω 

The problem is that (almost all) trajectories of a Brownian motion


are very irregular ... (more precisely, have no bounded variation)

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

A first naive idea would consist to define the stochastic integral as


a limit obtained along each trajectory:

ΠS
0
T
φˆsdBs ‘ ˆω  lim
maxˆti ti 1  0 i 1

n
ti 1 ˆω  Bti ˆω   Bti 1 ˆω 

The problem is that (almost all) trajectories of a Brownian motion


are very irregular ... (more precisely, have no bounded variation)
and that this limit hence does not exist in general for a given
trajectory ω!

47 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti .

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti . In the Itô integral, we will fix yi ti 1 ;

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti . In the Itô integral, we will fix yi ti 1 ;
The stochastic integral will be defined as a limit of Riemann sums,

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti . In the Itô integral, we will fix yi ti 1 ;
The stochastic integral will be defined as a limit of Riemann sums, but
not in the sense of "almost sure convergence",

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti . In the Itô integral, we will fix yi ti 1 ;
The stochastic integral will be defined as a limit of Riemann sums, but
not in the sense of "almost sure convergence", but in "quadratic
convergence" (L2 ).

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

Contrarily to the Riemann integral, the result can depend on the point
yi > ti 1 , ti . In the Itô integral, we will fix yi ti 1 ;
The stochastic integral will be defined as a limit of Riemann sums, but
not in the sense of "almost sure convergence", but in "quadratic
convergence" (L2 ).
Let us recall that

Xn X in quadratic mean (or in L2  iff

YXn  X Y2L2 E ˆXn  X 2  0

48 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

The Itô integral is actually defined following a rigorous mathematical


construction, that we will not recall here.

49 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

The Itô integral is actually defined following a rigorous mathematical


construction, that we will not recall here. We can get a good intuition of
it using the following result (Riemann’s representation):
If f  R R continuous,

49 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

The Itô integral is actually defined following a rigorous mathematical


construction, that we will not recall here. We can get a good intuition of
it using the following result (Riemann’s representation):
If f  R R continuous, and if we consider the partition of 0, T  given
by ti i Tn with 0 B i B n,

49 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

The Itô integral is actually defined following a rigorous mathematical


construction, that we will not recall here. We can get a good intuition of
it using the following result (Riemann’s representation):
If f  R R continuous, and if we consider the partition of 0, T  given
by ti i Tn with 0 B i B n, then the Itô integral of f ˆBt  w.r.t. ˆBt  is
obtained as the L2 limit of Riemann sums:

49 / 80
Stochastic Integral Stochastic integral - Definition

Stochastic integral

The Itô integral is actually defined following a rigorous mathematical


construction, that we will not recall here. We can get a good intuition of
it using the following result (Riemann’s representation):
If f  R R continuous, and if we consider the partition of 0, T  given
by ti i Tn with 0 B i B n, then the Itô integral of f ˆBt  w.r.t. ˆBt  is
obtained as the L2 limit of Riemann sums:

Qf
n

i 1
‰Bti 1 Ž ‰Bti  Bti 1 Ž ÐL
2

S 0
T
f ˆBs  dBs

where the limit is taken in the sense of the quadratic convergence.

49 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function

50 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function


Proposition (Gaussian integrals)
Let f be a continuous deterministic function,

50 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function


Proposition (Gaussian integrals)
Let f be a continuous deterministic function, then

S
t
Xt  f ˆsdBs
0

has a Gaussian distribution

50 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function


Proposition (Gaussian integrals)
Let f be a continuous deterministic function, then

S
t
Xt  f ˆsdBs
0

has a Gaussian distribution with zero mean and variance:

S
t
VarˆXt  f 2 ˆu du.
0

50 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function


Proposition (Gaussian integrals)
Let f be a continuous deterministic function, then

S
t
Xt  f ˆsdBs
0

has a Gaussian distribution with zero mean and variance:

