FM9590 Stochastic Processes with Applications in

Finance and Actuarial Science

Marcos Escobar-Anel ([email protected])

Statistical and Actuarial Sciences,

Western University

SAS. Western, Fall 2024

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 1 / 190
 Introduction

1 Introduction

2 Results from statistics and stochastic analysis

3 Financial markets in continuous time

4 The Black-Scholes formula

5 Stochastic volatility models

6 Numerical Methods.
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science 2 / 190
 Introduction

Details
Lecture time and place:
Thursday 8 : 30 − 11 : 20 am. WSC248.

Office hours: Wednesday, 8 : 00-10 : 00am.

Recommended Textbooks:
Tomas Bjork (2009). Arbitrage Theory in Continuous Time. Oxford University
Press, Oxford.
Bingham & Kiesel (2004): Risk-neutral valuation, 2nd ed.
Note that these textbooks will only be used as a guide. The instructor will use
his own set of course notes during lectures.

Exams and Marking Scheme:

Two Assignments: 30% (equal weight, 27/Sept, 8/Nov)
Midterm: 25% (100 min., 17/Oct)
Final Exam: 45% (3 hrs., 11/Dec)

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 2 / 190
 Introduction

Approximate Course Content, Timing and Exams:

Week Sept 2-6:

σ-algebra, Borel σ−algebra, probability space, random variables, moment
generating and characteristic functions. Radon-Nikodym derivative, conditional
expectations, (natural) filtration, stochastic processes.

Sept 9-13:
Stopping times, modes of convergence. Variations of paths, Quadratic
variation, p− variation, Quadratic covariation, differentiability. Definition of
Brownian motion (BM). Pertinent results from normal distribution.
d-dimensional Wiener process.

Sept 16- 20:

Quadratic variation and the non-differentiability of the BM’s sample path.
Martingale, super and sub-martingales, local martingales, Martingale
decomposition. Square integrable processes and bounded in mean square.

Sept 23-27:
Riemann integral versus Itô integral. Itô integral of an elementary integrand.
Properties of Itô integral: linearity and the martingale property. Itô isometry. Itô
integral of a general integrand.

A1, Due 27-Sept.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 3 / 190
 Introduction

Approximate Course Content, Timing and Exams:

Sept 30 - Oct 4:
Itô’s process. Itô’s differentiation rule. Derivation of Itô’s formula. Quadratic
variation of an Itô process. Quadratic variation of a geometric Brownian
motion. Itô’s formula in multidimensions. The product rule.

Oct 7-11:
Example of Itô process: Vasicek model, Geometric Brownian motion.
Existence and Uniqueness of solutions of Stochastic Differential Equations
(SDE). Change of measure (Girsanov’s theorem) and applications in finance.
Martingale Representation theorem.

Oct 14-18:
(Reading Week for undergrads).
Midterm, 17-Oct,

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 4 / 190
 Introduction

Approximate Course Content, Timing and Exams:

Oct 21- 25:

Feynman-Kac’s and discounted Feynman-Kac formulas. Numéraire processes.
Self-financing and admissible strategies. Investor’s wealth and gains
processes.

Oct 28- Nov 1:

Arbitrage opportunities. Equivalent Martingale measure (EMM) and Arbitrage.
First FTAP. Contingent claims and replicating strategies. Complete markets
and second FTAP. Risk-neutral valuation and Bayes’ theorem. Change of
Numéraire.

Nov 4-8:
The Black-Scholes model. Change of measure, EMM and completeness in the
BS-model. The Black-Scholes Formula. Proofs via PDE and Risk-neutral
valuation. The Greeks: Delta (-hedging), Gamma, Theta, Vega, Rho. Implied
volatility and volatility surface. Options on Dividend paying assets.

A2, Due 8-Nov.

Nov 11-15:
Volatility modeling: volatility as a deterministic function, generalized BS.
stochastic volatility, Heston 1993 model. The Feller condition. The
characteristic function. stochastic volatility, Stein and Stein 1991 model.
stochastic volatility, Hull and White 1987 model.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 5 / 190
 Introduction

Approximate Course Content, Timing and Exams:

Nov 18- 22:

Simulation of a Brownian motion. Euler discretization: Simulating a general Itô
process. Unbiased simulation of a GBM. MonteCarlo simulation and
properties. Pricing options. Confidence intervals for option prices.

Nov 25- 29:

Introduction to the pricing of default-free zero-coupon bonds. Term structure
modelling and example of classical models. The bond price in the Vasiček
model. Bond price solution via the PDE approach.

Dec 2- 6:
Introduction to the pricing of default-free zero-coupon bonds. Term structure
modelling and example of classical models. The bond price in the Vasiček
model. Bond price solution via the PDE approach.

Dec 9 -22
Final Exam, Wednesday Dec 11, 9am - noon.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 6 / 190
 Introduction

Overview

Chapter I: Results from statistics and stochastic analysis

Chapter II: Financial markets in continuous time
Chapter III: The Black-Scholes formula
Chapter IV: Stochastic volatility models
Chapter V: Numerical methods

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 7 / 190
 Results from statistics and stochastic analysis

Chapter I
Results from statistics and stochastic
analysis

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 8 / 190
 Results from statistics and stochastic analysis

Definition 1 (Measure space: the triplet (Ω, F, IP)).

The sample space Ω is a non-empty set.

A σ-algebra F on Ω is a system of subsets of Ω satisfying:
1. Ω ∈ F.
2. A ∈ F ⇒ Ac := Ω\A ∈ F.
3. For each sequence {An }n∈N , with An ∈ F for all n ∈ N, it holds that
[ 
An ∈ F.
n∈N

The tuple (Ω, F) is called measurable space.

A map IP : F → [0, ∞] on the measurable space (Ω, F) is called measure, if
IP(∅) = 0 and each sequence {An }n∈N of disjoint sets from F satisfies:
[  X
IP An = IP(An ).
n∈N n∈N

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science 9 / 190
 Results from statistics and stochastic analysis

Definition 2 (Finite measure).

IP is called finite, if IP(Ω) < ∞.

Definition 3 (σ-finite measure).

IP is called σ-finite, if there is an ascending sequence {An }n∈N of measurable sets,

S
i.e. An ⊂ An+1 , ∀ n ∈ N, with n∈N An = Ω such that IP(An ) < ∞ ∀ n ∈ N.

Definition 4 (Probability measure).

A measure IP on (Ω, F) satisfying IP(Ω) = 1 is called probability measure. The

triplet (Ω, F, IP) is called probability space.

Definition 5 (Equivalent measures).

Let IP and Q be measures on the measurable space (Ω, F).

Q is called absolutely continuous with respect to IP (notation: Q ≪ IP), if

IP(A) = 0 ⇒ Q(A) = 0, ∀A ∈ F.

IP and Q are equivalent measures (notation: IP ∼ Q), if

IP(A) = 0 ⇔ Q(A) = 0, ∀A ∈ F.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science10 / 190
 Results from statistics and stochastic analysis

Definition 6 (Random variable).

Let (Ω, F, IP) be a probability space.

A mapping X : Ω → R, with X −1 (B) := {ω ∈ Ω : X (ω) ∈ B} ∈ F for all
B ∈ B(R), is called a (real-valued) random variable. Here, B(R) denotes the
Borel σ-algebra, i.e. the smallest σ-algebra containing all open sets from R.

X (ω) = x is called realisation of X .

The probability measure IPX (B) = IP(X −1 (B)) = IP(X ∈ B), B ∈ B(R), is called
distribution of X .

A random variable X is called discrete, if X only takes values xi , i ∈ I, where I

is a (at most) countable set.

The Borel σ-algebra associated to the random variable X is defined as:

βX = {X −1 (B), B ∈ B(R)}

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science11 / 190
 Results from statistics and stochastic analysis

Definition 7 (Density).

Let X be a random variable on (Ω, F, IP). A function f : R → R is called density

(PDF) of X if
P
Discrete case: f (x) = IP(X = x) = px , with px ≥ 0 and i∈I pxi = 1, where I is
the (at most) countable range of X . R∞
Continuous case:
R f (x) ≥ 0 for all x ∈ R, −∞ f (x)dx = 1, and
IP(X ∈ A) = A f (x)dx, A ∈ B(R).

Definition 8 (Distribution function).

Let X be a random variable on (Ω, F, IP). The function F : R → [0, 1], with

F (x) = IP(X ≤ x) = IP({ω ∈ Ω : X (ω) ≤ x}),

is called (cumulative) distribution function (cdf) of X .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science12 / 190
 Results from statistics and stochastic analysis

Definition 9 (Expected value).

Let X be a random variable. Assuming existence, the value

 P
 i∈I xi pi , if X is discrete ,
IE [X ] :=
 ∞
R
−∞ xf (x)dx, if X is continuous ,

is called expected value of X .

One can similarly define IE [g(X )] for a function g and/or for a vector X .

Definition 10 (Expected value given event B).

Let X be a random variable. The value

IP({X =xi }∩B)
 P

 i∈I∩B xi IP(B)
, if X is discrete ,
IE [X |B] :=
 R IP′ ({X ≤x}∩B)

ℜ∩B x IP(B)
dx, if X is continuous ,

is called conditional expected value of X given B.

≤x}∩B)
where IP′ (X ≤ x ∩ B) = ∂ IP({X∂x

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science13 / 190
 Results from statistics and stochastic analysis

Definition 11 (Moment generating function).

Let X be a random variable and ϵ > 0 such that IE [etX ] < ∞ for all t with |t| < ϵ.
Then,
MX (t) := IE [etX ], |t| < ϵ,
is called the moment generating function of X .

Definition 12 (Characteristic function).

Let X be a random variable. The function

ϕX (t) := IE [eitX ], t ∈ R,

is called the characteristic function of X .

Remark 1.
From Inverse Fourier Transform:
Z
1
FX′ (x) = fX (x) = e−itx ϕX (t)dt
2π ℜ

Levy’s Theorem:
T
e−ita − e−itb
Z
1
FX (b) − FX (a) = lim ϕX (t)dt
2π T →∞ −T it

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science14 / 190
 Results from statistics and stochastic analysis

Definition 13 (Variance, skewness, kurtosis, covariance, correlation).

Assume X and Y to be a sufficiently integrable random variable such that the

expressions below are well defined. Let µ := IE [X ].
The variance of X is defined as: Var(X ) := σ 2 := IE [(X − µ)2 ].
The skewness of X is defined as:
(X − µ)3
 
skewness(X ) := IE .
σ3

The kurtosis of X is defined as:

(X − µ)4
 
kurtosis(X ) := IE .
σ4

The covariance of X and Y is defined as:

Cov(X , Y ) := IE [(X − IE [X ])(Y − IE [Y ])] .

The correlation of X and Y is defined as:

p
Cor(X , Y ) = Cov(X , Y )/ Var(X )Var(Y ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science15 / 190
 Results from statistics and stochastic analysis

Definition 14 (Normal distribution).

A random variable X is normally distributed with parameters µ ∈ R, σ > 0, i.e.

X ∼ N (µ, σ 2 ), if its density function is given by

(x − µ)2
 
1
f (x) = √ exp − 2
, x ∈ R.
0.15
0.10 2πσ 2σ
f(x)

0.05
0.00

−5 0 5 10

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science16 / 190
 Results from statistics and stochastic analysis

Lemma 1 (Moment generating and characteristic function).

The moment generating and the characteristic function of X ∼ N (µ, σ 2 ) are

σ2t 2 σ2t 2
   
MX (t) = exp µt + , ϕX (t) = exp iµt − , t ∈ R.
2 2

Remark 2 (Properties).

The mean of a normal distribution can be calculated as MX′ (t)|t=0 = µ.

The variance is given by MX′′ (t)|t=0 − (MX′ (t)|t=0 )2 = σ 2 .
The skewness of a normal distribution is 0, the kurtosis is 3.

Lemma 2 (Linear transformations).

1. Let X ∼ N (µ, σ 2 ), a, b ∈ R. Then aX + b ∼ N (aµ + b, (aσ)2 ).

2. Let X ∼ N (µX , σX2 ), Y ∼ N (µY , σY2 ) and (X , Y ) be jointly normal. Then
X ± Y ∼ N (µX ± µY , σX2 + σY2 + 2ρσX σY ), where ρ = Cor(X , Y ).
3. If X and Y are normal and uncorrelated, then:
(X , Y ) is jointly normal ⇔ X and Y are independent.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science17 / 190
 Results from statistics and stochastic analysis

Definition 15 (Multivariate Density).

Let X = (X1 , ..., Xd ) be a random vector on (Ω, F, IP). A function f : Rd → R is

called density of X if
P
Discrete case: f (x) = IP(X = x) = px , with px ≥ 0 and i∈I pxi = 1, where I is
the (at most) countable range of X . R∞ R∞
Continuous case:
R f (x) ≥ 0 for all x ∈ Rd , −∞ ... −∞ f (x)dx1 ...dxd = 1, and
IP(X ∈ A) = A f (x)dx, A ∈ B(R).

Definition 16 (Distribution function and Independence).

Let X be a random vector on (Ω, F, IP). The function F : Rd → [0, 1], with

F (x) = IP(X1 ≤ x1 , ..., Xd ≤ xd ) = IP({ω ∈ Ω : X (ω) ≤ x}),

is called (cumulative) distribution function (cdf) of X .

The random variables are independent iff
IP(X1 ≤ x1 , ..., Xd ≤ xd ) = IP(X1 ≤ x1 )...IP(Xd ≤ xd )

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science18 / 190
 Results from statistics and stochastic analysis

Definition 17 (Multivariate normal distribution).

A random vector X = (X1 , . . . , Xd ) is multivariate normal distributed with

parameters µ ∈ Rd , Σ ∈ Rd×d positive definite, i.e. X ∼ Nd (µ, Σ), if its density
function is given by
 
1 1 ′ −1
fX (x1 , . . . , xd ) = d 1
exp − (x − µ) Σ (x − µ) ,
(2π) 2 |Σ| 2 2

where Σ is the covariance matrix and |Σ| its determinant.

Lemma 3 (Moment generating and characteristic function).

The moment generating and the characteristic function of X are given by

   
′ 1 ′ 1
MX (t) = IE [et X ] = exp µ′ t + t ′ Σt , ϕX (t) = IE [eit X ] = exp iµ′ t − t ′ Σt ,
2 2

where t = (t1 , . . . , td ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science19 / 190
 Results from statistics and stochastic analysis

Two dimensional Normal Distribution

0.015

0.010

0.005
10
5
0.000
−10 0
−5 x2
0 −5
x1 5
10 −10

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science20 / 190
 Results from statistics and stochastic analysis

Definition 18 (Log-normal distribution).

A random variable X is log-normally distributed with parameters µ ∈ R, σ > 0, i.e.

X ∼ LN (µ, σ 2 ), if its density function is given by

(log x − µ)2
 
1
f (x) = √ exp − 2
1{x>0} .
σx 2π 2σ

σ = 0.25
σ = 0.5
1.5

σ=1
σ=2
1.0
f(x) (mu=0)

0.5
0.0

0 2 4 6 8

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science21 / 190
 Results from statistics and stochastic analysis

Lemma 4 (Working with the log-normal distribution).

For log-normal random variables, the moment-generating function exists for

t ≤ 0, but is not known in closed form.
Nevertheless, the moments can be calculated by

n2 σ 2

IE [X n ] = exp nµ + , n ∈ N.
2
Mean and variance of a log-normal distribution are thus given by
σ2 2 2
IE [X ] = eµ+ 2 , Var(X ) = (eσ − 1) · e2µ+σ .

Let Xm ∼ LN (µ, σm 2 ), for m = 1, 2, . . . , n, be independent random variables

with the samePmean µ. Then, Y = nm=1 Xm is log-normally distributed, i.e.

