Stochastic Finance and Risk Modelling

—
1 - From probability to stochastic process

Gaoyue Guo

Gaoyue Guo ST7 - 71 CentraleSupélec 1

Outline

Probability

Independence

Conditional expectation

Equivalent probabilities

Gaoyue Guo ST7 - 71 CentraleSupélec 2

 Probability

Outline

Probability

Independence

Conditional expectation

Equivalent probabilities

Gaoyue Guo ST7 - 71 CentraleSupélec 3

 Probability

Measurability
Definition (measurable space)
A collection of subsets F is said to be a σ−algebra on Ω if
I ∅, Ω ∈ F ;
I A ∈ F =⇒ Ac := Ω \ A ∈ F ;
I (An )n≥1 ⊂ F =⇒ ∪n≥1 An ∈ F.
(Ω, F) is called a measurable space.

Definition (measurable function)

For measurable spaces (Ω, F), (Σ, G), the map f : Ω → Σ is said to be
measurable if

f −1 (B) := {ω ∈ Ω : f (ω) ∈ B} ∈ F, ∀B ∈ G.

In particular, f −1 (B) denotes the inverse image of B by f .

Gaoyue Guo ST7 - 71 CentraleSupélec 4
 Probability

Some algebras on Ω
I (Fλ )λ∈Λ is a family of algebras. So it is with ∩λ∈Λ Fλ .
I Let C be a set of some subsets of Ω, define
\
σ(C) := F.
F : algebra, F ⊃C

σ(C) is called the algebra generated by C.

I For (fλ )λ∈Λ with fλ : (Ω, F) → (Σλ , Gλ ), set σ(fλ : λ ∈ Λ) := σ(CΛ ),
where

CΛ := {fλ−1 (B) : B ∈ Gλ , λ ∈ Λ}.

I Let (Ωn , Fn ) be a sequence of measurable spaces. Define the
n≥1
product algebra on the product space Πn≥1 Ωn by

Π⊗


n≥1 Fn := σ {Πn≥1 An : An ∈ Fn } .

Gaoyue Guo ST7 - 71 CentraleSupélec 5

 Probability

Probability
Definition (measure)
A map µ : F → [0, ∞] is called a measure if
I µ(∅) = 0 ;
X
I µ(∪n≥1 An ) = µ(An ) for any (An )n≥1 ⊂ F of disjoint sets.
n≥1
µ is called a probability if µ(Ω) = 1. Denote by P(Ω, F) the set of
probabilities. For µ ∈ P(Ω, F), (Ω, F, µ) is called a probability space.

µ is supported in K ⊂ Ω if µ(K c ) = 0. If Ω is Polish, let supp(µ) be the

smallest closed subset in which µ is supported.

Definition (pushforward measure)

Let µ ∈ P(Ω, F) and f : Ω → Σ be measurable. Set f# µ ∈ P(Σ, G) by
f# µ(B) := µ(f −1 (B)) for all B ∈ G.

Gaoyue Guo ST7 - 71 CentraleSupélec 6

 Probability

Random variable
Definition (random variable)
Let Σ = Rd and G = B(Rd ) := σ({A : A is open in Rd }). Any
measurable function X : Ω → Rd is called a r.v. on (Ω, F).

For every P ∈ P(Ω, F), PX := X# P ∈ P(Rd , B(Rd )) denotes the

law/distribution of X under P. Define supp(X ) := supp(PX ).

Definition (expectation)
Z
Fix P ∈ P(Ω, F). A r.v. X is integrable if |X (ω)|P(dω) < ∞. Denote
Ω
the expectation of X by
Z
E[X ] := X (ω)P(dω).
Ω

Gaoyue Guo ST7 - 71 CentraleSupélec 7

 Probability

Related terminology

Definition (Lp (Ω, F, P))

For each p > 0, denote by Lp (Ω, F, P) the space of r.v.s X s.t. |X |p is
integrable.

Definition (variance/covariance)
Let d = 1. If X , Y ∈ L2 (Ω, F, P), denote by Cov (X , Y ) the covariance
between X and Y
 
Cov (X , Y ) := E (X − E[X ])(Y − E[Y ]) = E[XY ] − E[X ]E[Y ].