S
t
VarˆXt  f 2 ˆu du.
0

Moreover, if we consider the partition of 0, T  given by ti iT


n
for 0 B i B n and
ti‡ > ti 1 , ti , then

50 / 80
Stochastic Integral Stochastic integral: properties

Integral of a deterministic function


Proposition (Gaussian integrals)
Let f be a continuous deterministic function, then

S
t
Xt  f ˆsdBs
0

has a Gaussian distribution with zero mean and variance:

S
t
VarˆXt  f 2 ˆu du.
0

Moreover, if we consider the partition of 0, T  given by ti iT


n
for 0 B i B n and
ti‡ > ti 1 , ti , then

Q f t‡ S
n T
lim ˆ i  ˆBti  Bti 1  f ˆsdBs ,
n ª i 1 0

where the limit is taken in the sense of quadratic convergence and


convergence in probability.

We see that here the choice of ti‡ > ti 1 , ti  is free: it does not need to be ti 501 ./ 80
Stochastic Integral Stochastic integral: properties

Other properties

Other important properties of the Itô integral are as follows:

R
The Itô integral is defined in general for a square integrable
t
process Xt . The resulting process, 0 X ˆsdB ˆs is a continuous
martingale w.r.t. ˆBt .1

1
Actually it is equal almost surely to a continuous martingale
51 / 80
Stochastic Integral Stochastic integral: properties

Other properties

Other important properties of the Itô integral are as follows:

R
The Itô integral is defined in general for a square integrable
t
process Xt . The resulting process, 0 X ˆsdB ˆs is a continuous
martingale w.r.t. ˆBt .1
(Itô isometry)
For all 0 B s B t and for all ˆXt  square integrable, we have:

S S
t 2 t
E ‹ X ˆω, u  dBu   E X ˆω, u 2 du  .(1)
s s

1
Actually it is equal almost surely to a continuous martingale
51 / 80
Stochastic Integral Stochastic integral: properties

Other properties

Other important properties of the Itô integral are as follows:

R
The Itô integral is defined in general for a square integrable
t
process Xt . The resulting process, 0 X ˆsdB ˆs is a continuous
martingale w.r.t. ˆBt .1
(Itô isometry)
For all 0 B s B t and for all ˆXt  square integrable, we have:

S S
t 2 t
E ‹ X ˆω, u  dBu   E X ˆω, u 2 du  .(1)
s s

This equality is also true actually if considering conditional


expectations knowing Bs .

1
Actually it is equal almost surely to a continuous martingale
51 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example:

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

QBi
ti 1 ‰∆i B Ž

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

QBi
ti 1 ‰∆i B Ž QBi
‰ ti 1 Bti  Bt2i 1 Ž

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

QBi
ti 1 ‰∆i B Ž QB B
i
‰ ti 1 ti  Bt2i 1 Ž

1
2
Q B i
2
‰ ti  Bt2i 1 Ž  ‰Bti 
2
Bti 1 Ž 

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

QBi
ti 1 ‰∆i B Ž QB B
i
‰ ti 1 ti  Bt2i 1 Ž

1
2
Q B i
‰ ti
2
 Bt2i 1 Ž  ‰Bti 
2
Bti 1 Ž 

1 2
2
2
‰Bt  B0 Ž 
1
2
Q ∆B
i
‰ i Ž
2

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

In some cases, this integral can be explicitly computed using the


Riemann representation.

Example: R 0
t
Bs dBs ?

If we denote ∆i B  Bti  Bti 1 and ∆i t  ti  ti 1 , we have:

QBi
ti 1 ‰∆i B Ž QB B
i
‰ ti 1 ti  Bt2i 1 Ž

1
2
Q B i
‰ ti
2
 Bt2i 1 Ž  ‰Bti  Bti 1 Ž
2


1 2
2
2
‰Bt  B0 Ž 
1
2
Q ∆B
i
‰ i Ž
2 1 2 1
B 
2 t 2
Q ∆B
i
‰ i Ž
2
,

52 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

Q ∆B
We can see that
2
‰ i Ž t
i

in L2 .

