1 Probability theory

1.1 Basics
Consider a finite sample space Ω

Ω = {ω1 , ω2 , . . . , ωM }, M < ∞.

Define a probability measure P on Ω such that

P ({ωi }) = P (ωi ) = pi > 0, i = 1, . . . , M
M
X
pi = 1.
i=1

For every subset A of Ω, A ⊆ Ω, we have that

X
P (A) = P (ωi ).
ωi ∈A

A random variable X on Ω is a mapping

X : Ω −→ R.

The expectation of X is defined as

M
X
E[X] = X(ωi )P (ωi ).
i=1

1.2 Sigma-algebras and information

It is important to know which information is available to investors. This is formalized using
σ-algebras and filtrations.

Definition 1 A collection F of subsets of Ω is called a σ-algebra (or σ-field) if the following

hold.
1. ∅ ∈ F.
2. If A ∈ F then Ac ∈ F.
∞
[
3. If An ∈ F, n = 1, 2, . . . then ∈F
n=1

Remark 1 When working on a finite sample space Ω condition 3 will reduce to

3.’ If A ∈ F and B ∈ F then A ∪ B ∈ F.

Example 1 The following are examples of σ-algebras.

1. F = 2Ω = {A|A ⊆ Ω}, the power set of Ω.
2. F = {∅, Ω}, the trivial σ-algebra.
3. F = {∅, A, Ac , Ω}.
✷

1
Definition 2 A set P = {A1 , . . . , An } of nonempty subsets of the sample space Ω is called a
(finite) partition of Ω if
n
[
1. Ai = Ω,
i=1

2. Ai ∩ Aj = ∅ for i 6= j.
The σ-algebra consisting of all possible unions of the Ai :s (including the empty set) is called
the σ-algebra generated by P and is denoted by σ(P).

Remark 2 On a finite sample space every σ-algebra is generated by a partition.

When making decisions investors may only use the information available to them. This is
formalized by measurability requirements.

Definition 3 A function X : Ω −→ {x1 , . . . , xK } is F-measurable if

X −1 (xi ) = {ω ∈ Ω|X(w) = xi } ∈ F for all xi

If X is F-measurable we write X ∈ F.

Remark 3 Let F = σ(P). Then a function X : Ω −→ R is F-measurable if and only if X

is constant on each set Ai , i = 1, . . . , n.

This captures the idea that based on the available information we should be able to determine
the value of X.
Measurability is preserved under a lot of operations which is the content of the next propo-
sition.

Proposition 1 Assume that X and Y are F-measurable. Then the following hold:
1. For all real numbers α and β the functions

αX + βY, X ·Y

are F-measurable.

2. If Y (ω) 6= 0 for all ω, then

X
Y
is F-measurable.

3. If {Xn }∞
n=1 is a (countable) sequence of measurable functions, then the functions

sup Xn , inf Xn , lim sup Xn , lim inf Xn ,

n n n n

are F-measurable.

Definition 4 Let X be a function X : Ω −→ R. Then F = σ(X) is the smallest σ-algebra

such that X is F-measurable.
If X1 , . . . , Xn are functions such that Xi : Ω −→ R, then G = σ(X1 , . . . , Xn ) is the smallest
σ-algebra such that X1 , . . . , Xn are G-measurable.

2
The next proposition formalizes the idea that if Z is measurable with respect to a certain
σ-algebra, then “the value of Z is completely determined by the information in the σ-algebra”.

Proposition 2 Let X1 , . . . , Xn be mappings such that Xi : Ω −→ R. Assume that the map-

ping Z : Ω −→ R is σ(X1 , . . . , Xn )-measurable. Then there exists a function f : Rn −→ R
such that
Z(ω) = f (X1 (ω), . . . , Xn (ω)).

We also need to know what is meant by independence. Recall that two events A and B on a
probability space (Ω, F, P ) are independent if

P (A ∩ B) = P (A) · P (B).

For σ-algebras and random variables on (Ω, F, P ) we have the following definition.

Definition 5 The σ-algebras F1 , . . . , Fn are independent if

n n
!
\ Y
P Ai = P (Ai ) whenever Ai ∈ Fi , i = 1 . . . , n.
i=1 i=1

Random variables X1 , . . . , Xn are independent if σ(X1 ), . . . , σ(Xn ) are independent.

1.3 Stochastic processes and filtrations

Let N = {0, 1, 2, 3, . . .}.

Definition 6 A stochastic process {Sn }∞

n=0 on the probability space (Ω, F, P ) is a mapping

S : N × Ω −→ R

such that for each n ∈ N

Sn (·) : Ω −→ R
is F-measurable.

Note that Sn (ω) = S(n, ω). We have that for a fixed n

ω −→ S(n, ω)

is a random variable. For a fixed ω

n −→ S(n, ω)

is a deterministic function of time, called the realization or sample path of S for the outcome
ω.

Remark 4 In this course we will mostly be looking at a fixed time horizon so the process
will only live up until time T , that is we will be looking at processes {Sn }Tn=0 .

3
A stochastic process generates information and as before this is formalized in terms of σ-
algebras, only now there will be a time dimension as well.

Definition 7 Let {Sn }∞ n=0 be random process on (Ω, F, P ). The σ-algebra generated by S
over [0, t] is defined by
FtS = σ{Sn ; n ≤ t}.