Define further the variance of X by Var (X ) := Cov (X , X ).

Gaoyue Guo ST7 - 71 CentraleSupélec 8

 Probability

Theorem of transport

Theorem
Fix P ∈ P(Ω, F) and denote Y := g (X ). Then Y is a r.v. and
Y ∈ L1 (Ω, F, P) iff g ∈ L1 (Rd , B(Rd ), PX ), i.e.
Z
E[|Y |] < ∞ ⇐⇒ |g (x)|PX (dx) < ∞.
Rd

When Y ∈ L1 (Ω, F, P), one has

Z
E[Y ] = E[g (X )] = g (x)PX (dx).
Rd

Gaoyue Guo ST7 - 71 CentraleSupélec 9

 Probability

List of common distributions

1. X (resp. PX ) has (resp. is) binomial distribution B(n, p) if
supp(X ) = {0, . . . , n} and
n!
P(X = k) = PX ({k}) = p k (1 − p)n−k , k = 0, . . . , n;
k!(n − k)!
2. X has Poisson distribution P(λ) if supp(X ) = N and
e −λ λk
PX ({k}) = , k = 0, 1, . . . ;
k!
3. X has Gaussian distribution N (m, σ 2 ) if
1 2 2
PX (dx) = √ e −(x−m) /2σ dx;
2πσ
4. X has uniform distribution U(a, b) if supp(X ) = [a, b] and
1[a,b] (x)
PX (dx) = dx.
b−a
Gaoyue Guo ST7 - 71 CentraleSupélec 10
 Independence

Outline

Probability

Independence

Conditional expectation

Equivalent probabilities

Gaoyue Guo ST7 - 71 CentraleSupélec 11

 Independence

Independence

Fix (Ω, F, P).

Definition (independence of algebras)

The algebras G1 , . . . , Gn ⊂ F are independent if

P(∩ni=1 Ai ) = Πni=1 P(Ai ), ∀Ai ∈ Gi , 1 ≤ i ≤ n.

A family of algebras (Gλ )λ∈Λ is said to be independent if Gλ1 , . . . , Gλn are

independent for all λ1 , . . . , λn ∈ Λ and n ≥ 1.

Definition (independence of r.v.s)

A family of r.v.s (Xλ )λ∈Λ is independent if (σ(Xλ ))λ∈Λ is independent.

Gaoyue Guo ST7 - 71 CentraleSupélec 12

 Independence

More theorems
Let (Xn )n≥1 be a sequence of i.i.d. r.v.s on (Ω, F, P).

Theorem (Law of large numbers)

Assume X1 ∈ L1 (Ω, F, P) and set m := E[X1 ]. Then
 
X1 (ω) + · · · + Xn (ω)
P ω ∈ Ω : lim =m = 1.
n→∞ n

Namely, (X1 + · · · + Xn )/n converges a.s. to m.

Theorem (Central limit theorem)

Assume further X1 ∈ L2 (Ω, F, P) and set σ 2 := Var (X1 ). Then
Pn
k=1√(Xk − m) L
−→ N (0, 1).
nσ

Gaoyue Guo ST7 - 71 CentraleSupélec 13

 Conditional expectation

Outline

Probability

Independence

Conditional expectation

Equivalent probabilities

Gaoyue Guo ST7 - 71 CentraleSupélec 14

 Conditional expectation

Conditional expectation in L1
Theorem (Existence and uniqueness)
Let X ∈ L1 (Ω, F, P) and G ⊂ F. There is a unique G−measurable r.v.,
denoted by E[X |G] ∈ L1 (Ω, F, P) and called the conditional expectation of
X knowing G, s.t.

E[XZ ] = E[E[X |G]Z ], ∀bounded and G-mesurable Z .

We may replace Z by 1A for A ∈ G. Further, if G is generated by (Yλ )λ∈Λ ,

we write E[X |G] = E[X |Yλ , λ ∈ Λ].