53 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

Q ∆B
We can see that
2
‰ i Ž t
i

in L2 . We then get:

53 / 80
Stochastic Integral Stochastic integral: properties

Stochastic integral: explicit calculation

Q ∆B
We can see that
2
‰ i Ž t
i

in L2 . We then get:
S 0
t
Bs dBs
1 2 1
B  t.
2 t 2

Remark
We just saw that “ˆ∆i B 2 behaves like ∆i t”, i.e., formally:
2
ˆdBt  dt

53 / 80
Stochastic Integral Stochastic integral: properties

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

54 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:

55 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:


b bˆs, ω  adapted integrable process

55 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:


b bˆs, ω  adapted integrable process
σ σ ˆs, ω  adapted square-integrable process

55 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:


b bˆs, ω  adapted integrable process
σ σ ˆs, ω  adapted square-integrable process
We can hence construct a new process X ˆt  by integration of b and σ:

55 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:


b bˆs, ω  adapted integrable process
σ σ ˆs, ω  adapted square-integrable process
We can hence construct a new process X ˆt  by integration of b and σ:

X ˆt  X ˆ0  S 0
t
bˆsds  S 0
t
σ ˆsdB ˆs

55 / 80
Itô formula and Stochastic Differential Equations Differential notation

Stochastic differential: notation

Introduction of the differential notation:


b bˆs, ω  adapted integrable process
σ σ ˆs, ω  adapted square-integrable process
We can hence construct a new process X ˆt  by integration of b and σ:

X ˆt  X ˆ0  S 0
t
bˆsds  S 0
t
σ ˆsdB ˆs

We say that X is an Itô process, and we use the notation

dX ˆt  bˆt dt  σ ˆt dB ˆt 

55 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

The Itô stochastic integral leads to other calculation rules than in the
classical calculus:
Classical calculus:
1 2
2
f ˆt  S 0
t
f ˆsdf ˆs

Stochastic calculus:
1 2
2
B ˆt 
??
S 0
t
B ˆsdB ˆs ??

56 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

One can easily see that:

S 0
t
B ˆs dB ˆs
1 2
2
1
B ˆt   t
2

57 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

One can easily see that:

S 0
t
B ˆs dB ˆs
1 2
2
1
B ˆt   t
2
 2 S0
t
B ˆs dB ˆs  S 0
t
ds B 2 ˆt 

57 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

One can easily see that:

S 0
t
B ˆs dB ˆs
1 2
2
1
B ˆt   t
2
 2 S0
t
B ˆs dB ˆs  S 0
t
ds B 2 ˆt 

which means by using differential notations:

57 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

One can easily see that:

S 0
t
B ˆs dB ˆs
1 2
2
1
B ˆt   t
2
 2 S0
t
B ˆs dB ˆs  S 0
t
ds B 2 ˆt 

which means by using differential notations:

d ˆB 2 ˆt  2B ˆt dB ˆt   dt

57 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential: calculation rules

One can easily see that:

S 0
t
B ˆs dB ˆs
1 2
2
1
B ˆt   t
2
 2 S0
t
B ˆs dB ˆs  S 0
t
ds B 2 ˆt 

which means by using differential notations:

d ˆB 2 ˆt  2B ˆt dB ˆt   dt
the usual chain rule does not work anymore...