Q
n 2 ).
Y ∼ LN (nµ, m=1 σm
Let Y ∼ N (µ, σ 2 ). Then X := eY ∼ LN (µ, σ 2 ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science22 / 190
 Results from statistics and stochastic analysis

Definition 19 (Poisson distribution).

A random variable X is Poisson distributed with intensity parameter λ > 0, i.e.

X ∼ Poi(λ), if its density function is given by

λk e−λ
f (k ) = IP(X = k ) = , k ∈ {0, 1, 2, . . . }.
0.25
0.20
0.15 k!
f(x) (lambda=3)

0.10
0.05
0.00

1 2 3 4 5 6 7 8 9 10 11 12

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science23 / 190
 Results from statistics and stochastic analysis

Lemma 5 (Moment generating and characteristic function).

The moment generating and the characteristic function are given by
t it −1)
MX (t) = eλ(e −1) , ϕX (t) = eλ(e .

Remark 3 (Properties).
A Poisson distribution is often used to model the random number of discrete
occurrences during a given time-interval. The expected number of occurrences
is λ = IE [X ].
The mean and variance of the Poisson distribution is µ = σ 2 = λ, respectively.
1
Moreover, skewness = λ− 2 , kurtosis = 3 − λ−1 .
Take independent X ∼ Poi(λX ), Y ∼ Poi(λY ). Then X + Y ∼ Poi(λX + λY ).

Example 1.

The Poisson distribution appears, for example, in the following counting events:
The number of defaults, shocks to a stock, claims to an insurance company for
a given period.
The number of white blood cells found in a cubic centimetre of blood.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science24 / 190
 Results from statistics and stochastic analysis

Definition 20 (Exponential distribution).

A random variable X is exponentially distributed with parameter λ > 0, i.e.

X ∼ Exp(λ), if its density function is given by

0.4
f (x) = λ · e−λx 1{x≥0} , λ > 0.

λ = 0.25
λ = 0.5
λ = 0.75
λ=2
0.3
0.2
f(x)

0.1
0.0

0 2 4 6 8

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science25 / 190
 Results from statistics and stochastic analysis

Lemma 6 (Moment generating and characteristic function).

The moment generating and the characteristic function are given by
−1
it −1
  
t
MX (t) = 1− , ϕX (t) = 1− .
λ λ

Remark 4 (Properties).

The cdf is given by F (x) = (1 − e−λx )1{x≥0} .

The moments of the exponential distribution are µ = 1/λ, σ 2 = 1/λ2 ,
skewness = 2, and kurtosis = 9.

Example 2.

The exponential distribution is, for example, used to model:

The time until a radioactive particle decays. The time to the next earthquake.
The time until the next Poisson jump (shock in the market).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science26 / 190
 Results from statistics and stochastic analysis

Theorem 1 (Strong law of large numbers).

Let {X (i) }i∈N be a sequence of i.i.d. random variables with IE [|X (1) |] < ∞. Then, the
sample average converges almost surely (a.s., to be defined later) to the expected
value, i.e.
n
1 X (i) n→∞
X −→ IE [X (1) ] a.s.
n
i=1

Theorem 2 (Central limit theorem).

Let {X (i) }i∈N be a sequence of i.i.d. random variables with mean µ and variance
σ 2 < ∞. For Sn := X (1) + X (2) + . . . + X (n) , it holds that
Sn − nµ n→∞
Y (n) := √ −→ N (0, 1) in distribution.
σ n

Remark 5.
Both theorems hold irrespectively of the shape of the original distribution. Modes of
convergence are recalled later.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science27 / 190
 Results from statistics and stochastic analysis

Theorem 3 (Radon-Nikodým).

Let IP and Q be measures on the measurable space (Ω, F), where IP is σ-finite and
Q is finite. Then Q ≪ IP holds, if and only if there exists an integrable, IP-a.s.
non-negative function f (e.g. points where negative has zero probability), such that
Z
Q(A) = f (ω)d IP(ω), ∀A ∈ F.
A

The function f is called Radon-Nikodým derivative of Q with respect to IP.

Notation:
dQ
f = .
d IP

Example 3 (Radon-Nikodým).

Consider a coin with IP(’head’ ) = IP(’tail’ ) = 0.5.

Change the probabilities to Q(’head’ ) = 0.6 = 1 − Q(’tail’ ).
What is f ?

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science28 / 190
 Results from statistics and stochastic analysis

Definition 21 (Conditional expectation under σ-algebra).

Let X be an integrable random variable on the probability space (Ω, F, IP) and let
G ⊂ F be a sub-σ-algebra of F. The conditional expectation of X under G is
defined as the IP-a.s. unique G-measurable function IE [X |G] satisfying
Z Z
X (ω)d IP(ω) = IE [X |G](ω)d IP(ω), ∀A ∈ G.
A A

Remark 6 (Conditional expectation: a simple case).

Let B1 , . . . , Bn be a partition of Ω and G = σ{B1 , . . . , Bn }. For an integrable X on

(Ω, F, IP), it holds that
n
X
IE [X |G](ω) = 1Bi (ω)IE [X |Bi ],
i=1

where IE [X |Bi ] is the conditional expectation value of X given Bi .

Example 4.

Ω = {ω1 , . . . , ω6 }, X (ωi ) = i, IP(ωi ) = 1/6, G = σ({ω1 , ω2 , ω3 }, {ω4 , ω5 , ω6 }).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science29 / 190
 Results from statistics and stochastic analysis

Lemma 7 (Properties of the conditional expectation).

Let X be an integrable random variable on the probability space (Ω, F, IP) and let
G ⊂ F be a sub-σ-algebra of F. Then
1. IE [X |{∅, Ω}] = IE [X ].
2. IE [X |F] = X , IP-a.s.
3. If X is G-measurable, then IE [X |G] = X , IP-a.s.
4. Taking out what is known:
If Z is an integrable random variable and X is G-measurable, then

IE [ZX |G] = XIE [Z |G], IP-a.s.

5. Linearity: Let Z be an integrable random variable and a, b ∈ R, then

IE [aX + bZ |G] = aIE [X |G] + bIE [Z |G], IP-a.s.

6. Monotonicity: Let Z be a random variable satisfying X ≤ Z , then

IE [X |G] ≤ IE [Z |G], IP-a.s.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science30 / 190
 Results from statistics and stochastic analysis

Lemma 8 (Properties of the conditional expectation (cont.)).

Let X be an integrable random variable on the probability space (Ω, F, IP) and let
G ⊂ F be a sub-σ-algebra of F. Then
1. Tower property: For each sub-σ-algebra H ⊂ G ⊂ F we get

IE [IE [X |G]|H] = IE [X |H], IP-a.s.

2. Especially, if H = {∅, Ω}, this implies

IE [IE [X |G]|H] = IE [X ].

3. If X is independent of G (e.g. βX ∩ G = ∅), then

IE [X |G] = IE [X ], IP-a.s.

4. Conditional version of Jensen’s inequality:

Let φ(X ) be integrable. For a real valued convex function φ, it holds that

φ(IE [X |G]) ≤ IE [φ(X )|G], IP-a.s.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science31 / 190
 Results from statistics and stochastic analysis

Remark 7.
Regarding the interpretation of the conditional expectation, the particular σ-algebra
represents the level of information based on which the expectation is computed.
The σ-algebra F itself stands for complete information.
The sub-σ-algebra G ⊂ F stands for partial information.
The trivial σ-algebra G = {∅, Ω} contains no information.

Definition 22 (Filtration).

A filtration F on (Ω, F, IP) is defined as an ascending family of σ-algebras

F = {Ft }t≥0 ,

satisfying
Fs ⊂ Ft ⊂ F, ∀ 0 ≤ s ≤ t < ∞.
The quadruple (Ω, F, F, IP) is called filtered probability space.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science32 / 190
 Results from statistics and stochastic analysis

Definition 23 (The usual conditions).

The usual conditions (u.c.) are said to hold, if

1. F is complete, i.e. F0 contains all subsets of IP-null sets of F,
2. F is right continuous, i.e.
\
Ft = Ft+ := Fs , ∀t ≥ 0.
s>t

Remark 8.
Ft represents the level of information up to time t, whereas F = {Ft }t≥0 represents
the flow of information with respect to time.

If the usual conditions are not satisfied, the filtration can be completed:
In the case of a not completed filtration, this is called IP-completion.
In the case of a not right continuous filtration, it is called IP-augmentation.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science33 / 190
 Results from statistics and stochastic analysis

Definition 24 (Stochastic process).

A stochastic process (of dimension d ∈ N) is a family

X = {Xt }t≥0

of random variables (d-dimensional vectors), defined on a filtered probability space

(Ω, F, F, IP). The stochastic process X is said to be . . .
. . . adapted to the filtration F, if Xt is measurable w.r.t. Ft for all t ≥ 0.
. . . measurable, if the mapping X : [0, ∞) × Ω → Rd is
B([0, ∞)) × F-B(Rd )-measurable, where B(M) is the Borel-σ-algebra on the
set M.
. . . progressively measurable, if X : [0, t) × Ω → Rd is
B([0, t)) × Ft -B(Rd )-measurable for all t ≥ 0.
. . . predictable, if X is measurable with respect to the predictable σ-algebra P.
P is generated in [0, T ] × Ω by all adapted and left-continuous processes.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science34 / 190
 Results from statistics and stochastic analysis

Remark 9 (Economic interpretation).

We explicitly try to model the dynamic behavior of random (stock) price

movements and information. It is natural to assume that some events which
are random in the future become known as time goes on.
When we later make predictions about future events (e.g. when we compute
expectations) we keep the probability measure fixed, but we condition on
different filtrations, i.e. the information given at the present time t (which is Ft ).

Adapted processes reveal their time t value with Ft . This is a natural

assumption for stock processes - one observes the time t value precisely at
time t (and not prior to t).
Left-continuous processes are anticipating in the sense that lims↗t Xs = Xt .
This means that the time t value is announced. This situation is realistic for
investment strategies, where the investor has to make an investment decision
for her or his time t portfolio composition based on knowing the (stock) market
up to (but not including) time t.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science35 / 190
 Results from statistics and stochastic analysis

Definition 25 (Natural filtration).

Let (Ω, F, F, IP) be a filtered probability space and let X be a stochastic process,
adapted to F. The natural filtration FX is defined as the family of σ-algebras

Ft := σ(Xs : 0 ≤ s ≤ t), ∀ t ≥ 0.

It is the smallest σ-algebra (i.e. the intersection of all σ-algebras satisfying this
property) such that X is an adapted process.

Remark 10.
The intersection of (any number of) σ-algebras is again a σ-algebra.
The natural filtration is a generalization of the Borel σ − algebra generated by a r.v.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science36 / 190
 Results from statistics and stochastic analysis

Definition 26 (Stopping time).

Let (Ω, F, F, IP) be a filtered probability space. A stopping time (w.r.t. the filtration
F) is a B([0, ∞])-measurable random variable

τ : Ω → [0, ∞]

satisfying
{τ ≤ t} = {ω ∈ Ω : τ (ω) ≤ t} ∈ Ft , ∀ t ≥ 0.

Lemma 9.
Let τ1 and τ2 be stopping times. Then

min{τ1 , τ2 } =: τ1 ∧ τ2 , max{τ1 , τ2 } =: τ1 ∨ τ2

are stopping times, too.

In particular, τ1 ∧ t is a stopping time for all t ≥ 0.
An example of a positive random variable that is not a stopping time: Last exit
time.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science37 / 190
 Results from statistics and stochastic analysis

Definition 27 (Modes of convergence).

All random variables considered below are suitably integrable and defined on
(Ω, F, IP). A sequence {X (n) }n∈N of rvs converges to the rv X . . .
. . . almost surely (a.s.), if
 
IP lim X (n) = X = 1.
n→∞

. . . in probability, if for all ϵ > 0:

 
lim IP |X (n) − X | > ϵ = 0.
n→∞

. . . in distribution, if for all bounded and continuous functions f : R → R:

IE [f (X (n) )] → IE [f (X )], (n → ∞).

. . . in mean square (L2 (IP)), if

IE (X (n) − X )2 → 0,
 
(n → ∞).

(this last one is key to develop Ito’s calculus).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science38 / 190
 Results from statistics and stochastic analysis

Lemma 10 (Modes of convergence - relations).

Working on the probability space (Ω, F, IP), the following relations hold:
almost surely (a.s.) ⇒ in probability
in mean square ⇒ in probability
in probability ⇒ in distribution
Furthermore, given additional conditions, the following inversions hold:
in probability ⇒ almost surely, if for all ϵ > 0:
 
lim IP sup |X (m) − X | > ϵ = 0.
n→∞ m≥n

in probability ⇒ in mean square, if |Xn | ≤ Y ∈ L2 (IP) for all n ≥ 1 (a.s.)

in distribution ⇒ in probability, if the limit X is a constant (a.s.)

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science39 / 190
 Results from statistics and stochastic analysis

Definition 28 (The p’th-variation of a continuous process X ).

For a continuous process X on (Ω, F, F, IP), define the p’th variation as

⟨X , X ⟩pt := lim
X
|Xtk +1 − Xtk |p ,
∆tk →0
tk ≤t

where the limit is taken in probability, p > 0 and 0 = t1 < · · · < tn = t.

If p = 2 (the quadratic variation) then the superscript p is suppressed.

If p = 1 then it is called Total Variation and if finite then Bounded (total) variation
or simply finite variation.

Processes (and hence deterministic functions) with differentiable paths are of

finite variation.
Processes of finite variation have a quadratic variation of zero.
Note
RT
̸ ⟨X , X ⟩1T = lim∆tk →0 tk ≤t |Xtk +1 − Xtk |.
P P
0 Xt dt = lim∆tk →0 tk ≤t Xtk |tk +1 − tk | =
If
R TX is′′ differentiable then ′
P P
0 Xt dt = lim∆tk →0 tk ≤t Xtk |tk +1 − tk | = lim∆tk →0 tk ≤t |Xtk +1 − Xtk | only if X is
increasing.
If f (t) has countable points of non-differentiability then ⟨f ⟩1T < ∞ and ⟨f ⟩2T = 0.
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science40 / 190
 Results from statistics and stochastic analysis

Definition 29 (The quadratic covariation).

Write ⟨M, M⟩ for ⟨M⟩ and extend ⟨◦⟩ to the bilinear form (called quadratic
covariation) ⟨◦ , ◦⟩. For this, use the polarization identity

1 1
⟨M, N⟩ := (⟨M + N, M + N⟩ − ⟨M − N, M − N⟩) := (⟨M + N⟩ − ⟨M − N⟩) .
4 4

The polarization identity reflects the Hilbert-space structure of the inner

product ⟨◦ , ◦⟩.
If N is of finite variation, M ± N has the same quadratic
p variation as M, so
⟨M, N⟩ = 0 (Hint: use Cauchy-Schwartz: ⟨M, N⟩ ≤ ⟨M⟩⟨N⟩)

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science41 / 190
 Results from statistics and stochastic analysis

Example 5 (The quadratic covariation).

Let X be continuous with existing quadratic variation ⟨X ⟩t . For a continuous

function t 7→ f (t) of finite variation, it holds that:

⟨X , f ⟩t = 0.

We will study Brownian motion processes next. For two independent Brownian
motions W (1) and W (2) . Then:
D E
W (1) , W (2) = 0.
t

Sketch of the proofs.

For the first claim, observe ⟨X + f ⟩t = ⟨X ⟩t . √
To show the second claim, note that (W (1) + W (2) )/ 2 is again (in distribution)
a Brownian motion (we will come back to this later).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science42 / 190
 Results from statistics and stochastic analysis

Definition 30 (Brownian motion or Wiener process).

Given (Ω, F, F, IP), an adapted process W = {Wt }t≥0 is called Brownian motion, if
W satisfies the following properties.
1. W0 = 0, IP-a.s.
2. W has independent increments, i.e. the random variables

(Wt − Ws ), (Wv − Wu )

are independent for all 0 ≤ u < v ≤ s < t < ∞.