We interpret FtS as the information generated by observing S over the time interval [0, t].
More generally information developing over time is formalized by filtrations. They are families
of increasing σ-algebras.

Definition 8 A filtration F = {Fn }n≥0 on (Ω, F, P ) is an indexed family of σ-algebras on

Ω such that
1. Fn ⊆ F, n ≥ 0,

2. if m ≤ n then Fm ⊆ Fn .

Remark 5 As stated before, we will mostly be looking at a fixed time horizon in this
course so the filtration will only live up until time T , that is we will be looking at filtrations
F = {Fn }Tn=0 .

For stochastic process the following measurability conditions are relevant.

Definition 9 Given a filtration F and a random process S on (Ω, F, P ) we say that S is

adapted to F if
Sn ∈ Fn for all n ≥ 0,
and S is predictable with respect to F if

Sn ∈ Fn−1 for all n ≥ 1.

1.4 Conditional expectation

Let X be a random variable on (Ω, F, P ) and G a σ-algebra such that G ⊆ F. In this section
we aim to define the expectation of X given the information in G, or conditional on G, E[X|G].
We will do this in three steps.
1. First we will define the expectation of X given a set B ∈ F, such that P (B) 6= 0, i.e.
E[X|B]. Recall that
M
X X
E[X] = X(ωi )P (ωi ) = X(ω)P (ω)
i=1 ω∈Ω

Now it would seem natural (?) to use the normalized probabilities

P (ωi )
on B.
P (B)
We thus define
X P (ωi ) 1 X
E[X|B] = X(ωi ) = X(ω)P (ω)
ωi ∈B
P (B) P (B) ω∈B

4
 Example 2 Consider the finite sample space Ω = {ω1 , ω2 , ω3 } endowed with the power
σ-algebra F = 2Ω , and a probability measure P such that P (ωi ) = 1/3, i = 1, 2, 3.
Furthermore let B1 = {ω1 , ω2 }, B2 = {ω3 }, P = {B1 , B2 }, and G = σ(P). Finally, let
(
1, if ω = ω1
X(ω) = I{ω1 } (ω) =
0, otherwise.

Then we have that

X 1 1 1 1
E[X] = X(ω)P (ω) = 1 · +0· +0· =
ω∈Ω
3 3 3 3

and that
1 1 1 1 1
X  
E[X|B1 ] = X(ω)P (ω) = 1· +0· = ,
P (B1 ) ω∈B 1/3 + 1/3 3 3 2
1

whereas
1 X 1 1
E[X|B2 ] = X(ω)P (ω) = · 0 · = 0.
P (B2 ) ω∈B 1/3 3
2

2. Next we will define the expectation of X conditional on a partition P of Ω. Suppose

that P = {B1 , . . . , BK } and that P (Bi ) 6= 0, i = 1, . . . , K. Note that for any random
variable Y measurable with respect to σ(P) we have that if ωi ∈ Bj

Y (ωi ) = E[Y |Bj ]

since Y is constant on each Bi . This means that

K
X
Y (ω) = E[Y |Bi ]IBi (ω),
i=1

where IBi denotes the indicator function of Bi , i.e.

(
1, if ω ∈ Bi
IBi (ω) =
0, otherwise.

We now define
K
X
E[X|P](ω) = E[X|Bi ]IBi (ω).
i=1

Note that this means that E[X|P] is a random variable Z such that

1. Z ∈ σ(P) and that

2. for all B ∈ σ(P) we have that
X X
Z(ω)P (ω) = X(ω)P (ω).
ω∈B ω∈B

5
 Example 3 Continuing on Example 2 we can compute
2
X 1
E[X|P] = E[X|Bi ]IBi (ω) = · IB1 (ω) + 0 · IB2 (ω).
i=1
2
✷

3. Now we are ready to give the general definition of E[X|G].

Definition 10 Consider a random variable X on (Ω, F, P ) and a σ-algebra G such

that G ⊆ F. The conditional expectation of X given G denoted E[X|G] is any random
variable Z such that

1. Z ∈ G, and
2. for all A ∈ G we have that
X X
Z(ω)P (ω) = X(ω)P (ω).
ω∈A ω∈A

The proposition below states some properties of the conditional expectation.

Proposition 3 The conditional expectation has the following properties. Suppose that X and
Y are random variables on (Ω, F, P ) and that α, β ∈ R. Let G be a σ-algebra such that G ⊆ F.
Then the following hold.
1. Linearity.
E[αX + βY |G] = αE[X|G] + βE[Y |G].

2. Monotonicity. If X ≤ Y then
E[X|G] ≤ E[Y |G].
3.

E[E[X|G]] = E[X].

4. If H is σ-algebra such that H ⊆ G ⊆ F then

(i) E[E[X|H]|G] = E[X|H],

(ii) E[E[X|G]|H] = E[X|H].

Thus “the smallest σ-algebra always wins”.

5. Jensen’s inequality. If ϕ is a convex function, then

ϕ(E[X|G]) ≤ E[ϕ(X)|G].

6. If X is independent of G then
E[X|G] = E[X].

7. Taking out what is known. If X ∈ G then

E[XY |G] = X · E[Y |G].