57 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula: case of f ˆBt 

Theorem (Itô formula)

58 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula: case of f ˆBt 

Theorem (Itô formula)

Let f  R R twice differentiable, then a.s. for all t > 0, T :

58 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula: case of f ˆBt 

Theorem (Itô formula)

Let f  R R twice differentiable, then a.s. for all t > 0, T :

f ˆBt  f ˆ0   S 0
t
f œ ˆBs  dBs 
1
2
S
0
t
f œœ ˆBs  ds. (2)

58 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula: case of f ˆBt 

Theorem (Itô formula)

Let f  R R twice differentiable, then a.s. for all t > 0, T :

f ˆBt  f ˆ0   S 0
t
f œ ˆBs  dBs 
1
2
S
0
t
f œœ ˆBs  ds. (2)

In differential notations:
1 œœ
d ˆf ˆBt  f œ ˆBt dBt  f ˆBt dt
2
The proof is a consequence of Taylor development of order 2 and of
the fact that a Brownian motion has a quadratic variation on 0, T 
equal to T , leading to an additional term w.r.t standard calculus.

58 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x ,

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt .

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt . We will assume that f is twice differentiable.

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt . We will assume that f is twice differentiable.
Theorem (Itô formula in dim 2)

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt . We will assume that f is twice differentiable.
Theorem (Itô formula in dim 2)

For each function f twice differentiable, we have the representation

f ˆt, Bt  f ˆ0, 0  S t

0 ∂x
∂f
ˆs, Bs  dBs

 S 0
t ∂f
∂t
ˆs, Bs  ds 
1 t ∂2f
S
2 0 ∂x 2
ˆs, Bs  ds a.s.

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt . We will assume that f is twice differentiable.
Theorem (Itô formula in dim 2)

For each function f twice differentiable, we have the representation

f ˆt, Bt  f ˆ0, 0  S t

0 ∂x
∂f
ˆs, Bs  dBs

 S 0
t ∂f
∂t
ˆs, Bs  ds 
1 t ∂2f
S
2 0 ∂x 2
ˆs, Bs  ds a.s.

In differential notation, this becomes:

59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Itô formula with time and space variables

We now consider a function of two variables: f ˆt, x , and we are


interested in d ˆf ˆt, Bt . We will assume that f is twice differentiable.
Theorem (Itô formula in dim 2)

For each function f twice differentiable, we have the representation

f ˆt, Bt  f ˆ0, 0  S t

0 ∂x
∂f
ˆs, Bs  dBs

 S 0
t ∂f
∂t
ˆs, Bs  ds 
1 t ∂2f
S
2 0 ∂x 2
ˆs, Bs  ds a.s.

In differential notation, this becomes:

∂f ∂f 1 ∂2f
d ˆf ˆt, Bt  ˆt, Bt  dBt  ˆt, Bt  dt  ˆt, Bt  dt
∂x ∂t 2 ∂x 2
59 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

Remark that the calculation rule is identical if we consider f ˆt, Xt 


where dXt bˆt dt  σ ˆt dB ˆt  (with b, σ satisfying the ad-hoc
integrability properties).
Actually, the different formulations of the Itô formula suggest the
following formal calculation rules:

60 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

Remark that the calculation rule is identical if we consider f ˆt, Xt 


where dXt bˆt dt  σ ˆt dB ˆt  (with b, σ satisfying the ad-hoc
integrability properties).
Actually, the different formulations of the Itô formula suggest the
following formal calculation rules:

dt.dt 0
dB ˆt .dB ˆt  dt
dB ˆt .dt 0

60 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

Remark that the calculation rule is identical if we consider f ˆt, Xt 


where dXt bˆt dt  σ ˆt dB ˆt  (with b, σ satisfying the ad-hoc
integrability properties).
Actually, the different formulations of the Itô formula suggest the
following formal calculation rules:

dt.dt 0
dB ˆt .dB ˆt  dt
dB ˆt .dt 0
We recover the Itô formula by applying a Taylor development of order 2
combined with those rules

60 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

dYt

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

∂f
dYt ˆt, Bt  dt 
∂t

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

∂f ∂f
dYt ˆt, Bt  dt  ˆt, Bt  dBt 
∂t ∂x

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

∂f ∂f 1 ∂2f
dYt ˆt, Bt  dt  ˆt, Bt  dBt  ˆt, Bt  dBt dBt
∂t ∂x 2 ∂x 2

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Formal calculation rules

E.g. Case Y ˆt  f ˆt, B ˆt :