3. W has normally distributed increments, i.e.

Wt+h − Wt ∼ N (0, h), ∀ h > 0.

4. W has IP-a.s. continuous paths.

Theorem 4 (Wiener).

Brownian motion exists.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science43 / 190
 Results from statistics and stochastic analysis

Remark 11 (The natural filtration of the Brownian motion).

If F = FW , then the independence of the increments is equivalent to stating that

(Wt − Ws ) is independent of Fs for all 0 ≤ s < t < ∞.
The IP-completion of FW is right (and left) continuous.

Definition 31 (Brownian motion - Wiener process - of dimension d ∈ N).

Let (Ω, F, F, IP) be a filtered probability space. W satisfying

(1) (d)
W ′ = (W (1) , . . . , W (d) ) = {(Wt , . . . , Wt )}t≥0

is called a d-dimensional Wiener process, if W (j) , j = 1, . . . , d, are independent

Wiener processes.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science44 / 190
 Results from statistics and stochastic analysis

Sample paths of a Brownian motion

4
Brownian Motion

2
0
−2
−4

0 2 4 6 8 10

time t

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science45 / 190
 Results from statistics and stochastic analysis

Lemma 11 (Properties of the Brownian motion).

Wt is normally distributed with expected value 0 and variance t.

The covariance of Wt and Ws is Cov(Wt , Ws ) = min{t, s}.
With probability one, t 7→ Wt is nowhere differentiable.
The paths of Brownian motion are of unbounded variation on every interval.
The paths of Brownian motion are unbounded, i.e.


IP limsupt≥0 Wt = ∞ = IP (liminft≥0 Wt = −∞) = 1.

Each path of the Brownian motion has infinitely many roots near the origin and
near infinity.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science46 / 190
 Results from statistics and stochastic analysis

Theorem 5 (Lévy).

The quadratic variation of a Brownian path over [0, t] exists and equals t, i.e.

⟨W ⟩t = t.

This limit holds in mean square and, hence, in probability.

We only sketch the proof (limit in mean square).

This limit even holds IP-a.s. (difficult to prove).
If we consider the theorem over [0, t + dt], [0, t] and subtract, we can write the
result formally as
(dWt )2 = dt.
This symbolism is later seen to be the essence of Itô calculus.
In Lévy’s theorem, the converse also holds:

If M is any continuous, square-integrable martingale on (Ω, F, F, IP), with

M0 = 0 and quadratic variation t, then M is a Brownian motion.

Remark 12.
The total variation is unbounded,

⟨W ⟩1t = ∞.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science47 / 190
 Results from statistics and stochastic analysis

Lemma 12 (Properties of the Brownian motion (cont.)).

Several processes on (Ω, F, F, IP), where F = FW , deduced from the Brownian

motion W itself, are again (in distribution) Brownian motions.
Reflected Brownian motion:

Xt := −Wt , ∀t ≥ 0.

Self similarity:
1
W 2, Xtc := c > 0, ∀t ≥ 0.
c ct
Time-inverted Brownian motion:

Xt := tW1/t , ∀t > 0, X0 := 0.

Restarted Brownian motion:

Xt := Wτ +t − Wτ , ∀t ≥ 0,

for some stopping time τ .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science48 / 190
 Results from statistics and stochastic analysis

Definition 32 (Martingale).

Let (Ω, F, F, IP) be a filtered probability space. A stochastic process X is called

martingale w.r.t. (IP, F), if
1. X is F-adapted.
2. Each Xt is integrable, i.e. IE [|Xt |] < ∞ for all t ≥ 0.
3. The best prediction about the future is the present value, i.e.

IE [Xt |Fs ] = Xs , IP-a.s. ∀ 0 ≤ s ≤ t < ∞.

(⇔ IE [ XXst |Fs ] = 1, or IE [Xt − Xs |Fs ] = 0)

Definition 33 (Supermartingale, submartingale).

If the last equation is replaced by an inequality, we call the process X . . .

. . . supermartingale, if IE [Xt |Fs ] ≤ Xs , IP-a.s. ∀ 0 ≤ s ≤ t < ∞.
. . . submartingale, IE [Xt |Fs ] ≥ Xs , IP-a.s. ∀ 0 ≤ s ≤ t < ∞.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science49 / 190
 Results from statistics and stochastic analysis

Remark 13.

Martingales represent fair games, i.e. situations in which there is no drift.

In a statistical situation, where we have

data = signal + noise,

martingales represent the noise component.

There is an extensive theory (on semimartingales, i.e. a local martingale plus a
finite variation process) build on such decompositions, see for instance Protter
(1992).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science50 / 190
 Results from statistics and stochastic analysis

Example 6 (Some martingales).

Show that the following processes are martingales. Let W be a Wiener process on
(Ω, F, F, IP) and F = FW .
Xt := Wt and F = FW .
Xt := Wt2 − t and F = FW .
Xt := exp(σWt − 1/2σ 2 t), for σ > 0 and F = FW .
Collecting information about a random variable: Let Y be an integrable
random variable on (Ω, F, F, IP). Then the stochastic process

Xt := IE [Y |Ft ], ∀ t ≥ 0,

is a martingale.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science51 / 190
 Results from statistics and stochastic analysis

Definition 34 (Local martingale).

Let (Ω, F, F, IP) be a filtered probability space. An adapted, càdlàg

(right-continuous with left limits) process X is called a local martingale w.r.t.
(IP, F), if there exists a sequence of increasing stopping times, {τn }n∈N , such that:
- limn→∞ τn = ∞ IP-a.s. and
- {Xt∧τn 1{τn >0} }t≥0 is a martingale for all n ∈ N.

{τn }n∈N is said to reduce X and is called a localizing (or fundamental) sequence
for X .

Remark 14.
Example, processes that explode in finite time, e.g. IE [|Xt |] = ∞ for some t < ∞ are
not martingales but rather local martingales. (in [0.t) is a martingale but not in
[0, ∞). )

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science52 / 190
 Results from statistics and stochastic analysis

Theorem 6 (Conditions for a local martingale to be a martingale).

A local martingale X is a martingale if and only if for all T > 0 the set

{Xτ | τ stopping time with τ ≤ T }

is uniformly integrable, i.e. limc→∞ supτ ≤T IE [|Xτ |1|Xτ |≥c ] = 0.

Let X be a local martingale such that

IE [sup |Xs |] < ∞, ∀ t ≥ 0.

s≤t

Then X is a martingale.
If IE [sups≥0 |Xs |] < ∞, then X is a uniformly integrable martingale. In particular,
a bounded local martingale is a uniformly integrable martingale.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science53 / 190
 Results from statistics and stochastic analysis

Definition 35 (Square integrable processes).

Let (Ω, F, F, IP) be a filtered probability space and let X be an F-adapted stochastic
process. X is a L2 [0, T ]-process, if X is progressively measurable and
"Z #
T
||X ||2 := IE Xt2 dt < ∞.
0

The set of all L2 [0, T ]-processes defines a linear vector space, denoted by V T .

Think of Xt as a function of Wt . Note ||X ||2 is not the QV of XT .

Definition 36 (Processes bounded in mean square).

If X satisfies the stronger condition

sup IE [Xt2 ] < ∞,

t≥0

then X is said to be bounded in L2 (i.e. X is square integrable, X ∈ L2 [0, T ])

Recall a continuous (IP- a.s.) bounded function on a compact interval is Riemann

integrable.
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science54 / 190
 Results from statistics and stochastic analysis

Definition 37 (The spaces M2 , cM2 , and cM20 ).

For a martingale M on (Ω, F, F, IP), write M ∈ M2 if M is bounded in L2 , i.e.

sup IE [Mt2 ] < ∞.

t≥0

Further, write M ∈ M20 if M0 = 0. Finally, write cM2 , cM20 for the subclasses of
(pathwise) continuous M.

Definition 38 (The quadratic variation of X ∈ cM2 ).

For a process X ∈ cM2 , we have the decomposition (Doob-Mayer)

Xt2 = X02 + Mt + At ,

with M a continuous martingale and A a continuous and increasing process. We write

⟨X ⟩t := At

where ⟨X ⟩ is the quadratic variation of X .

Rt 2
Note If Xt = a(t)Wt + b(t) with a,b of finite variation then ⟨X ⟩t = ⟨aW ⟩t = 0 a (s)ds
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science55 / 190
 Results from statistics and stochastic analysis

Remark 15 (Motivation: the stochastic integral).

Stochastic integration was introduced by Kiyoshi Itô (born 1915) in 1944,

hence the name Itô calculus.
It gives a meaning to the expression
Z t Z t
XdY := Xs (ω)dYs (ω),
0 0

for suitable stochastic processes X (the integrand) and Y (the integrator).

We only consider the basic case with Brownian motion as integrator.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science56 / 190
 Results from statistics and stochastic analysis

Remark 16 (Motivation: the stochastic integral (cont.)).

Riemann-Stieltjes (RS) integrals

Z t
f (s)dg(s)
0

have as integrator g the difference of two monotone increasing functions (f is any

function). Such R t functions are (locally) of bounded variation.
Note the r.v. 0 Xs dg(s) is well defined as a RS integral pathwise.
Since the paths of Brownian motion are of infinite (unbounded) variation on every
interval, it is not possible to interpret a stochastic integral pathwise (i.e. ω-by-ω) as a
Riemann-Stieltjes integral. Consequently, Riemann-Stieltjes and Itô integrals must
fundamentally be different.
To define Itô integrals such as
Z t Z t
XdY = Xs (ω)dWs (ω),
0 0

we start with the simplest possible integrand X and extend successively to more
complicated stochastic processes.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science57 / 190
 Results from statistics and stochastic analysis

Definition 39 (The stochastic integral for indicator functions).

If Xt (ω) := 1(a,b] (t), where 0 ≤ a < b < ∞, there is only one plausible way to define
Rt
0 Xs dWs . Define

Z t Z t  0, if t ≤ a,
Xs (ω)dWs (ω) = 1(a,b] (s)dWs (ω) := Wt − Wa , if a < t ≤ b,
0 0 
Wb − Wa , if t ≥ b.
Rt
= a dWs (ω) for a < t ≤ b.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science58 / 190
 Results from statistics and stochastic analysis

Definition 40 (The stochastic integral for sums of indicator functions).

If X is a linear combination of indicators, i.e.

n
X
Xt (ω) := ci 1(ai ,bi ] (t),
i=1

we extend the first definition by linearity. We obtain

Z t n
X Z t
Xs dWs := ci 1(ai ,bi ] (s)dWs .
0 i=1 0

The following questions need to be answered:

How to extend this from constants ci to suitable random variables?
How to simplify the (long and inconvenient) three-line expressions per indicator
above?

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science59 / 190
 Results from statistics and stochastic analysis

Definition 41 (The stochastic integral for simple processes).

We call a stochastic process X simple, if there is a partition of [0, T ]

0 = t0 < t1 < . . . < tn = T < ∞

and uniformly bounded Ftk -measurable random variables (i.e. ∃ C ∈ R : |ξk | ≤ C for
all k = 0, . . . , n and ω) and if Xt (ω) can be written in the form
n−1
X
Xt (ω) := ξi (ω)1(ti ,ti+1 ] (t), 0 ≤ t ≤ T.
i=0

Then, if tk ≤ t < tk +1 ,
Z t −1
kX
It (X ) := Xs dWs = ξi (Wti+1 − Wti ) + ξk (Wt − Wtk )
0 i=0
n−1
X
= ξi (Wt∧ti+1 − Wt∧ti ).
i=0

Note that by definition: I0 (X ) = 0, IP-a.s.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science60 / 190
 Results from statistics and stochastic analysis
Pn−1
For example, define Xt (ω) = i=0 Wti (ω)1(ti ,ti+1 ] (t)

This is Xt (ω) = Wt0 (ω)1(t0 ,t1 ] (t) + ... + Wtn −1 (ω)1(tn−1 ,tn ] (t) then
Rt
It (X ) = 0 Xs dWs = n−1
P
i=0 Wti (Wti+1 ∧τ − Wti ∧τ )

.
Note Stratanovich proposed:
Rt Wti +Wti+1
It (X ) = 0 Xs dWs = n−1
P
i=0 2 (Wti+1 ∧τ − Wti ∧τ )

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science61 / 190
 Results from statistics and stochastic analysis

Remark 17.
What could be an economic interpretation of this result?

In Finance, one does not know the future value of the stock but rather present
value, i.e. take n = T = 2, ti = i, i = 0, 1, 2, then:

Xt (ω) = ξ0 (ω)1(0,1] + ξ1 (ω)1(1,2]

At time t = 1 and for a specific ω ∈ Ω one knows the stock price is X1 = ξ0 (ω) but

one does not know what the price will be at t = 1 + ϵ for any ϵ > 0.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science62 / 190
 Results from statistics and stochastic analysis

We collect some properties of the stochastic integral defined so far.

Lemma 13 (Properties of the stochastic integral for simple processes).

Linearity, i.e. for simple processes X , Y and a, b ∈ R

It (aX + bY ) = aIt (X ) + bIt (Y ).

The martingale relation holds, i.e.

IE [It (X )|Fs ] = Is (X ), IP-a.s. ∀ 0 ≤ s < t < ∞.

Hence, It (X ) is a (continuous) martingale.

The stochastic integral for simple integrands is essentially a martingale (called a

martingale transform), and the above is essentially the proof that martingale
transforms are martingales.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science63 / 190
 Results from statistics and stochastic analysis

Further properties of the stochastic integral for simple functions.

Theorem 7 (Itô isometry for simple stochastic processes).

Let X be a simple stochastic process.

We have the Itô isometry
h i Z t 
IE (It (X ))2 = IE Xs2 ds .
0

Moreover, for s < t, it holds that

h i Z t 
IE (It (X ) − Is (X ))2 |Fs = IE Xu2 du , IP-a.s.
s

Remark 18.
Rt
Note the quadratic variation of the process It (X ) is 0 Xs2 ds. Hence The previous
result provides the expected QV of It (X ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science64 / 190
 Results from statistics and stochastic analysis

Remark 19.
Rt
The Itô isometry suggests that 0 Xs dWs should be defined only for processes
X from L2 [0, ∞), i.e. square integrable
Z t 
IE Xs2 ds < ∞, ∀ t ≥ 0.
0

In other words, if Xt ∈ L2 then It is in L2 and a martingale (bounded).

This means we can transfer convergence on a suitable L2 -space of stochastic
processes to a suitable L2 -space of martingales. This provides an L2 -theory of
stochastic integration, for which Hilbert-space (vector
hR space
i with inner product)
2 t 2
methods are available (i.e. ∥It ∥ = E[⟨I⟩t ] = E 0 Xs ds )
In all financial applications, we have in mind that there is a fixed time-horizon T
(e.g. an option is written at time t = 0 and expires at t = T ). Then the above is
"Z # Z
T T h i
2
IE Xs ds = IE Xs2 ds < ∞.
0 0

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science65 / 190
 Results from statistics and stochastic analysis

Remark 20.
So far, we know how to integrate simple processes. We now seek a class of
integrands, which can be suitably approximated by simple integrands. It turns out
that:
1. The suitable class of integrands is the class of (B([0, ∞)) ⊗ F)-measurable,
Rt  
F-adapted processes X with 0 IE Xs2 ds < ∞ for all t > 0.
2. Each such X may be approximated byRa sequence of simple integrands X (n) ,
t
so that the stochastic integral It (X ) = 0 Xs dWs may be defined as the limit of
(n)
R t (n)
It (X ) = 0 Xs dWs as n tends to infinity.
3. The
R t properties from both lemmas above remain true for the stochastic integral
0 Xs dWs defined by 1. and 2.
4. Details of this construction are given in Øksendal (2000).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science66 / 190
 Results from statistics and stochastic analysis

Finally, we are in the situation to define the Itô integral.

Definition 42 (The Itô integral).

Let X be a stochastic process with properties as described in the remark above.