Proposition
For X ∈ L1 (Ω, F, P), one has :
(a) (tower property) E[E[X |G]|H] = E[X |H] for all H ⊂ G ⊂ F ;
(b) (linearity) X 7→ E[X |G] is a linear map ;
(c) (positivity) If X ≥ 0 a.s., then E[X |G] ≥ 0 a.s. ;
(d) (representation)
Gaoyue Guo E[X |Y ] = g (Y
ST7) - for
71 some measurable function g.
CentraleSupélec 15
 Conditional expectation

Properties of conditional expectation

Proposition
Let X ∈ L1 (Ω, F, P) and G ⊂ F.
(a) E[X |G] = X iff X is G-measurable.
(b) If X is independent of G, then E[X |G] = E[X ].
(c) Let Y is G-measurable s.t. XY ∈ L1 (Ω, F, P). Then
E[XY |G] = Y E[X |G].
(d) If ϕ : R → R is convex, then ϕ (E[X |G]) ≤ E[ϕ(X )|G].

Theorem
Let (Xn )n≥1 ⊂ L1 (Ω, F, P) converge almost surely to X . Then
lim E[Xn |G] = E[X |G] almost surely if one of the following conditions
n→∞
holds
(i) |Xn | ≤ Z ∈ L1 (Ω, F, P) for all n ≥ 1 ;
(ii) Xn ≥ 0 increases to X almost surely.
Gaoyue Guo ST7 - 71 CentraleSupélec 16
 Conditional expectation

A special case : X ∈ L2 (Ω, F, P)

Note that L2 (Ω, F, P) is a Hilbert space endowed with the scalar product
hX , Y i := E[XY ].

Proposition
Let G ⊂ F.
(a) L2 (Ω, G, P) is a closed subspace of L2 (Ω, F, P).
(b) E[X |G] is identified as the projection of X in L2 (Ω, G, P), i.e.

hX − E[X |G], Z i = 0, ∀Z ∈ L2 (Ω, G, P).

(c) If X , Y1 , . . . , Yn form a Gaussian vector, then with

G = σ(Y1 , . . . , Yn ), one has
n
X n
X
E[X |G] = a0 + ai Yi s.t. X − ai Yi is independent of G.
i=1 i=1

Gaoyue Guo ST7 - 71 CentraleSupélec 17

 Equivalent probabilities

Outline

Probability

Independence

Conditional expectation

Equivalent probabilities

Gaoyue Guo ST7 - 71 CentraleSupélec 18

 Equivalent probabilities

Change of measure

Let Z ∈ L1 (Ω, F, P) be positive a.s. and E[Z ] = 1. Define Q ∈ P(Ω, F) as

follows : For any A ∈ F, set
Z
Q(A) := Z (ω)dP(ω).
A

Then Q ∈ P(Ω, F).

Z is called the derivative of Radon-Nikodym of Q with respect to P,
dQ
denoted by Z = .
dP

Gaoyue Guo ST7 - 71 CentraleSupélec 19

 Equivalent probabilities

Radon-Nikodym’s derivative

Let (Ω, F, P) be a probability space endowed with filtration (Ft )t∈I .

Let Z ≥ 0 be a r.v. s.t. E[Z ] = 1. Define Q as follows :
Z
Q(A) = Z (ω) dP(ω), ∀A ∈ F.
A

Then Q defines a probability measure on Ω and Z is called the

dQ
Radon-Nikodym derivative of Q w.r.t. P, denoted by Z = .
dP
Proposition
Let X ≥ 0 be a r.v.. It holds EQ [X ] = EP [ZX ].

Gaoyue Guo ST7 - 71 CentraleSupélec 20

 Equivalent probabilities

Equivalent probabilities

Let P, Q be two probabilities on (Ω, F).

Definition (Absolutely continuous probabilities)

Q is absolument continue w.r.t. P, denoted by Q ≤ P) if for every A ∈ F
P(A) = 0 =⇒ Q(A) = 0.

Definition
P and Q are equivalent if Q ≤ P and P ≤ Q.

I If Q is given previously, then Q is absolutely continuous w.r.t. P.

I Assume further P(Z > 0) = 1, then P and Q are equivalent.

Gaoyue Guo ST7 - 71 CentraleSupélec 21