∂f ∂f 1 ∂2f
dYt ˆt, Bt  dt  ˆt, Bt  dBt  ˆt, Bt  dBt dBt
∂t ∂x 2 ∂x 2
1 ∂2f ∂2f
 ˆt, Bt  dt.dt  ˆt, Bt  dt.dB ˆt 
2 ∂t 2 ∂x∂t

∂f ∂f 1 ∂2f
ˆt, Bt  dt  ˆt, Bt  dBt  ˆt, Bt  dt
∂t ∂x 2 ∂x 2

61 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential equations

Stochastic calculus allows to express efficiently the dynamics of


stochastic processes, given in the form of stochastic differential
equations.
Many continuous time financial models are expressed by means of
such equations.
For instance, the stock price in Black-Scholes model is assumed to
satisfy the following SDE:

dSt µSt dt  σSt dBt

whose solution is easily obtained by applying Itô lemma to the new


process Yt lnˆSt :
St S0 eˆµ 2 σ t σBt
1 2

62 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic differential equations

Another important example is the Ornstein-Uhlenbeck process, behind


Vasicek model:
dXt aˆb  Xt dt  σdBt
The solution of this equation can be easily obtained up to an Itô
integral, allowing a lot of explicit computations.
In general, properties of the Itô integral are useful to get properties of
solutions to these equations, assumed to be satisfied by market
variables (e.g. moments, type of distribution...). These properties are
useful in order to control accuracy of simulations, to ease calibration
on historical data,...

63 / 80
Itô formula and Stochastic Differential Equations Itô formula

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

64 / 80
Valuation and arbitrage Arbitrage opportunity

Arbitrage opportunity

Suppose again that we are in a market with d risky assets and a risk
free asset (risk free rate r ).
An arbitrage opportunity is a self-financing strategy2 is a zero-cost
strategy that can lead to a positive terminal gain, without any
probability of intermediate loss.

2
no consumption, no injection of money
65 / 80
Valuation and arbitrage Arbitrage opportunity

Arbitrage opportunity

Suppose again that we are in a market with d risky assets and a risk
free asset (risk free rate r ).
An arbitrage opportunity is a self-financing strategy2 is a zero-cost
strategy that can lead to a positive terminal gain, without any
probability of intermediate loss.
Formally, if we denote by Vt ˆφ the value of the portfolio obtained
when following this strategy, this means:

P t > 0, T , Vt ˆφ C 0
¦ 1, P VT ˆφ A 0 A 0

P V 0 ˆφ  0 1

2
no consumption, no injection of money
65 / 80
Valuation and arbitrage Arbitrage opportunity

Arbitrage opportunity

Suppose again that we are in a market with d risky assets and a risk
free asset (risk free rate r ).
An arbitrage opportunity is a self-financing strategy2 is a zero-cost
strategy that can lead to a positive terminal gain, without any
probability of intermediate loss.
Formally, if we denote by Vt ˆφ the value of the portfolio obtained
when following this strategy, this means:

P t > 0, T , Vt ˆφ C 0
¦ 1, P VT ˆφ A 0 A 0

P V 0 ˆφ  0 1
N.B.: the assumption of self-financing is important: it is trivial to find strategies which
are not self-financing but verify the property above (exercise: it suffices to inject cash
just before maturity...)
2
no consumption, no injection of money
65 / 80
Valuation and arbitrage Arbitrage opportunity

Arbitrage-free pricing rules

In most pricing models, we will always assume that there is no


arbitrage opportunity.
A first consequence of absence of arbitrage is the law of one price:
two self-financing strategies with the same terminal payoff must have
the same value at all times, otherwise the difference would generate
an arbitrage.
This fact is at the origin of the pricing of financial products by using
replication arguments (replicating or hedging portfolio).