{X (n) }n∈N is a sequence of simple processes converging to X , i.e. the sequence
{X (n) }n∈N satisfies
Z t 
(n) 2
IE (Xs − Xs ) ds → 0, (n → ∞).
0

Then, the Itô integral of X is defined as

Z t Z t
(n)
Xs dWs := lim Xs dWs ,
0 n→∞ 0

thelimit being taken in L2 .

R 2 
t 2 R t (n) 2
(E 0 Xs dWs − 0 (Xs ) dWs → 0 in n)

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science67 / 190
 Results from statistics and stochastic analysis

Example 7.

Compute
Z t
1 2 1
Ws dWs = W − t.
0 2 t 2

Rt
Note the contrast with ordinary (Newton-Leibniz) calculus! ( 0 xdx = 21 t 2 , more
Rt
generally 0 g(x)dg(x) = 12 g 2 (t))
Itô calculus requires the second term on the right - the Itô correction term -
which arises from the quadratic variation of W .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science68 / 190
 Results from statistics and stochastic analysis

Definition 43 (Itô process).

An adapted stochastic process X on (Ω, F, F, IP), with

Z t Z t
Xt := x0 + b(s, Xs )ds + σ(s, Xs )dWs ,
0 0

is called Itô process. The functions b (drift) and σ (dispersion or volatility) map
from R+ × R to R. To simplify notation, this integral equation is often expressed
symbolically in differential form. In terms of the stochastic differential equation
(SDE)
dXt = bdt + σdWt , X0 = x0 ,
where the arguments of b and σ are surpressed for notational convenience.

Remark 21 (The dynamics of f (Xt )).

Theorem 8 (Itô’s formula) is the stochastic analogue to the chain rule of

ordinary calculus.
Given dXt and f ∈ C 2 , it gives meaning to the stochastic differential df (Xt ) of
the process f (Xt ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science69 / 190
 Results from statistics and stochastic analysis

Remark 22 (Martingale characterization).

Let X be an Itô process on (Ω, F, F, IP) with

dXt = b(t, Xt )dt + σ(t, Xt )dWt , X 0 = x0 .

If X is a martingale, then b ≡ 0 almost surely.

From the definition of the Itô integral, it follows that for suitably adapted and
RT
measurable σ, that fulfills the integrability condition IE [ 0 σ(t, Xt )2 dt] < ∞ for all
T > 0, we have that the solution to

dXt = σ(t, Xt )dWt , X0 = x0 ,

is an FW -martingale.
If dXt = Xt σt dWRt
for a progressively measurable process σ and if Novikov’s
1 T 2
condition IE [e 2 0 σt dt ] < ∞ for all T > 0, then X is a martingale which is given
by Rt 1 t 2
R
Xt = X0 e 0 σs dWs − 2 0 σs ds , t ≥ 0.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science70 / 190
 Results from statistics and stochastic analysis

Remark 23 (Rules for the quadratic variation).

This table can be used as a shorthand for the corresponding properties of the
quadratic (co-)variation.

· dt dWt
dt 0 0
dWt 0 dt

Example 8 (The quadratic variation of an Itô process).

We find (using the above table)

(dXt )2 = d ⟨X ⟩t = (bdt + σdWt )2

= b2 (dt)2 + 2bσdtdWt + σ 2 (dWt )2
= 0 + 0 + σ 2 dt.

This equation is required in Itô’s formula whenever d ⟨X ⟩t has to be simplified.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science71 / 190
 Results from statistics and stochastic analysis

Theorem 8 (Itô’s formula - simple version).

If X is an Itô process and f ∈ C 2 , then f (X ) has stochastic differential

1
df (Xt ) = f ′ (Xt )dXt + f ′′ (Xt )d ⟨X ⟩t , f (X0 ) = f (x0 ).
2
Writing out the integrals, we get
Z t Z t
1
f (Xt ) = f (x0 ) + f ′ (Xs )dXs + f ′′ (Xs )d ⟨X ⟩s .
0 2 0

Theorem 9 (Itô’s formula - general version).

If X is an Itô process and f ∈ C 1,2 , then f = f (t, Xt ) has stochastic differential

 
1 1
df = ft dt + fx dXt + fxx d ⟨X ⟩t = ft + bfx + σ 2 fxx dt + σfx dWt .
2 2

That is in integral form (writing f0 for f (0, x0 ), the initial value of f )

Z t Z t
1
f = f0 + (ft + bfx + σ 2 fxx )ds + σfx dWs .
0 2 0

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science72 / 190
 Results from statistics and stochastic analysis

Theorem 10 (Multidimensional Itô formula).

Let W = (W (1) , . . . , W (m) ) be an m-dimensional Brownian motion and

X = (X (1) , . . . , X (n) ) be an n-dimensional Itô process with
m
(i) (j)
X
dXt = bi (t, Xt )dt + σij (t, Xt )dWt , i = 1, . . . , n.
j=1

The function f : [0, ∞) × Rn → R is twice continuously differentiable in each

argument. Then, for all t ≥ 0
n n n
X (i) 1 XX D E
df (t, Xt ) = ft dt + fxi dXt + fxi xj d X (i) , X (j) ,
2 t
i=1 i=1 j=1

where
D E m
X
d X (i) , X (j) = σik (t, Xt )σjk (t, Xt )dt.
t
k =1

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science73 / 190
 Results from statistics and stochastic analysis

Theorem 11 (The product rule).

Let X (1) and X (2) be Itô processes with

(i) (i) (i) (i)
dXt = bi (t, Xt )dt + σi (t, Xt )dWt , i = 1, 2.

Then D E
(1) (2) (2) (1)
d(X (1) X (2) )t = Xt dXt + Xt dXt + d X (1) , X (2) .
t

Again, note the difference to ordinary calculus, where the last term is missing.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science74 / 190
 Results from statistics and stochastic analysis

Example 9 (The Vasicek model).

Consider the following SDE, which is often used to model short-rates, for the
process r = {rt }t≥0 on (Ω, F, F, IP)

drt = k (θ − rt )dt + σdWt , r0 > 0,

where all coefficients are positive constants.

This SDE can be solved as follows:

1. Determine the SDE for ekt rt via the product rule.
2. Calculate rt by integrating the SDE from 1.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science75 / 190
 Results from statistics and stochastic analysis

Consider the SDE:

dXt = b(t, Xt )dt + σ(t, Xt )dWt , X 0 = x0 .

Theorem 12 (Existence and uniqueness of solutions of a SDE).

Assume that b and σ are continuous functions, satisfying that for arbitrary positive
constants T and N and for all x, y ∈ R, | x |, | y |≤ N and 0 ≤ t ≤ T ,
The Lipschitz condition:

|b(t, x) − b(t, y )| + |σ(t, x) − σ(t, y )| ≤ K |x − y |.

The growth condition:

b2 (t, x) + σ 2 (t, x) ≤ K 2 (1 + x 2 ).

For some constant K > 0 (depending possibly on T and N). Then there exists a
unique solution X , which is adapted to the filtration of the Brownian motion W .

The first condition holds if b and σ have continuous first partial derivatives w.r.t x,
and the second condition holds when they both have at most linear growth in x for
large x and bounded for arbitrarily small x.
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science76 / 190
 Results from statistics and stochastic analysis

Remark 24 (Motivation: Geometric Brownian motion (GBM)).

Our motivation / aim is to model the time evolution of a stock price St .

Consider how the stock S changes in some small time-interval from time t (the
present) to time t + dt (the near future). Writing dSt for the change St+dt − St
in S, the discrete return of S in this interval is dSt /St .
It is economically reasonable to expect this return to decompose into two
components, a systematic part and a random part.
The systematic part could plausibly be modelled by µdt, where µ is some
parameter representing the mean rate of return of the stock.
The random part could plausibly be modelled by σdWt , where dWt represents
the noise term driving the stock price dynamics.
σ is a second parameter, describing how much the stock price fluctuates. This
parameter is called the volatility of the stock.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science77 / 190
 Results from statistics and stochastic analysis

Putting this together, we obtain the following stochastic differential equation:

Definition 44 (Geometric Brownian motion).

The solution of the SDE

dSt = St (µdt + σdWt ), S0 > 0,

is called geometric Brownian motion with drift µ and volatility σ > 0.

This corrects Bachelier’s model of 1900 (a model without the factor St on the
right - missing the interpretation in terms of returns) which had a positive
probability for negative stock prices.
Modeling stock prices via a geometric Brownian motion was suggested by Paul
A. Samuelson (1965). In part for this, Samuelson received the Nobel Prize in
Economics in 1970.

Lemma 14 (Geometric Brownian motion).

The SDE for Geometric Brownian motion has the unique solution
  
St = S0 exp µ − 0.5σ 2 t + σWt .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science78 / 190
 Results from statistics and stochastic analysis

Example 10 (Geometric Brownian motion).

The SDE for Geometric Brownian motion has the unique solution
n  o
St = S(0) exp µ − 0.5σ 2 t + σWt .

Proof: We let n  o
f (t, x) := S(0) exp µ − 0.5σ 2 t + σx .

The partial derivatives of f are

 
ft = µ − 0.5σ 2 f , fx = σf , fxx = σ 2 f .

With Xt = Wt , one has dXt = dWt , (dX )2 = dt. Itô’s lemma gives

df (t, Xt ) = ft dt + fx dW + 0.5fxx (dW )2

  
= f µ − 0.5σ 2 dt + σdW + 0.5σ 2 dt
= f (µdt + σdW ).

Finally, the initial condition is satisfied as f (0, W (0)) = S(0).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science79 / 190
 Results from statistics and stochastic analysis

Sample path and expectation of a GBM with µ = 0.1, σ 2 = 0.1, and S0 = 10

20
Geometric Brownian Motion

18
16
14
12
10

0 2 4 6 8 10

time t

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science80 / 190
 Results from statistics and stochastic analysis

Remark 25 (Motivation: Girsanov’s theorem).

Consider independent N (0, 1) random variables Z1 , . . . , Zn on (Ω, F, IP).

Given γ = (γ1 , . . . , γn ), define a new probability measure Q on (Ω, F) via
( n n
)
X 1X 2
Q(dω) := exp γi Zi (ω) − γi IP(dω).
2
i=1 i=1

Then
( n n
)
X1X 2
Q(Zi ∈ dzi , ∀i) = exp γi zi − γi IP(Zi ∈ dzi , ∀i)
2
i=1 i=1
( n n n
) n
1 X 1X 2 1X 2 Y
= n exp γi zi − γi − zi dzi
(2π) 2 2 2
i=1 i=1 i=1 i=1
n
( )
1 1X 2
= n exp − (zi − γi ) dz1 . . . dzn .
(2π) 2 2
i=1

Hence, the Zi are independent and N (γi , 1)-distributed under Q.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science81 / 190
 Results from statistics and stochastic analysis

Theorem 13 (Girsanov’s theorem).

Let W = (W (1) , . . . , W (d) ) be a d-dimensional BM on (Ω, F, F, IP). With

γ = {γt }t≥0 , a suitable (adapted) d-dimensional process satisfying Novikov’s
condition (E[exp {0.5⟨γ⟩t }] < ∞), let
 Z t
1 t
Z 
Lt := exp − γs′ dWs − ∥γs ∥2 ds .
0 2 0

Define Z t
(i) (i) (i)
W̃t := Wt + γs ds.
0
Under the equivalent probability measure Q with Radon-Nikodým derivative

dQ
= LT ,
d IP

then the process W̃ = (W̃ (1) , . . . , W̃ (d) ) is a d-dimensional Brownian motion.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science82 / 190
 Results from statistics and stochastic analysis

Remark 26 (Girsanov’s theorem: common use).

In the one-dimensional case with constant γ, i.e. for γt ≡ γ, a change of

measure by the Radon-Nikodým derivative
 
1
Lt = exp −γWt − γ 2 t
2

corresponds to a change of drift from µ to µ − γσ.

If F = FW is the Brownian filtration, any pair of equivalent probability measures
Q ∼ IP on F = FT is a Girsanov pair, i.e.

dQ
= Lt
d IP Ft

with L defined as above.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science83 / 190
 Results from statistics and stochastic analysis

Theorem 14 (Representation theorem).

Let M = {Mt }t≥0 be a martingale on (Ω, F, F, IP) with respect to the Brownian
filtration F = FW . Then
Z t
Mt = M0 + Hs dWs , ∀ t ≥ 0,
0

with H = {Ht }t≥0 a suitable process such that (i.e. Ht ∈ L2 )

Z t
Hs2 ds < ∞, t ≥ 0,
0

with IP-probability one.

This means that all Brownian martingales may be represented as stochastic
integrals with respect to Brownian motion.
As a corollary, let C be an FT -measurable random variable with IE [|C|] < ∞.
Then, there exists a process H, as above, such that
Z T
C = IE [C] + Hs dWs .
0

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science84 / 190
 Results from statistics and stochastic analysis

Consider the SDE:

dXt = b(t, Xt )dt + σ(t, Xt )dWt , X 0 = x0 .

Theorem 15 (Feynman-Kac formula ).

Assume b and σ satisfies Lipschitz, growth conditions and h is a measurable

function, then the solution F = F (t, x) of the PDE

1
Ft + bFx + σ 2 Fxx = 0
2
with final condition F (T , x) = h(x) has the stochastic representation

F (t, x) = IE [ h(XT )| Xt = x] ,

where X satisfies the stochastic differential equation

dXs = b(s, Xs )ds + σ(s, Xs )dWs

with initial condition Xt = x.

We assume a solution exist: i.e. there is a finite twice differential function F

satisfying the PDE.
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science85 / 190
 Results from statistics and stochastic analysis

Consider the SDE:

dXt = b(t, Xt )dt + σ(t, Xt )dWt , X 0 = x0 .

Theorem 16 (Discounted Feynman-Kac formula).

Assume b and σ satisfies Lipschitz, growth conditions and h is a measurable

function, then the solution F = F (t, x) of the PDE

1
Ft + bFx + σ 2 Fxx = rF
2
with final condition F (T , x) = h(x) has the stochastic representation
h i
F (t, x) = IE e−r (T −t) h(XT ) Xt = x ,

where X satisfies the stochastic differential equation

dXs = b(s, Xs )ds + σ(s, Xs )dWs

with initial condition Xt = x.

If Xt is a vector then
X 1X 2
Ft + bi Fxi + σi,j Fxi xj = rF
2
i i,j

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science86 / 190
 Financial markets in continuous time

Chapter II
Financial markets in continuous time

Literature for this chapter:

Bingham & Kiesel (2004): Risk-neutral valuation, 2nd ed.
Björk (2004): Arbitrage theory in continuous time.
Delbaen & Schachermayer (2006): The mathematics of arbitrage.
Zagst (2002): Interest rate management.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science87 / 190
 Financial markets in continuous time

Assumptions
Throughout this chapter, let T > 0 be the terminal time horizon.
Uncertainty in the market is modeled by the filtered probability space
(Ω, F, F, IP). The filtration F is assumed to satisfy the usual conditions of
completeness and right-continuity.
Consider d + 1 basic assets, whose price processes are modeled by the
stochastic processes S (0) , . . . , S (d) .

Definition 45 (Numéraire).

A numéraire is a price process which is IP-a.s. strictly positive for t ∈ [0, T ].

In the present setting, use S (0) as numéraire.

Historically, the money market account was usually preferred as numéraire. We
shall later see situations where it is more efficient to use other processes as
numéraire.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science88 / 190
 Financial markets in continuous time

We impose the following set of assumptions on the financial markets.

1. No market frictions:
No transaction costs,
no bid-ask spread,
no taxes,
no restrictions on short sales.
2. No default risk: Implying the same interest rate for borrowing and lending.
3. Competitive markets: Market participants act as price takers.
4. Rational agents: Market participants prefer more to less.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science89 / 190
 Financial markets in continuous time

Definition 46 (Trading strategy).

We call an Rd+1 -valued predictable process φ = {φt }t≥0

(0) (d)

φt = φt , . . . , φt , t ∈ [0, T ],

a trading strategy (or dynamic portfolio process).

Assumption:
R t (i) (i)
Each φ(i) is sufficiently integrable such that 0 φs dSs is well defined.
Interpretation:
(i)
φt denotes the number of shares of asset i held in the portfolio at time t.
Why predictable?
This number has to be determined on the basis of information available before
time t, i.e. the investor selects her or his time t portfolio after observing the
prices St− .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science90 / 190
 Financial markets in continuous time

Definition 47 (Value process, gains process, self-financing).

The value of the portfolio φ at time t is given by

d
(i) (i)
Vtφ := φ′t St =
X
φt St , ∀ t ∈ [0, T ].
i=0

V φ = {Vtφ }t∈[0,T ] is called the value (or wealth) process of the strategy φ.
The gains process Gφ , based on the strategy φ, is defined as
d Z t
(i) (i)
Gtφ :=
X
φs dSs , ∀ t ∈ [0, T ].
i=0 0

A trading strategy φ is called self-financing, if the wealth process V φ satisfies

Vtφ = V0φ + Gtφ , ∀ t ∈ [0, T ].

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science91 / 190
 Financial markets in continuous time

Definition 48 (Discounted processes).

The discounted price process S̃ is defined as

St 
(1) (d)

S̃t := (0)
= 1, S̃t , . . . , S̃t , ∀ t ∈ [0, T ],
St
(i) (i) (0)
with S̃t = St /St , for i = 1, . . . , d.
The discounted wealth process Ṽ φ is defined as
d
Vtφ (0) (i) (i)
Ṽtφ :=
X
(0)
= φt + φt S̃t , ∀ t ∈ [0, T ].
St i=1

Finally, the discounted gains process G̃φ is given by

d Z t
(i) (i)
G̃tφ :=
X
φs d S̃s , ∀ t ∈ [0, T ].
i=1 0

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science92 / 190
 Financial markets in continuous time

Lemma 15 (Numéraire invariance).

Self-financing strategies remain self-financing under a change of numéraire.

Lemma 16 (Self-financing trading strategies).

φ is self-financing if and only if

Ṽtφ = Ṽ0φ + G̃tφ , ∀ t ∈ [0, T ].

We observe that a self-financing strategy is completely determined by its initial

value and the components φ(1) , . . . , φ(d) .
Any set ofR predictable processes φ(1) , . . . , φ(d) , such that the stochastic
integrals φ(i) d S̃ (i) exist, can be uniquely extended to a self-financing strategy
φ with specified initial value Ṽ0φ = v0 by setting the cash holding as

d Z t d
(0) (i) (i) (i) (i)
X X
φt = v0 + φs d S̃s − φt S̃t , ∀ t ∈ [0, T ].
i=1 0 i=1

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science93 / 190
 Financial markets in continuous time

Definition 49 (Arbitrage opportunity).

A self-financing trading strategy φ is called an arbitrage opportunity, if the wealth

process V φ satisfies the following set of conditions:
Zero initial value (no required capital), i.e.

V0φ = 0.

Non-negative terminal value (no loss), i.e.

φ
IP(VT ≥ 0) = 1.

A positive probability of profit, i.e.

φ
IP(VT > 0) > 0.

An arbitrage opportunity can be seen as the chance to create terminal wealth

without initial cost (or risk). This should not be possible in a well-functioning market.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science94 / 190
 Financial markets in continuous time

Definition 50 (Equivalent martingale measure).

A probability measure Q, defined on (Ω, F), is an equivalent martingale measure

(EMM) if
Q is equivalent to IP, i.e. Q ∼ IP,
the discounted price processes S̃ (i) , i = 1, . . . , d, are Q-martingales.

Lemma 17 (Equivalent martingale measure).

(0) Rt
Assume St = Bt := exp( 0 r (s)ds) for some deterministic, non-negative, and
smooth function t 7→ r (t). Then Q ∼ IP is a martingale measure if and only if every
asset price process S (i) has price dynamics under Q of the form
(i) (i) (i)
dSt = r (t)St dt + dMt ,

where M (i) is a Q-martingale (e.g. dMt = Hs dWsQ , WsQ a BM under Q).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science95 / 190
 Financial markets in continuous time

In what follows, we always assume as given the martingale measure Q ∼ IP.

Definition 51 (Admissible strategies).

A self-financing strategy φ is called Q-admissible, if

Z t d Z t
(i) (i)
G̃tφ =
X
φ′s d S̃s = φs d S̃s
0 i=1 0

is a Q-martingale.

By definition, S̃ is a martingale and G̃φ is the stochastic integral w.r.t. S̃.

The set (linear space) of all Q-admissible strategies is denoted as Φ(Q).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science96 / 190
 Financial markets in continuous time

Theorem 17 (EMMs and arbitrage).

Assume that there exists an EMM Q ∼ IP. Then the market model does not contain
any arbitrage opportunities in Φ(Q).

Remark 27 (The converse statement).

The converse, i.e. no arbitrage implying the existence of an EMM, is basically true
as well. One requires, however, a more technical and stronger definition of no
arbitrage: No free lunch with vanishing risk (NFLVR), see Delbaen &
Schachermayer (1994). This establishes the so called first fundamental theorem
of asset pricing.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science97 / 190
 Financial markets in continuous time

Definition 52 (Contingent claim).

A contingent claim X is an FT -measurable random variable such that

X
(0)
=: X̃ ∈ L1 (Ω, FT , Q).
ST

(e.g. E Q [| X |] < ∞)

Definition 53 (Attainable).

A contingent claim X is called attainable, if there exists at least one Q-admissible

trading strategy φ, such that
VTφ = X .
We call such a trading strategy φ a replicating strategy for X .

Definition 54 (Complete market).

The financial market model M is said to be complete, if every contingent claim X ,

(0)
with X /ST ∈ L1 (Ω, FT , Q), is attainable.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science98 / 190
 Financial markets in continuous time
(0)
We now link attainability to martingale representations. Let X̃ := X /ST .

Lemma 18 (Martingale representation).

Let X be a contingent claim and Q be an EMM. Then, X is attainable if and only if

the Q-martingale
Mt := IEQ [X̃ |Ft ], t ∈ [0, T ],
admits a representation of the form
d Z t
(i) (i)
X
Mt = x + φs d S̃s , t ∈ [0, T ],
i=1 0

for some constant x and a Q-admissible trading strategy φ.

Finally, we characterize market completeness.

Theorem 18 (Second fundamental theorem of asset pricing).

A market model that admits at least one EMM Q is complete if and only if Q is
unique.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science99 / 190
 Financial markets in continuous time

Remark 28 (Arbitrage price process).

If the claim X is attainable, X can be replicated by a portfolio φ ∈ Φ(Q).

This means that holding the portfolio φ and holding the contingent claim X are
equivalent from a financial point of view, since they produce the same terminal
payoff.
In the absence of arbitrage, the arbitrage price process ΠXt of the contingent
claim (e.g. price at time t of claim X ) must therefore satisfy

ΠXt = Vtφ , ∀ t ∈ [0, T ].

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
100 / 190
 Financial markets in continuous time

Theorem 19 (Risk-neutral valuation).

The arbitrage price process of any attainable claim X is given by the risk-neutral
valuation formula
h X i
(0) (0)
ΠXt = St IEQ (0) Ft = St IEQ X̃ Ft , ∀ t ∈ [0, T ].
 
ST

Thus, for any two replicating portfolios φ, ψ ∈ Φ(Q)

Vtφ = Vtψ , ∀ t ∈ [0, T ].

Lemma 19 (Bayes formula).

Assume 0 ≤ s < t ≤ T < ∞ and let Y be some Ft -measurable and integrable

random variable. Further, let Z be some positive and FT -measurable random
variable with IEIP [Z ] = 1. Define the martingale

Zt := IEIP [Z |Ft ], ∀ t ∈ [0, T ].

Further, define the measure Q ∼ IP via dQ/d IP = Z . Then, it holds that

1
IEQ [Y |Fs ] = IEIP [YZt |Fs ].
Zs

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
101 / 190
 Financial markets in continuous time

In some situations, using an alternative numéraire simplifies the computation of the

required expectation. This technique is introduced and justified below.
Theorem 20 (Change of numéraire).

Let N = {Nt }t≥0 be a numéraire such that N/S (0) is a Q-martingale. Define the
new measure
(0)
dQN Nt S
Ft := ηt = (0) 0 .
dQ S N0
t

Then, the processes S (i) /N are QN -martingales. Moreover, for attainable X

i h XhX i
(0)
Ft = Nt IEQN
St IEQ (0)
Ft .
NT
ST
h i
(0)
Hence the price of the claim can be found solving St IEQ X(0) Ft or solving
ST
h i
X
Nt IEQN NT Ft whichever is simpler.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
102 / 190
 The Black-Scholes formula

Chapter III
The Black-Scholes formula

Literature for this chapter:

Black & Scholes (1973): The pricing of options and corporate liabilities, Journal
of Political Economy, (81), pp. 637–654.
Bingham & Kiesel (2004): Risk-neutral valuation, 2nd ed.
Hull (2008): Options, futures, and other derivatives, 6th ed.
Wilmott (2007): Paul Wilmott introduces quantitative finance, 2nd ed.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
103 / 190
 The Black-Scholes formula

Definition 55 (The classical Black-Scholes model).

The classical Black-Scholes model, defined on (Ω, F, FW , IP), is given by

dBt = rBt dt, B0 = 1,

dSt = St (bdt + σdWt ), S0 = s0 > 0,

with constant coefficients b ∈ R, r , σ ∈ R+ . The solution of the SDE of the

stock-price process is a geometric Brownian motion.

Compared to the seminal model of Bachelier (1900), where

St = S0 + bt + σWt ,

it is not possible for a stock in the BS-model to reach negative values.

Lemma 20 (Discounted stock price process).

Using the bank account as numéraire, i.e. S̃t := St /Bt , one finds


d S̃t = S̃t (b − r )dt + σdWt .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
104 / 190
 The Black-Scholes formula

Remark 29 (Some first remarks on the Black-Scholes model).

GBM is a reasonable (but not perfect) model for stock price movements.
Shortfalls and extensions are discussed later on.
A major advantage of GBM is its analytical tractability. A vast number of exotic
options can be priced in closed form within this context. This tractability is
typically lost when extensions of the model are considered.
A surprising observation is that the model is robust for hedging. Even though
there might be better models for the pricing of derivatives, BS hedging
strategies turn out to perform very well.
We shall see that the BS-model is free of arbitrage and is complete. This is
extremely convenient for theoretical considerations. However, one should be
aware that as soon as more realistic extensions of the model are considered,
one typically loses the completeness of the model.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
105 / 190
 The Black-Scholes formula

Definition 56 (The generalized Black-Scholes model).

The generalized Black-Scholes model consists of d + 1 assets - the bank

account B and d stocks S (i) , indexed by i = 1, . . . , d. These assets are modeled via
the SDEs

dBt = rt Bt dt, B0 = 1,
 n 
(i) (i) (i) (ij) (j) (i) (i)
X
dSt = St bt dt + σt dWt , S0 = s0 > 0,
j=1

where W = W (1) , . . . , W (n)

is an n-dimensional Brownian motion.
(i)
The coefficients represent the instantaneous short rate ( rt ), drift ( bt ), and
(ij)
volatility ( σt ) at time t.
These coefficients fulfill certain integrability and measurability conditions to
guarantee (unique) solutions of the respective SDEs.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
106 / 190
 The Black-Scholes formula

Remark 30 (Change of measure in the classical BS-model).

Recall: Any equivalent measure Q ∼ IP is a Girsanov pair

dQ
= Lt ,
d IP Ft

with  Z t
1 t 2
Z 
Lt = exp − γs dWs − γs ds .
0 2 0
By Girsanov’s theorem:
dWt = d W̃t − γt dt,
where W̃ is a Q-Brownian motion.
Thus, the Q-dynamics for S̃ are

d S̃t = S̃t ((b − r )dt + σdWt )

 
= S̃t (b − r − σγt )dt + σd W̃t .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
107 / 190
 The Black-Scholes formula

Remark 31 (EMM in the classical BS-model).

Since S̃ has to be a martingale w.r.t. Q, we must have

b − r − σγt = 0, ∀ t ∈ [0, T ].

Therefore, we must choose (in Girsanov’s theorem)

b−r
γt ≡ =: γ,
σ
where γ is called market price of risk.
This shows that the martingale measure Q is unique in this model.
The Q-dynamics of S are
 
dSt = St rdt + σd W̃t .

Theorem 21 (The classical BS-model is free of arbitrage).

This is implied by the existence of the martingale measure Q ∼ IP.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
108 / 190
 The Black-Scholes formula

Remark 32 (Completeness of the classical BS-model).

The next question addressed is completeness of the BS-model.

We know that there exists a unique martingale measure Q ∼ IP (constructed
via γ = (b − r )/σ in Girsanov’s theorem), implying completeness.
However, one can directly show completeness: Let the contingent claim X be
in L1 (Ω, FT , Q), i.e. IEQ [|X |] < ∞. The martingale representation theorem then
allows to show the existence of a replicating strategy.

Theorem 22 (The classical BS-model is complete).

The proof is performed along the outline of the remark above.

If n > d then market is incomplete (no replicating strategy), not unique EMM (no
arbitrage oppor.) Examples: stochastic volatility models.
If n < d then infinitely many replicating strategies, no EMM (arb. opport., no γ such
that µi − σi γ = r for all i.)

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
109 / 190
 The Black-Scholes formula

Theorem 23 (Completeness of the generalized BS-model).

The following statements are equivalent:

1. There exists a unique EMM Q ∼ IP for the discounted price processes S̃.
2. The generalized BS-model is complete, i.e. every contingent claim X with
X /BT ∈ L1 (Ω, FT , Q) is attainable.
3. Equality n = d holds (the number of assets d equals the number of sources of
risk n) and the volatility matrix σt is non-singular Q-a.s. for Lebesque-a.e.
t ∈ [0, T ].

Proof: See Bingham and Kiesel (2004).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
110 / 190
 The Black-Scholes formula

Example 11 (Futures and forwards in the BS-model).

1. When the stock St is discounted by the bond Bt , the martingale measure Q is

unique. As the BS-model is complete, we only require a contingent claim X to
be FT -measurable and integrable to be conveniently priced by the risk-neutral
valuation formula.
2. This allows to compute the time t value of a forward contract with forward price
K and maturity T , which is given by

ΠKt = St − Ke−r (T −t) .

3. From this, we can obtain the forward price F (t, T ) of a forward contract with
maturity T on the stock S. We obtain

F (t, T ) = er (T −t) St .

Recall: F (t, T ) is the K such that ΠKt = 0.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
111 / 190
 The Black-Scholes formula

Remark 33 (The Black-Scholes formula).

Consider a European call option with terminal payoff C(ST , T ) = (ST − K )+ in a

BS-model. The price (at time 0 ≤ t ≤ T ) of this option was first derived by Black
and Scholes (1973) via a hedging argument. Since then, several other derivations
of the Black-Scholes formula are known:
1. The Black-Scholes partial differential equation (PDE) is derived via hedging
arguments (or via the Feynman-Kac formula). The solution of this PDE is the
Black-Scholes formula.
2. The Q-expectation of the discounted payoff (ST − K )+ is computed directly.
3. The Black-Scholes formula can also be derived as the limit of a sequence of
(suitably parameterized) Cox-Ross-Rubinstein models (Binomial trees).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
112 / 190
 The Black-Scholes formula

Theorem 24 (The BS-call formula).

For a European call option with terminal payoff C(ST , T ) = (ST − K )+ , we obtain
the following formula. The Black-Scholes price process of a European call option
with strike K and maturity T > t is given by

C(St , t) = St Φ d1 (St , T − t) − Ke−r (T −t) Φ d2 (St , T − t) .