66 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 1: Fundamental theorem of asset pricing

We say that a probability measure Q is equivalent to P is they have the


same set of possible scenarios:

QˆA 0  PˆA 0

An equivalent measure is Q called a martingale measure if discounted


assets (ert St t are martingales under Q.

67 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 1: Fundamental theorem of asset pricing

We say that a probability measure Q is equivalent to P is they have the


same set of possible scenarios:

QˆA 0  PˆA 0

An equivalent measure is Q called a martingale measure if discounted


assets (ert St t are martingales under Q.
In practice, one can see that the expected return under Q of all assets
is then equal to the risk free rate.

67 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 1: Fundamental theorem of asset pricing

We say that a probability measure Q is equivalent to P is they have the


same set of possible scenarios:

QˆA 0  PˆA 0

An equivalent measure is Q called a martingale measure if discounted


assets (ert St t are martingales under Q.
In practice, one can see that the expected return under Q of all assets
is then equal to the risk free rate. This is why such a measure is also
called a risk neutral probability measure.

67 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 1: Fundamental theorem of asset pricing

We say that a probability measure Q is equivalent to P is they have the


same set of possible scenarios:

QˆA 0  PˆA 0

An equivalent measure is Q called a martingale measure if discounted


assets (ert St t are martingales under Q.
In practice, one can see that the expected return under Q of all assets
is then equal to the risk free rate. This is why such a measure is also
called a risk neutral probability measure.

67 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 1: Fundamental theorem of asset pricing

One can show that whatever the model used, a market is arbitrage free
if and only if there exists an equivalent martingale measure3 .
Theorem (First fundamental theorem of asset pricing)
A market model defined by ˆΩ, F , Ft , P and asset prices ˆSt t > 0,T  is
arbitrage free if and only if there exists a probability measure Q  P
such that the discounted asset prices ˆert St  are martingales with
respect to Q.

In that case, one can show that derivative products can be priced
(without arbitrage) by using the expectation under Q of their
discounted payoff: EQ payoff ˆT  erT 

3
see [Harrison-Kreps, 1979], [Harrison-Pliska, 1983], [Delbaen, Schachermayer,
1994]
68 / 80
Valuation and arbitrage Fundamental results of asset pricing

Result 2: Market Completeness

A self-financing strategy ˆφ0t , φt  is a replication strategy for a derivative


product on S 1 , ..., S d delivering a stochastic payoff H at T if:

S S
T T
H V0  φt dSt  φ0t dSt0 P  a.s.
0 0

A market is said complete if any contingent claim admits a replication


strategy.

Theorem (Second fundamental theorem of asset pricing)


An arbitrage free market defined by the assets ˆSt0 , St1 , ..., Std , described as
stochastic processes on ˆΩ, F , Ft , P, is complete if and only if there exists a
unique martingale measure Q  P.

69 / 80
Valuation and arbitrage Fundamental results of asset pricing

In summary: pricing and hedging of contingent assets

1 Equivalence between the no arbitrage opportunity assumption


(NAO) and the existence of a martingale measure / risk-neutral
measure
2 Equivalence between completeness of the market and
uniqueness of the martingale / risk neutral measure
3 In an arbitrage free complete market, the price of any contingent
asset is equal to the expectation under the risk-neutral measure of
its discounted payoff

70 / 80
Valuation and arbitrage Fundamental results of asset pricing

In summary: pricing and hedging of contingent assets

Pricing
In a complete and arbitrage free market, if H is a contingent asset
whose payoff is paid in T , its price Πt ˆH  at instant t is given by:

Πt ˆ H  EQ er ˆT t  H ˆT SFt 

Hedging
In a complete and arbitrage free market, any contingent asset H is
replicable by a self-financing strategy φ‡ :

VT ˆΦ‡ ˆω  H ˆT , ω  ¦ ω>Ω

71 / 80
Valuation and arbitrage Fundamental results of asset pricing

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

72 / 80
Girsanov theorem

Change of measure in a continuous time framework

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model,

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r .