The functions d1 (s, t) and d2 (s, t) are given by

σ2 σ2
log(s/K ) + (r + 2 )t log(s/K ) + (r − 2 )t
d1 (s, t) = √ , d2 (s, t) = √ ,
σ t σ t
where Φ(x) is the cumulative distribution function of the N (0, 1)-distribution.

Remark 34 (The BS-put formula).

Using similar arguments as in the following proofs (or the put-call parity)

C(St , t) + Ke−r (T −t) = P(St , t) + St ,

the BS-put price is found to be

P(St , t) = e−r (T −t) K Φ − d2 (St , T − t) − St Φ − d1 (St , T − t) .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
113 / 190
 The Black-Scholes formula

Remark 35 (Proof of the BS-formula via the BS PDE).

We assume the time t value of a call in the BS-model to be a smooth function

C(St , t) of the current stock price and time.
Now consider a portfolio Π consisting of one call C and a short position in ξ
stocks. The value of this portfolio at time t is

Πt = C(St , t) − ξSt .

We use Itô’s formula to find the dynamics of the portfolio at time t. We obtain
(omitting arguments for notational simplicity)

∂ ∂ 1 ∂2
dΠt = C dt + C dSt + σ 2 St2 2 C dt − ξdSt .
∂t ∂S 2 ∂ S

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
114 / 190
 The Black-Scholes formula

Remark 36 (Proof of the BS-formula via the BS PDE - 2).

We rearrange the last line to

∂ 1 ∂2   ∂ 
dΠt = C + σ 2 St2 2 C dt + C − ξ dSt .
∂t 2 ∂ S ∂S
∂
If we now choose ξ = ∂S C, then the portfolio is immune (at time t) to changes
in S. Such a reduction of randomness / risk is called hedging.
The dynamics of the hedged portfolio are
∂ 1 ∂2 
dΠt = C + σ 2 St2 2 C dt.
∂t 2 ∂ S
The (Nobel Prize winning) idea is to recognize that the return of a riskless
portfolio must agree with the return of the riskless bank account. Every other
return creates an arbitrage opportunity! Therefore

dΠt = r Πt dt.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
115 / 190
 The Black-Scholes formula

Remark 37 (Proof of the BS-formula via the BS PDE - 3).

Putting these equations together, we find

∂ 1 ∂2   ∂ 
C + σ 2 St2 2 C dt = r C − St C dt
∂t 2 ∂ S ∂S
or differently, which is known as the BS partial differential equation (PDE)

∂ 1 ∂2 ∂
C + σ 2 St2 2 C + rSt C − rC = 0.
∂t 2 ∂ S ∂S
So far, we did not use that C is a call option. In fact, the same derivation holds
for all options! The specific payoff of C is now absorbed into the boundary
condition of the solution of the above PDE. It is required that

C(ST , T ) = (ST − K )+ .

We can now verify (which is done later in an exercise) that the suggested
BS-formula fulfills the required BS PDE and the boundary condition.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
116 / 190
 The Black-Scholes formula

Remark 38 (Proof of the BS-formula via the BS PDE - 4).

Uniqueness of the solution can be shown using results on PDEs.

Be aware that our derivation is heuristic in the sense that we had to assume (in
the beginning) that the BS-formula is a smooth function of t and S.
One point in this (original) derivation is also missing: The suggested trading
strategy is not self-financing!
Still, a mathematically precise derivation of the BS PDE can be found using the
(discounted) Feynman-Kac formula.
Solving the BS PDE is also tricky, we just verify the solution as exercise. A
roadmap for the construction of the solution is given in Wilmott (2007).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
117 / 190
 The Black-Scholes formula

Remark 39 (Proof of the BS-formula via risk-neutral valuation).

We have shown that pricing in the BS-market can be done using the
risk-neutral valuation formula.
Hence, the price of the call option is the discounted expected future payoff,
where the expectation is computed w.r.t. the martingale measure Q.
Recall that the Q-solution of dSt = St (rdt + σd W̃t ) is given by
 1

ST = S0 exp r − σ 2 T + σ W̃T .
2
We now have to compute (w.l.o.g. t = 0)
h i
C(S0 , 0) = IEQ e−rT (ST − K )+ .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
118 / 190
 The Black-Scholes formula

Remark 40 (Proof of the BS-formula via risk-neutral valuation - 2).

Define the set of all ω ∈ Ω for which the option is drawn as

A := {ω ∈ Ω : ST ≥ K }.

This allows to rewrite the pricing formula to

h i
IEQ e−rT (ST − K )+ = e−rT IEQ [ST 1A ] − e−rT K Q(A).

A computation using the Q-solution of ST gives

σ2
  
log(S0 /K ) +
Q(A) = P(ST > K ) = P σ W̃T > log(K /S0) − (r − )T . . . = Φ √
2 σ

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
119 / 190
 The Black-Scholes formula

Remark 41 (Proof of the BS-formula via risk-neutral valuation - 3).

Simplifying the other term is more tricky. At first, we write

h 2
i
e−rT IEQ [ST 1A ] = e−rT S0 IEQ e(r −0.5σ )T +σW̃T 1A .

We now introduce another measure R ∼ Q, with

dR 2
= eσW̃T −0.5σ T ,
dQ

where Wt∗ = W̃t − σt is the R-BM.

A lengthy computation shows

log(S0 /K ) + (r + 0.5σ 2 )T
 
−rT
e IEQ [ST 1A ] = . . . = S0 R(A) = . . . = S0 Φ √ .
σ T

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
120 / 190
 The Black-Scholes formula

Remark 42 (Discussing the assumptions of the BS-formula).

The stock price process is a GBM:

This assumption is problematic, as stock returns are leptocurtic, heavy tailed,
and have jumps. Also, volatility clusters are omnipresent.

Generalizations of the BS-model using more general Lévy processes (with

jumps) are available. Also, there exists an extensive literature on stochastic
volatility models.
The interest rate is constant:
This is not true. However, explicit formulas for options are still known if the
term-structure of interest rates is a deterministic function of the time t. Also,
there exists an extensive literature on stochastic-interest rates.
There are no dividends:
This assumption will be relaxed (to some extent) in the following.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
121 / 190
 The Black-Scholes formula

Remark 43 (Discussing the assumptions of the BS-formula - 2).

Delta hedging can be done continuously:

This assumption is also problematic, since continuously rebalancing a portfolio
is impossible and existing bid-ask spreads would make such an approach too
expensive. If delta hedging (continuous and without costs) was possible,
options were redundant objects. The pure existence of option markets shows
that this assumption does not hold.
There are no transaction costs:
For some markets, transaction costs are very small, but still they exist.
However, the amount of transaction costs depends on the liquidity of the
respective market.
There are no arbitrage opportunities:
This is wrong. In fact, there is a whole industry exploiting even the smallest
arbitrage opportunity. Important for the derivation of our formulas is that there
is no arbitrage opportunity within the model!

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
122 / 190
 The Black-Scholes formula

Remark 44 (Motivation: The ’Greeks’).

The Black-Scholes formula for European call / put options is a function of

several underlying variables (or risk factors).
We can analyze the impact of these underlying parameters on the prices of call
and put options.
The Black-Scholes option values depend on the (current) stock price S0 , the
volatility σ, the time to maturity T , the interest rate r , and the strike price
K . In this sense, we consider C and P as functions of (S0 , T , σ, r , K ).
The sensitivities of the option price (with respect to the first four parameters)
are called the Greeks and are widely used for hedging purposes.
We can determine the impact of these parameters by taking partial derivatives.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
123 / 190
 The Black-Scholes formula

Lemma 21 (Black-Scholes Delta).

The Delta of a Black-Scholes call option is calculated as the first-order partial

derivative of the option value with respect to the price of the stock. It is given by

∂C
∆C := = Φ(d1 ) > 0.
∂S0

Remark 45 (Delta of a put option).

Using the put-call parity, the Delta of the corresponding put option is found to be

∂P
∆P := = Φ(d1 ) − 1 < 0.
∂S0

Delta is the most important and widely used short-term risk measure.
The Delta of an option shows how the option value changes when the price of
the underlying marginally varies (under the assumption that everything else
stays constant).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
124 / 190
 The Black-Scholes formula

Example 12 (Black-Scholes Delta).

A stock is currently traded at S0 = 2 700. The exercise price of the considered

call and put option is K = 2 600, the constant riskless interest rate is r = 3%,
the volatility is σ = 20%, and the time to maturity is two months, i.e. T = 0.167
years.
The premium for the call (resp. put) option is

C = 154.18, P = 41.19.

The option deltas for the call (put) option are

Φ(d1 ) = 0.7136, Φ(d1 ) − 1 = −0.2864.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
125 / 190
 The Black-Scholes formula

Black-Scholes Delta (Call), r = 0.1, K = 100, σ = 0.15, and T = 1

1.0
60
50

0.8
40

0.6
Option Value

Delta
30

0.4
20

0.2
10

0.0
0

60 80 100 120 140 60 80 100 120 140

Stock Price Stock Price

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
126 / 190
 The Black-Scholes formula

Lemma 22 (Black-Scholes Gamma).

The Gamma of a Black-Scholes option is calculated as the second-order partial

derivative of the option value with respect to the underlying price S0 . It is given by
(for both, the put and the call option)

∂2C ∂2P φ(d1 )

Γ := 2
= 2
= √ > 0,
∂S0 ∂S0 S0 σ T

where φ(x) is the density of a N (0, 1)-distribution.

Gamma is the sensitivity of Delta w.r.t. the underlying price.

Gamma shows the change of Delta under a marginally varying underlying.
It displays the stability of Delta (under the assumption that everything else
stays constant). In this regard, it measures how often a position must be
re-hedged in order to keep it Delta-neutral.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
127 / 190
 The Black-Scholes formula

Example 13 (Black-Scholes Gamma).

Using the same options and parameters as in Example 12, the Gamma of the
put and call option at time t = 0 is

Γ = 0.0015.

This means that the Delta of both options changes about 0.0015, if the stock
price increases by 1 unit of currency.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
128 / 190
 The Black-Scholes formula
Black-Scholes Gamma (Call), r = 0.1, K = 100, σ = 0.15, and T = 1

1.0
0.8
0.6
Delta

0.4
0.2
0.0

60 80 100 120 140

Stock Price
0.030
0.025
0.020
Gamma

0.015
0.010
0.005
0.000

60 80 100 120 140

Stock Price

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
129 / 190
 The Black-Scholes formula

Lemma 23 (Black-Scholes Theta).

Let r , S0 , and σ be constant. The Theta of a Black-Scholes option is calculated as

the first-order partial derivative of the option value with respect to time T ( t = 0)
and is given by

∂C S0 σφ(d1 ) ∂P S0 σφ(d1 )
ΘC := = √ + Kre−rT Φ(d2 ), ΘP := = √ − Kre−rT Φ(−d2 ).
∂T 2 T ∂T 2 T
Caution: Some authors introduce Theta as the derivative with respect to t:

∂C St σφ(d1 ) ∂P St σφ(d1 )
=− √ − Kre−r (T −t) Φ(d2 ), =− √ + Kre−r (T −t) Φ(−d2 ).
∂t 2 T −t ∂t 2 T −t

Theta is the sensitivity of the option price w.r.t. the time to maturity T (under
the assumption that everything else stays constant).
Theta measures the result of decay in time on an option.
It is especially important for short-maturity options, as the probability of a call
(or put) reaching moneyness falls with shorter maturity, which lowers the option
value.
Note "at the money" means S0 = K , "our of the money" is S0 < K (for Calls),
"in the money" implies S0 > K (for Callls)
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science
130 / 190
 The Black-Scholes formula
Black-Scholes Theta (ΘC , w.r.t. t), r = 0.1, K = 100, and σ = 0.15

60
50
40
Option Value

30
20
10
0

0 2 4 6 8 10

Time to Maturity
−4
−6
Theta

−8
−10

at the money
in the money
−12

out of the money

0 2 4 6 8 10

Time to Maturity

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
131 / 190
 The Black-Scholes formula

Lemma 24 (Black-Scholes Vega).

The Vega of a Black-Scholes option is calculated as the first-order partial derivative

of the option value with respect to the volatility σ. We obtain for the put and call
option
∂C ∂P √
V := = = S0 T φ(d1 ) > 0.
∂σ ∂σ

Vega is the sensitivity of the option price w.r.t. the volatility σ.

Vega shows the change of the option price under a marginally varying volatility
(given the assumption that everything else stays constant).
Vega is always positive, i.e. an increasing volatility always leads to increasing
option prices. Later on, this turns out to be important for the definition of
implied volatilities.
option values approaches S as σ → ∞, while it approaches S − Ke−r (T −t) as
σ→0

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
132 / 190
 The Black-Scholes formula
Black-Scholes Vega (Call), r = 0.1, K = S0 = 100, and T = 1

20
Option Value

15
10

0.0 0.1 0.2 0.3 0.4 0.5

Volatility
30
20
Vega

10
0

0.0 0.1 0.2 0.3 0.4 0.5

Volatility

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
133 / 190
 The Black-Scholes formula

Lemma 25 (Black-Scholes Rho).

The Rho of a Black-Scholes option is calculated as the first-order partial derivative

of the option value with respect to the riskless interest rate and is therefore given by

∂C
ρC := = TKe−rT Φ(d2 ) > 0,
∂r
∂P
ρP := = −TKe−rT Φ(−d2 ) < 0.
∂r

Rho is the sensitivity of the option price w.r.t. the riskless interest rate.
Rho shows the change of the price under a marginally varying riskless interest
rate (under the assumption that everything else stays constant).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
134 / 190
 The Black-Scholes formula
Black-Scholes Rho (Call), K = S0 = 100, σ = 0.15, and T = 1

20
Option Value

15
10

0.00 0.05 0.10 0.15 0.20 0.25

Interest Rate
70
65
Rho

60
55
50

0.00 0.05 0.10 0.15 0.20 0.25

Interest Rate

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
135 / 190
 The Black-Scholes formula

Example 14 (Delta hedging).

A bank sells 2 000 calls with a Delta per call of 0.6.

When the underlying increases by 1 Euro, the option position decreases by
2 000 · 0.6 Euro, implying a loss of 1 200 Euro.
This (hypothetical) loss can be hedged by buying 1 200 stocks (note that the
Delta per stock is obviously 1).
The resulting position has a Delta of

−2 000 · 0.6 + 1 200 = 0

and is therefore Delta neutral.

The bank is Delta-hedged against smaller movements in the stock price and is
especially not exposed to losses from changes in option values.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
136 / 190
 The Black-Scholes formula

Example 15 (Delta-Gamma hedging).

A bank sells 2 000 calls with a Delta of 0.6 and a Gamma of 0.015 (per call).
The total Delta is −1 200, the total Gamma is −30.00.
Another option on the same stock is available in the market with a Delta of 0.5
and a Gamma of 0.02.
Gamma neutrality can be created by buying 1 500 units of the second option
(note that the Gamma per stock is obviously 0).
Delta neutrality can be created by additionally buying 450 stocks with Delta 1
and Gamma 0.
The total position satisfies:

Gamma: −2 000 · 0.015 + 1 500 · 0.02 + 450 · 0 = 0,

Delta: −2 000 · 0.6 + 1 500 · 0.5 + 450 · 1 = 0.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
137 / 190
 The Black-Scholes formula

Remark 46 (Motivation: Implied volatility).

One of the main issues raised, when using the Black-Scholes formula, is the
question of modeling / selecting the volatility σ. All other required quantities
(S0 , T , r , K ) are (more or less) directly observable.
Before we can implement the Black-Scholes formula to price options, we first
have to specify σ.
Recall that Vega is given by

∂C √
V := = S0 T φ(d1 ) > 0.
∂σ
The important thing to note is that Vega is always positive.
Next, since Vega is positive and C is differentiable, C is a strictly increasing
function of σ.
Turning this round, σ is a continuous (differentiable) strictly increasing function
of C.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
138 / 190
 The Black-Scholes formula

Definition 57 (Implied volatility).