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 
rS ˆt dt  σS ˆt dW ˆt   ˆδ  r S ˆt dt

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 
rS ˆt dt  σS ˆt dW ˆt   ˆδ  r S ˆt dt
rS ˆt dt  σS ˆt dW ‡ ˆt 

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 
rS ˆt dt  σS ˆt dW ˆt   ˆδ  r S ˆt dt
rS ˆt dt  σS ˆt dW ‡ ˆt 

where we denote by

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 
rS ˆt dt  σS ˆt dW ˆt   ˆδ  r S ˆt dt
rS ˆt dt  σS ˆt dW ‡ ˆt 

δr
where we denote by W ‡ ˆt  W ˆt   t
σ

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

When using a geometric Brownian motion like in


Black-Scholes-Merton model, we will try to identify an equivalent
measure, called risk – neutral, under which all financial assets have
an expected return equal to the risk free rate r . Once this measure is
known, the price of derivatives is known (numerically at least).
This will imply a change of drift of the process:

dS ˆt  δS ˆt dt  σS ˆt dW ˆt 
ˆr  δ  r S ˆt dt  σS ˆt dW ˆt 
rS ˆt dt  σS ˆt dW ˆt   ˆδ  r S ˆt dt
rS ˆt dt  σS ˆt dW ‡ ˆt 

δr
where we denote by W ‡ ˆt  W ˆt   t (Brownian motion with drift
σ
under the real world measure P).

73 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is

74 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is not a standard Brownian motion under the real


world probability (presence of a drift).

74 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is not a standard Brownian motion under the real


world probability (presence of a drift).
The Girsanov thm answers to the following questions:

74 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is not a standard Brownian motion under the real


world probability (presence of a drift).
The Girsanov thm answers to the following questions:
How to change measure s.t. W ‡ becomes a standard Brownian
motion under the new measure?

74 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is not a standard Brownian motion under the real


world probability (presence of a drift).
The Girsanov thm answers to the following questions:
How to change measure s.t. W ‡ becomes a standard Brownian
motion under the new measure?
More generally, what becomes the process W when we change
the measure, passing to an equivalent one?

74 / 80
Girsanov theorem

Change of measure in a continuous time framework

The process W ‡ is not a standard Brownian motion under the real


world probability (presence of a drift).
The Girsanov thm answers to the following questions:
How to change measure s.t. W ‡ becomes a standard Brownian
motion under the new measure?
More generally, what becomes the process W when we change
the measure, passing to an equivalent one?
For that purpose, we recall the concept of Radon-Nikodym derivative:
this concept captures the measure change, and allows to make explicit
computations under the new measure Q.

74 / 80
Girsanov theorem

Radon-Nykodym derivative and measure changes

A general result, called the Radon-Nikodym theorem, implies that if a


measure Q is equivalent to a measure P, then there exists a random
variable Z C 0 such that

75 / 80
Girsanov theorem

Radon-Nykodym derivative and measure changes

A general result, called the Radon-Nikodym theorem, implies that if a


measure Q is equivalent to a measure P, then there exists a random
variable Z C 0 such that

QˆA EP IA Z 

75 / 80
Girsanov theorem

Radon-Nykodym derivative and measure changes

A general result, called the Radon-Nikodym theorem, implies that if a


measure Q is equivalent to a measure P, then there exists a random
variable Z C 0 such that

QˆA EP IA Z 

Variable Z is called the Radon-Nikodym derivative of Q with respect


to P, or the density of Q with respect to P.

75 / 80
Girsanov theorem

Radon-Nykodym derivative and measure changes

A general result, called the Radon-Nikodym theorem, implies that if a


measure Q is equivalent to a measure P, then there exists a random
variable Z C 0 such that

QˆA EP IA Z 

Variable Z is called the Radon-Nikodym derivative of Q with respect


to P, or the density of Q with respect to P.
Hence, knowing Z means knowing Q: all expectations under Q can be
computed from expectations under P.
In particular, EP Z  1 (set A Ω in the relation above).