Assume as given the market price of a call option C market (T , K ).

Inverting the Black-Scholes formula for σ corresponds to choosing σ such that
the market value C, at which the call option is observed, is matched by the
call-formula:
C market (T , K ) = C(S0 , T , σTimplied
,K , r , K ).

The value σTimplied

,K obtained in this way is called implied volatility.

Remark 47 (Implied volatility).

For this procedure, a numerical routine (rooter) is required.

The implied volatility obtained in this way can be used to price other (more
exotic) options in the BS-framework.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
139 / 190
 The Black-Scholes formula

Implied volatility, given C market (T = 1, K = 100) = 11.2, r = 0.1, S0 = 100

13
Option Price in a BS!model

12
C=11.2
11
10

Impl. Vola
0.00 0.05 0.10 0.15 0.20

Volatility

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
140 / 190
 The Black-Scholes formula

Implied vol. surface of EURO STOXX 50, as of April 16, 2010, S0 = 2 949

"!
&"
&!
%"
&"'"!
%!
&!'&"
$"
%"'&!
$!
%!'%"
#"
$"'%!
#!
$!'$"
"
#"'$!
!
#!'#"
"'#!
!'"

Note that σTimplied

,K is not constant across different T and K .
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science
141 / 190
 The Black-Scholes formula

Implied volatility: some final considerations Derman (2004): My life as a quant.

Page 226: ”One of the things you learn repeatedly in a career in financial
modeling is the importance of units. You always want the prices of securities to
be quoted in a way that make it easy to compare their relative values. [...]”
”In the options world [...], price alone is an insufficient measure of value;
it’s impossible to tell whether 300 for an at-the-money put is more attractive
than 40 for a deep out-of-the-money put. A better measure of value is the
option’s implied volatility.”

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
142 / 190
 The Black-Scholes formula

Historical volatility (also called realized volatility)

Log-returns (over disjoint
periods of length h) in the BS-model are independent
and N (b − 0.5σ 2 )h, σ 2 h distributed under the physical measure IP. This
easily follows from the properties of W = {Wt }t≥0 .
Assume the asset to be observable at times

{0, h, 2h, . . . , nh} = {t0 , t1 , . . . , tn }.

The log-return over period i is Yi := log(Sti /Sti−1 ), i = 1, . . . , n.

The standard estimate for σh is
v
u
u 1 X n
σ̂h = t (Yi − Ȳ )2 .
n−1
i=1
√
This gives σ̂ hist = σ̂h / h.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
143 / 190
 The Black-Scholes formula

Dividends
Remark 48 (Options on dividend paying assets).

So far, we disregarded dividend payments.

For simplicity (to avoid jumps in the asset process and to remain in the present
stochastic framework), we now assume the holder of the stock to receive a
continuous and constant dividend yield.
This dividend is proportional to the stock price (with factor D). Hence, in
[t, t + dt], the holder of one stock receives the amount

DSt dt.

Again, assume the time t value of the option to be a (smooth) function of the
current time and stock value, denoted by V (St , t).
Now consider a portfolio Π consisting of one option V and a short position in ξ
stocks. The value of this portfolio at time t is

Πt = V (St , t) − ξSt .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
144 / 190
 The Black-Scholes formula

Dividends
Remark 49 (Options on dividend paying assets - the new PDE).

We again use Itô’s formula to find the dynamics of the portfolio at time t. We
obtain
∂ ∂ 1 ∂2
dΠt = Vdt + VdSt + σ 2 St2 2 Vdt − ξdSt − DξSt dt.
∂t ∂S 2 ∂S
Similar considerations as before lead to the following PDE for V :

∂ 1 ∂2 ∂
V + σ 2 S 2 2 V + (r − D)S V − rV = 0.
∂t 2 ∂S ∂S
The specific payoff of V is again absorbed into the boundary condition of the
solution of the above PDE.
The solution to the above PDE is the respective pricing formula under the
assumption of proportional dividends.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
145 / 190
 The Black-Scholes formula

Dividends
Remark 50 (Options on dividend paying assets - martingale pricing (1)).

Model (under the real-world measure IP):

dSt = St ((µ − D)dt + σdWt )

Reasoning: Dividend payments reduce the stock price.

Problem: RNVF requires a non-dividend paying asset.
(1)
Idea: Invest the dividend in the stock again. Start with position φ0 = 1 and
continue via
(1) (1) DSt dt (1)
dφt = φt = φt Ddt.
St
(1)
Thus φt = eDt , and the value of this position in St is Yt = eDt St .
Yt can be considered as a tradable, non-dividend paying asset.
Thus Yt fits to RNVF (and to the standard Black-Scholes framework).
Choose the EMM Q such that Ỹt := e−rt Yt is a Q-martingale.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
146 / 190
 The Black-Scholes formula

Dividends
Remark 51 (Options on dividend paying assets - martingale pricing (2)).

A payoff f (ST ) at time T is equivalent to f (e−DT YT ). Now we can apply RNVF.

This is equivalent to combining RNVF with the distribution of St under Q.
The Q-dynamics of St is
 
dSt = St (r − D)dt + σd W̃t .

Intuition: The Q-drift of St is reduced by the dividend payments

(same effect as for the IP-drift).
Formal explanation: Girsanov for Ỹt , then product rule for dSt = d(e(r −D)t Ỹt ).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
147 / 190
 The Black-Scholes formula

Dividends
Remark 52 (Calls and puts on dividend paying assets).

Computing the discounted Q-expectation gives the European call price

C(St , t) = St e−D(T −t) Φ d1 (St , T − t) − Ke−r (T −t) Φ d2 (St , T − t) .

The functions d1 (s, t) and d2 (s, t) are given by

σ2
log(s/K ) + (r − D + 2 )t
d1 (s, t) = √ ,
σ t
and
σ2
log(s/K ) + (r − D − 2 )t
d2 (s, t) = √ .
σ t
The price of the corresponding European put option is given by

P(St , t) = −St e−D(T −t) Φ − d1 (St , T − t) + Ke−r (T −t) Φ − d2 (St , T − t) .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
148 / 190
 Stochastic volatility models

Chapter IV
Stochastic volatility and jump models

Literature for this chapter:

Barndorff-Nielsen, O.E. and Shepard, N. (2001): Non-Gaussian
Ornstein-Uhlenbeck-based models and some of their uses in Financial
Economics (with discussion). Journal of the Royal Statistical Society, Series B,
Vol. 63, pp. 167–241.
Heston, S.L. (1993): A closed-form solution for options with stochastic volatility
with applications to bond and currency options. The Review of Financial
Studies, Vol. 6, pp. 327–343.
Merton, R.C. (1976): Option pricing when underlying stock returns are
discontinuous. Journal of Financial Economics, Vol. 3, pp. 125–144.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
149 / 190
 Stochastic volatility models

Literature for this chapter (cont.):

Hull, J. and White, A. (1987): The pricing of options on assets with
stochastic volatilities. The Journal of Finance, Vol. 42(2), pp.
281–300.
Raible, S. (2000): Lévy processes in finance. Theory, numerics,
and empirical facts. Dissertation, Universität Freiburg.
Stein, E.M. and Stein, J.C. (1991): Stock price distribution with
stochastic volatility: An analytic approach. Review of Financial
Studies, Vol. 4, pp. 727–752.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
150 / 190
 Stochastic volatility models

Remark 53 (Introduction to stochastic volatility models).

A core assumption of the BS-model is the asset-value process being a GBM.

This assumption is problematic, e.g., for the following reasons:
a) stock returns are leptocurtic and heavy-tailed;
b) stock processes have jumps;
c) volatility clusters are omnipresent.
To overcome these shortfalls, it is possible to:
a) and b): Use other distributions for returns than the normal distribution, i.e.
replace the Brownian motion of the model by more general Lévy processes.
Such a generalization allows for jumps and more realistic return distributions.
c): There exists an extensive literature on stochastic volatility models. These
models are used to explain phenomena such as volatility clusters and a
negative leverage effect. Such models are introduced in what follows.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
151 / 190
 Stochastic volatility models

Volatility as a deterministic function of t

Below, we successively loosen the restriction of a constant volatility.
In a first step, we choose t 7→ σ(t) to be a deterministic (positive) function.
In this setting, we can still derive many results from the BS-model.

Definition 58 (The generalized Black-Scholes model).

The generalized Black-Scholes model with deterministic coefficients consists

of a bank account B = {Bt }t≥0 and a risky asset S = {St }t≥0 . These assets are
modeled via the SDEs

dBt = r (t)Bt dt,B0 = 1,

dSt = St b(t)dt + σ(t)dWt , S0 > 0.

The coefficients represent the instantaneous short rate r (t) ≥ 0, drift b(t), and
volatility σ(t) > 0 at time t, which satisfy certain technical conditions to guarantee
that the model is well-defined.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
152 / 190
 Stochastic volatility models

Theorem 25 (The generalized Black-Scholes call formula).

Consider a European call option with payoff C(ST , T ) = (ST − K )+ . The

generalized Black-Scholes price process of the European call is given by
RT
C(St , t) = St Φ d1 (St , T − t) − Ke− t r (u)du Φ d2 (St , T − t) .

The functions d1 and d2 are given by

RT RT
r (u)du + 12
log(s/K ) + t t σ(u)2 du
d1 (s, t) = qR ,
T 2
t σ(u) du
s
Z T
d2 (s, t) = d1 (s, t) − σ(u)2 du.
t

The respective BS-put option formula is found by applying the put-call parity:
RT
P(St , t) = Ke− r (u)du



t Φ − d2 (St , T − t) − St Φ − d1 (St , T − t) .

The key isR the representation:

T 2
RT
ST = St e t (r (u)−σ (u)/2)du+ t σ(u)dWu
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science
153 / 190
 Stochastic volatility models

1-year rolling window volatility estimates of the Dax

The classical Black-Scholes model assumes a constant volatility.
In reality, however, volatility changes in random pattern over time.
0.030
0.025
Hist. daily vol. (1−year window)

0.020
0.015
0.010

0 1000 2000 3000 4000 5000

Dax Returns: 4.01.1988 − 13.11.2008

√
Note is σd and σy are vol per day and year resp. then σy = σd 250, e.g.
σd = 0.015 implies σy = 0.237

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
154 / 190
 Stochastic volatility models

Historical daily log-returns and simulated returns

Log−Return

0.05
−0.10

0 1000 2000 3000 4000 5000

Dax Returns: 4.01.1988 − 13.11.2008

Log−Return

0.05
−0.10

0 1000 2000 3000 4000 5000

Simulated Returns (with historical mean and volatility)

We observe that simulated BS-returns do not exhibit volatility clusters.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
155 / 190
 Stochastic volatility models

Why should one use stochastic volatility models?

We just saw that (realized) volatility of stocks, indexes, etc. changes over time.
A good stochastic model for the respective underlying takes this observation
into account and exhibits similar statistical properties.
Options on (realized) volatility become increasingly popular (vola-swaps,
moment swaps, etc.). Pricing such options necessarily requires a stochastic
model for the volatility. In this case, volatility is the underlying itself.
Trading strategies on volatility require a dynamic model for the volatility.
Falling stock prices induce more rebalancing (restructuring of portfolios) in
stock positions than rising stock prices. For instance, due to regulatory
requirements. This increases the volatility of the stock! Explaining such a
connection requires a correlation of stock movements and the volatility. Hence,
a stochastic model for both objects.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
156 / 190
 Stochastic volatility models

Definition 59 (Stochastic volatility model).

A (continuous) stochastic volatility model is a stock price model where the stock
price volatility follows its own Itô-process, i.e.

dSt = b(St , σt , t)dt + c(St , σt , t)dWtS , S0 > 0,

dσt = f (σt , t)dt + g(σt , t)dWtσ , σ0 > 0,

where the functions b, c, f , and g have to fulfill certain integrability and

measurability conditions such that the model is well-defined.

Instead of modeling the volatility {σt }t≥0 , some models prefer postulating some
SDE for the variance process vt = σt2 , instead. Changing from one
specification to the other is done via Itô’s formula.
Important: Due to the additional risk of a stochastic volatility, these models are
no longer complete. In the following, we state all models w.r.t. some martingale
measure Q and do not discuss the (difficult!) question of a reasonable change
of measure from IP to Q.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
157 / 190
 Stochastic volatility models

Remark 54 (Stochastic volatility model: Uncorrelated Brownian motions).

If, w.r.t. Q, b(St , σt , t) = rSt , c(St , σt , t) = σt St , and if the processes WtS and Wtσ
are uncorrelated Brownian motions, the price of a European call option can be
calculated as the call price in a classical Black-Scholes model with the average
volatility over the lifetime of the option, integrated over the distribution of the
average volatility (using Tower property), i.e.
Z ∞
CSV (S0 , 0) = CBS (S0 , 0; σ)hσ (σ)dσ,
0

where CBS (S0 , 0; σ) is the standard BS-price (with volatility being a function of
RT
1/2
time), σ = T1 0 σu2 du the average volatility, and hσ the density of the average
volatility.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
158 / 190
 Stochastic volatility models

Definition 60 (Heston (1993) model).

Heston’s model is defined (w.r.t. Q) by the SDEs

√
dSt = rSt dt + vt St dWtS , S0 > 0,
√
dvt = κ(θ − vt )dt + α vt dWtv , v0 > 0,

where the Brownian motions are correlated with parameter ρ ∈ (−1, 1), i.e.

dWtS dWtv = ρdt.

The parameters represent the following quantities:

r is the rate of return of the underlying asset;
θ is the long vol, i.e. the long run average variance;
κ is the mean reversion speed, i.e. the rate at which vt reverts to θ;
α is the vol of vol, i.e. the volatility of the variance process.
Note vt does not satisfy Lispchitz’s condition. It can be weakened to "locally
Lispchitz" (on any compact set).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
159 / 190
 Stochastic volatility models

Simulated returns in the Heston model

0.6
v(t)

0.0

0 2 4 6 8

Time
Log−Return(t)

0.1
−0.2

0 2 4 6 8

Time

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
160 / 190
 Stochastic volatility models

Remark 55 (Heston model).

The volatility process can reach zero, unless the Feller condition
holds:
1
κθ > α2 .
2

The Heston model achieves a calibration to today’s observed plain

vanilla option prices by balancing the probabilities of very high
volatility scenarios against those where future instantaneous
volatility drops to very low levels.
The average time the volatility stays at high or low levels is
measured/adjusted by the mean reversion scale 1/κ.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
161 / 190
 Stochastic volatility models

The characteristic function of sT := log(ST ) in the Heston model (w.r.t. Q) is

known in closed form.
This formula is required for the pricing of plain vanilla options via the
Fourier-pricing technique (as we shall see later on):

ϕsT (u) = IE eiusT

 


= exp iu(log(S0 ) + rT )
· exp θκα−2 ((κ − ραiu − d)T − 2 log((1 − ge−dT )/(1 − g)))

· exp v0 α−2 (κ − ραiu − d)(1 − e−dT )/(1 − ge−dT ) ,

where
κ − ραiu − d
q
d= (ραui − κ)2 + α2 (iu + u 2 ), g= .
κ − ραiu + d

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
162 / 190
 Stochastic volatility models

Advantages of the Heston model:

The model is simple to understand and interpret.
The characteristic function of log(St ) is known in closed-form.
Shortfalls of the Heston model:
Jumps in volatility or stock price are not supported.
Extension with jumps: Bates model, see Bates (1996).
Short-term options are underpriced (due to neglected jump risk).
Tedious calculations, due to the non-linearity of the Riccati-ODE (which occurs
in the derivation of the characteristic function of the log-stock price).
Further remarks:
Historical estimation of the model is difficult, possible approaches are the use
of empirical characteristic functions or moment matching.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
163 / 190
 Stochastic volatility models

Definition 61 (Stein and Stein (1991) model).