75 / 80
Girsanov theorem

Girsanov theorem

Theorem (Girsanov’s theorem)


Let ˜W ˆt, ω ; t > 0, T , ω > ٝ be a standard Brownian motion defined
on a probability space ˆΩ, F , P.
Let W ‡ be a process defined by:

W ‡ ˆt  W ˆt   θt

76 / 80
Girsanov theorem

Girsanov theorem

Theorem (Girsanov’s theorem)


Let ˜W ˆt, ω ; t > 0, T , ω > ٝ be a standard Brownian motion defined
on a probability space ˆΩ, F , P.
Let W ‡ be a process defined by:

W ‡ ˆt  W ˆt   θt

for some fixed θ in R.

76 / 80
Girsanov theorem

Girsanov theorem

Theorem (Girsanov’s theorem)


Let ˜W ˆt, ω ; t > 0, T , ω > ٝ be a standard Brownian motion defined
on a probability space ˆΩ, F , P.
Let W ‡ be a process defined by:

W ‡ ˆt  W ˆt   θt

for some fixed θ in R. Then the process W ‡ is a standard Brownian


motion under the new measure Q equivalent to P whose density is
defined by:

76 / 80
Girsanov theorem

Girsanov theorem

Theorem (Girsanov’s theorem)


Let ˜W ˆt, ω ; t > 0, T , ω > ٝ be a standard Brownian motion defined
on a probability space ˆΩ, F , P.
Let W ‡ be a process defined by:

W ‡ ˆt  W ˆt   θt

for some fixed θ in R. Then the process W ‡ is a standard Brownian


motion under the new measure Q equivalent to P whose density is
defined by:
θ2
Y ˆT  exp ŒθW ˆT   T ‘
2

76 / 80
agenda

Stochastic Calculus - Outline

1. Probabilitic framework
2. Martingales
3. Brownian Motion
4. Stochastic Integral
5. Itô formula and Stochastic Differential Equations
6. Valuation, arbitrage and martingale measures
7. Girsanov Theorem
8. Black-Scholes model

77 / 80
Black-Scholes model

Black-Scholes’ formula

Assumptions of Black-Scholes’ model


1 The market contains one risky asset (stock), supposed to follow a
geometric Brownian motion (dSt t dt  σSt dWt )
2 The market contains a risk free asset (saving account), with
(continuously compounded) rate of return r
3 The market is perfect
4 No dividend is distributed on the risky asset during the considered
period 4
5 No transaction costs, assets are infinitely divisible, no constraint
on short selling, no constraint for borrowing at the risk-free rate

4
see later for the case with dividends
78 / 80
Black-Scholes model

Black-Scholes formula: approach by martingale measure

To get the price of a European call (strike K , maturity T ) on the stock


in this model, a possible approach is :
1 Identify explicitly an equivalent martingale measure Q (risk neutral
measure)
2 Finding the dynamics of the stock price under that measure Q
3 Compute the price of the call as EQ erT ˆS ˆT   K  
Actually, Girsanov theorem allows to find the measure Q (point 1) and
to get the dynamics of the stock under that measure. It suffices to
apply this result with θ  µσr .
One can moreover show that Black-Scholes model is complete: we
have uniqueness of the risk neutral measure.

79 / 80
Black-Scholes model

Black-Scholes formula: approach by martingale measure

C ˆS, t  er ˆT t  S ª xdF ‡ x


K
ˆ e
 r ˆT  t 
S ª KdF ‡ x
K
ˆ 

F ‡ is the CDF of S ˆT  knowing S ˆt  S, leading to

C ˆS, t  Sֈd1   Ker ˆT t  ֈd2 

with
logˆS ~K ˆr  12 σ 2 ˆT t 
d1 º
σ T t

logˆS ~K ˆr  21 σ 2 ˆT t  º


d2 º d1  σ T  t
σ T t

Black-Scholes’ formula

80 / 80

Vous aimerez peut-être aussi