The model of Stein and Stein is defined by the SDEs (w.r.t. Q)

dSt = rSt dt + σt St dWtS , S0 > 0,

dσt = κ(θ − σt )dt + αdWtσ , σ0 > 0,

where the Brownian motions are correlated with parameter ρ ∈ (−1, 1), i.e.

dWtS dWtσ = ρdt.

The parameters have a similar interpretation as in Heston’s model.

In contrast to Heston’s model, where the variance follows a CIR-process, the
volatility in the Stein and Stein model follows a Vasicek process which might
become negative.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
164 / 190
 Stochastic volatility models

Simulated returns in the Stein and Stein model

0.50
sigma(t)

0.35

0 2 4 6 8

Time
Log−Return(t)

0.1
−0.2

0 2 4 6 8

Time

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
165 / 190
 Stochastic volatility models

Remark 56 (Stein and Stein model).

The distribution of the volatility converges to a Gaussian distribution with mean

θ and variance σ 2 /(2κ). This is found by solving the SDE of σ, recall the
example presented earlier.
Since the sign of σt might be interpreted as a sign modifier of the correlation,
we have the consequence that the sign of the correlation between movements
of the underlying and the volatility may change.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
166 / 190
 Stochastic volatility models

Advantages of the Stein and Stein model:

The characteristic function of log(St ) is known (but difficult).
The model is simple to interpret and understand.
Shortfalls of the Stein and Stein model:
May have negative volatilities.
The direction of the leverage effect may suddenly change.
Jumps are not supported.
The volatility is often quite small.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
167 / 190
 Stochastic volatility models

Definition 62 (Hull and White (1987) model).

The model of Hull and White is defined by the SDEs

√
dSt = rSt dt + vt St dWtS , S0 > 0,
dvt = µv vt dt + ξvt dWtv , v0 > 0,

where the Brownian motions are correlated with parameter ρ ∈ (−1, 1), i.e.

dWtS dWtv = ρdt.

The special case µv = α2 and ξ = 2α is also called Hagan model.

The parameters represent the following quantities:

µv is the rate of return of the variance process;
ξ is the vol of vol, i.e. the volatility of the variance process.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
168 / 190
 Stochastic volatility models

Remark 57 (Hull-White model).

Since vt is log-normally distributed, we have

√ 1 1 2
v0 e 2 µv t− 8 ξ t ,
IE [σt ] = IE [vt ] = v0 etµv ,
 1 2
  2 
Var(σt ) = v0 eµv t 1 − e− 4 ξ t , Var(vt ) = v02 e2µv t eξ t − 1 .

Advantages of the Hull and White model:

Simple to interpret and understand.
Shortfalls of the Hull and White model:
Jumps are not supported.
Often, the volatility is quite small.
The volatility is not mean-reverting.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
169 / 190
 Numerical Methods.

Chapter V
Numerical methods

Literature for this chapter:

Boyle, P., Broadie, M., and Glasserman, P. (1997): Monte Carlo Methods for
security pricing, Journal of Economic Dynamics and Control, Vol. 21, pp.
1267–1321.
Carr, P. and Madan, D. (1999): Option Valuation using the Fast Fourier
Transform, Journal of Computational Finance, Vol. 2(4), pp. 61–73.
Jäckel (2002): Monte Carlo methods in finance.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
170 / 190
 Numerical Methods.

Literature for this chapter (cont.):

Korn, Korn, & Kroisandt (2010): Monte Carlo methods and models in finance
and insurance.
Longstaff, F. and Schwartz, E. (2001): Valuing American options by simulation:
a simple least-square approach. Review of Financial Studies, Vol. 14(1), pp.
113–147.
Seydel (2009): Tools for computational finance, 4th ed.
Wilmott (2007): Paul Wilmott introduces quantitative finance, 2nd ed.
Wilmott (1997): The mathematics of financial derivatives.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
171 / 190
 Numerical Methods.

Monte Carlo methods

Remark 58 (Motivation: Monte Carlo methods).

Our models are described using the language of probability theory.

The terminal payoff of a derivative is an FT -measurable random variable.
Pricing corresponds to computing risk-neutral expectation values.
It is rarely possible to compute such an expectation value explicitly.
The idea of a MC-simulation is to:
a) Perform several experiments of the underlying security, and to
b) Estimate today’s price of the derivative based on these outcomes.
Hence, we first present simulation routines for the underlying processes.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
172 / 190
 Numerical Methods.

Algorithm 1 (Simulation of a Brownian motion).

This algorithm simulates a trajectory of the Brownian motion W = {Wt }t≥0 on the
equidistant grid 0 = t0 < t1 < · · · < tn = T , where ∆ ≡ ∆tk := tk +1 − tk .
1. Initialize t0 := 0, Wt0 := 0, and ∆ := T /n.
2. For j = 1, . . . , n do
2.1 tj := tj−1 + ∆.
2.2 Draw a N (0, 1)-distributed
√ r.v. Z , independent of the past.
2.3 Set Wtj := Wtj−1 + Z ∆.

Observation: This algorithm only requires N (0, 1)-distributed samples.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
173 / 190
 Numerical Methods.

Algorithm 2 (Euler discretisation: Simulating a general Itô process).

Consider the Itô process X = {Xt }t≥0 , given by the SDE

dXt = µ(Xt , t)dt + σ(Xt , t)dWt , X0 = x0 .

This algorithm simulates a trajectory of Xt on the (equidistant) grid

0 = t0 < t1 < · · · < tn = T ,

where ∆ ≡ ∆tk := tk +1 − tk .
1. Initialize t0 := 0, Xt0 =: x0 , Wt0 := 0, and ∆ := T /n.
2. For j = 0, . . . , n − 1 do
2.1 tj+1 := tj + ∆. √
2.2 Draw Z ∼ N (0, 1) (independent of the past) and set ∆W := Z ∆.
2.3 Set Xtj+1 := Xtj + µ(Xtj , tj )∆ + σ(Xtj , tj )∆W .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
174 / 190
 Numerical Methods.

Remark 59 (Euler discretisation: Simulating a general Itô process).

Observation: The algorithm only requires N (0, 1)-distributed samples and is

simple to implement; independent of the form of µ(Xt , t) and σ(Xt , t).
However, the drawback of the Euler discretisation is a small discretisation bias.
This bias results from assuming (freezing) µ(Xt , t) ≡ µ(Xtk , tk ) and
σ(Xt , t) ≡ σ(Xtk , tk ) over the interval [tk , tk +1 ).
The error of the approximation depends on the step size ∆ and is measured by
h i
ϵ(∆) = IE |XT − XT∆ | ,

where XT is the true solution of the SDE and XT∆ is the approximation obtained
with the Euler method (with the same BM).
For the Euler method, one can show that the error decreases as
√
ϵ(∆) ≤ c · ∆1/2 , i.e. ϵ(∆) ∈ O( ∆).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
175 / 190
 Numerical Methods.

Remark 60 (Unbiased simulation of a GBM).

To avoid the bias from the Euler discretisation, the following

approach allows to simulate a GBM without bias (note how we
exploit the specific structure of the GBM).
Recall the SDE of the GBM, i.e µ(St , t) = bSt and σ(St , t) = σSt

dSt = St (bdt + σdWt ), S0 = s0 > 0.

Using Itô’s formula, we find

1
d log(St ) = (b − σ 2 )dt + σdWt .
2
From this, we find the solution
 1 
St = S0 exp (b − σ 2 )t + σWt .
| 2 {z }
:=Xt
M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance
July
and17,Actuarial
2024 Science
176 / 190
 Numerical Methods.

Algorithm 3 (Unbiased and fast simulation of ST ).

If only the terminal value ST is required (do not use this method to produce a path
of S), it is enough to simulate Z ∼ N (0, 1) and to define

d
 1
√ 
ST = S0 exp b − σ2 T + σ T Z ,
2
d
where = means equal in distribution.

Algorithm 4 (Unbiased simulation of a path of S on a grid).

If a trajectory of S is required (e.g. for the pricing of path dependent options), we

proceed as follows. We consider the exponent of the exact solution, given by

1

Xt := b − σ 2 t + σWt .
2
This process can be simulated (without discretisation bias!) via Xt0 := 0 and

1
√
Xti+1 := Xti + b − σ 2 ∆ + σ ∆Z .
2
Finally, set Sti := S0 exp(Xti ), for i = 0, . . . , n.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
177 / 190
 Numerical Methods.

Remark 61 (Motivation and results: Monte Carlo simulation).

The origin for MC simulation is the computation of integrals as

Z 1
I= f (x)dx,
0

where f (x) is an integrable function with unknown anti-derivative.

Therefore, MC simulation is also known as MC integration.
In stochastic terms, this is interpreted as IE [f (U)], where U ∼ Uni [0, 1].
The definition of the Riemann integral suggests the approximation:
n
X k 1
I≈ f .
n n
k =1

Several related approximation (quadrature) formulas are known. However, the

crucial question is to carefully select the points at which f is evaluated.
⇒ In a MC simulation, these points are chosen at random.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
178 / 190
 Numerical Methods.

Remark 62 (Motivation and results: Monte Carlo simulation - 2).

Let U1 , . . . , Un be i.i.d. with Uk ∼ Uni [0, 1]. In a MC-simulation, we use

n
1X

In := f Uk
n
k =1

as an approximation for I.
It is important to notice that In itself is a random variable.
Therefore, we can apply probabilistic results to obtain statements about the
quality of the estimate In .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
179 / 190
 Numerical Methods.

Lemma 26 (Properties of In ).

We obtain (for square integrable f )

IE [In ] = I,
!
Z 1
1
Var (In ) = f 2 (x)dx − I 2 .
n 0

We conclude that:
1. In is an unbiased estimate for I. √
2. The variance (resp. standard deviation) of In decreases as 1/n (resp. 1/ n) in
the number of samples drawn, that is n.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
180 / 190
 Numerical Methods.

Lemma 27 (Properties of In - 2).

The strong law of large numbers can be used to show that In converges to I
with probability one, i.e. for a.e. ω ∈ Ω:

In → I, (n → ∞).

Using the central limit theorem we can even get the approximate distribution of
In . We find
d
 1

In ≈ N I, Var f (U) , U ∼ Uni [0, 1].
n
This result is later used to derive (asymptotic) confidence intervals for I.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
181 / 190
 Numerical Methods.

Remark 63 (Motivation: Pricing options via Monte Carlo simulation).

Recall: The fair price of an option / derivative D on a stock S with maturity T is

the expected discounted payoff (w.r.t. Q) of the option.
In the BS-model, the Q-dynamics of the stock price process are given by

dSt = rSt dt + σSt d W̃t .

For an option D with payoff h({St }t∈[0,T ] ) at time T , we find the price at time
t = 0 as h i
D(S0 , 0) = IEQ e−rT h({St }t∈[0,T ] ) .

This puts us back into the original MC-framework, where an expected value
(i.e. an integral) is to be computed.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
182 / 190
 Numerical Methods.

Algorithm 5 (Motivation: Pricing options via Monte Carlo simulation).

If the expected value

h i
D(S0 , 0) = IEQ e−rT h({St }t∈[0,T ] )

is not known, the following Monte Carlo approach is possible.

1. Simulate the (risk neutral) stock-price process {Ŝt }t∈[0,T ] .
2. For this realization, calculate the discounted payoff of the option
 
D̂ = e−rT h {Ŝt }t∈[0,T ] .

3. Repeat Steps 1. and 2. for i = 1, . . . , n, where n is ”sufficiently large”.

4. Calculate the average payoff of the option, i.e. set

1 
D̄n := D̂1 + . . . + D̂n .
n
This is our estimate of the option price.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
183 / 190
 Numerical Methods.

Remark 64 (Confidence intervals for option prices).

In many cases, D̄n is unbiased, i.e. IE [D̄n ] = D(S0 , 0), and strongly consistent,
i.e. D̄n − D(S0 , 0) → 0 a.s. (n → ∞). Only for path dependent options we
sometimes have to accept a small discretisation bias.
The central limit theorem allows to provide a confidence interval. Let
n
1 X
sn2 := (D̂i − D̄n )2
n−1
i=1

denote the sample variance of the option payoff and let z1−δ denote the 1 − δ
quantile of the standard normal distribution, i.e. Φ(z1−δ ) = 1 − δ. Then
 
sn sn
D̄n − z1−δ/2 √ , D̄n + z1−δ/2 √
n n

is an asymptotically valid confidence interval for D(S0 , 0) to the level 1 − δ.

Using this, we can conclude what ”n sufficiently large” means.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
184 / 190
 Numerical Methods.

Lemma 28 (Simulating X ∼ F (x) ).

All standard software packages provide uniformly distributed random variables

on [0, 1].
We therefore consider as given U ∼ Uni [0, 1].
To simulate an arbitrary random variable X with distribution F (x), it is enough
to know the quantile function F −1 (x) and to compute F −1 (U). Then
d
F −1 (U) = X .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
185 / 190
 Numerical Methods.

Algorithm 6 (Box-Muller algorithm).

Unfortunately, the quantile function Φ−1 (x) of the normal distribution is not
known in closed form, i.e. it has to be approximated numerically.
Here, the algorithm of Box and Müller is more efficient and exact.
Let U1 and U2 be independent and Uni [0, 1]-distributed.
Set
p
Y1 := −2 log(U1 ) cos(2πU2 ),
p
Y2 := −2 log(U1 ) sin(2πU2 ).

Then, Y1 and Y2 are independent N (0, 1)-distributed random variables.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
186 / 190
 Numerical Methods.

Remark 65 (Variance reduction).

Given two unbiased estimates (that can be computed with similar effort) for the
same quantity, it is natural to prefer the one with smaller variance (as for
instance, confidence intervals obtained from this estimate are smaller).
In a Monte Carlo context, a reduction of the standard deviation by the factor 10
corresponds to increasing the number of samples by 100.
The following examples for variance reduction methods in the context of option
pricing will be presented:
Antithetic variates;
Control variates;
Importance sampling.
Techniques that we do not cover are moment matching, stratified and Latin
hypercube sampling, and pseudo-Monte Carlo methods.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
187 / 190
 Numerical Methods.

Definition 63 (Antithetic variates).

The idea is to exploit symmetries in the generated random variables.

For instance, we have for U ∼ Uni [0, 1] and Z ∼ N (0, 1):
d d
U = 1 − U, Z = −Z .

In our context, we often work with

 1
√ 
ST = S0 exp r − σ2 T + σ T Z ,
2
where Z could be replaced by −Z without changing the distribution of ST .

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
188 / 190
 Numerical Methods.

Remark 66 (Using antithetic variates in derivative pricing).

Consider the problem of estimating the price of a standard call option.

Let Ĉi be the realization of the option price in the i-th Monte Carlo step
obtained from Zi ∼ N (0, 1) via
  1
√  +
Ĉi := e−rT S0 exp r − σ 2 T + σ T Zi − K ,
2

and let Či be the realization of the option price obtained from −Zi .
Now consider the estimate
n
1 X Ĉi + Či
C̄nanti := .
n 2
i=1

A heuristic argument for this modification is that antithetic pairs are more
regularly distributed, as each realization has its antithetic analogon. Note that
the sample mean of antithetic pairs Z and −Z is always zero.

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
189 / 190
 Numerical Methods.

Lemma 29 (Properties of C̄nanti ).

A first observation is that C̄nanti is unbiased if C̄n , the original estimate, is.
The next observation is that

Var (C̄nanti ) ≤ Var (C̄n ).

The computation of C̄nanti involves twice as many option values as the

computation of C̄n . Therefore, to be more efficient, we need

2Var (C̄nanti ) ≤ Var (C̄n ).

This relation holds for our call example (and for most standard options).

M. Escobar-Anel (Western University) FM9590 Stochastic Processes with Applications in Finance

July
and17,Actuarial
2024 Science
190 / 190