Advanced Derivatives Modelling:
Stochastic calculus toolkit
January 23, 2015
1 Expectation and conditional expectation operators
1.1 Properties of the expectation operator
Proposition 1.1 (Linearity). If α and β are real constants and X and Y are integrable
random variables, then
(1.1) E [αX + βY ] = αE [X] + βE [Y ] .
Proposition 1.2 (Jensen’s inequality). If φ is a convex, real-valued function defined on
R, and if X is an integrable random variable, then
(1.2) φ (E [X]) ≤ E [φ (X)] .
1.2 Properties of conditional expectations
We use the notion of filtration of information, where the space of events determined at
time t is denoted Ft . We thus have an increasing filtration of information indexed by time:
(1.3) F0 ⊂ Ft1 ⊂ · · · ⊂ . . .
The linearity and Jensen’s inequality properties are inherited by expectations conditioned
on a sigma-algebra. We recall three additional properties that are useful. First, condition-
ing on no information we obviously recover the unconditional expectation:
(1.4) E [X|F0 ] = E [X] .
1
Second, if we estimate a random variable based on an information set and then try to
estimate the estimate itself based on a smaller information set we get the same estimate
we would have got if we directly estimated the random variable based on the smaller set:
Proposition 1.3 (Tower Law). If X is integrable, then for any s < t
(1.5) E [E [X|Ft ] |Fs ] = E [X|Fs ]
Finally, if we condition on information that is independent of the random variable then we
get the same value as with the unconditional expectation:
Proposition 1.4 (Independence). If X is integrable and independent of F, then
(1.6) E [X|F] = E [X]
2 Martingale
Definition 2.1 (Martingale). A stochastic process (Xt )(t≥0) is a martingale with respect
to a measure P and a filtration (Ft )(t≥0) if
(2.1) EP [Xt |Fs ] = Xs , ∀t ≥ s.
3 Brownian motion
Definition 3.1 (Brownian motion). Let a stochastic process (Bt )(t≥0) be a collection of
random variables indexed by time t satisfying B0 = 0. Then the process (Bt )(t≥0) is a
Brownian motion if for all sequences of times t0 = 0 ≤ t1 ≤ · · · ≤ tm the increments
Bt1 − B0 , Bt2 − Bt1 , . . . , Btm − Btm−1
are independent and normally distributed with moments:
(3.1) E Bti+1 − Bti = 0,
(3.2) Var Bti+1 − Bti = ti+1 − ti
Using the independence of the increments and the linearity of the expectation operator,
we can deduce
(3.3) E [Bs Bt ] = s ∧ t.
A Brownian motion is a martingale with respect to its filtration (Ft )(t≥0) , i.e. for any
0 ≤ s ≤ t:
(3.4) E [Bt |Fs ] = Bs .
2
4 Ito calculus
4.1 Ito’s integral
Ito’s integral is defined for integrands that are adapted stochastic processes. It defines the
meaning of stochastic integrals with respect to a the measure of the Brownian motion:
Z t
(4.1) I(t) = f (u)dBu
0
"Obvious" properties of Ito integrals are path continuity (as a function of t) and linearity:
Proposition 4.1 (Linearity). For any constants α and β and adapted processes f (t) and
g(t)
Z t Z t Z t
(4.2) α f (u)dBu + β g(u)dBu = (αf (u) + βg(u)) dBu .
0 0 0
We recall three more important properties. First, Ito’s integral is a martingale
Proposition 4.2 (Martingale). For any adapted process integrable process f (u) and hori-
zon time t, the stochastic integral Z t
f (u)dBu
0
is a martingale.
By implication we have
Z t
(4.3) E f (u)dBu
= 0
0
Z t Z s
(4.4) E f (u)dBu |Fs = f (u)dBu
0 0
In addition, the quadratic variation of the stochastic integral is given by
Proposition 4.3 (Quadratic variation of Ito’s integral). The quadratic variation of the
stochastic integral up to a time t is given by
Z t Z t Z t
(4.5) f (u)dBu , f (u)dBu = f (u)2 du .
0 0 0
Finally, the variance of the stochastic integral is given by the expectation of the quadratic
variation:
3
Proposition 4.4 (Ito isometry). The Ito integral satisfies
"Z 2 #
t Z t
(4.6) E f (u)dBu = f (u)2 du
0 0
"Z 2 #
t Z s
(4.7) E f (u)dBu |Fs = f (u)2 du
0 0
Furthermore, in the case of deterministic integrands, i.e. functions of time only, the stochas-
tic integral is normally distributed.
4.2 Ito processes
Definition 4.1 (Ito process). Let (Bt )(t≥0) be a Brownian motion process with associ-
ated filtration (Ft )(t≥0) . If µ(u) and σ(u) are adapted stochastic processes satisfying the
integrability conditions:
Z t
(4.8) |µ(u)|du < ∞, ∀t > 0
0
Z t
2
(4.9) E σ(u) du < ∞, ∀t > 0
0
then a stochastic process (Xt )(t≥0) starting form a know value X0 and of the form
Z t Z t
(4.10) Xt = X0 + µ(u)du + σ(u)dBu
0 0
is called an Ito process.
4.3 Ito’s lemma
Proposition 4.5 (Ito’s formula). Let (Xt )(t≥0) denote an Ito process. If f (t, x) is function
with continuous partial derivatives ft (t, x), fx (t, x) and fxx (t, x), then
Z T Z T
1 T
Z
f (T, XT ) − f (0, X0 ) = ft (t, Xt )dt + fx (t, Xt )dXt + fxx (t, Xt )d[X, X]t
0 0 2 0
Z T Z T
1 2
(4.11) = ft (t, Xt ) + fx (t, Xt )µ(t) + fxx (t, Xt )σ(t) dt + fx (t, Xt )dBt
0 2 0
We often use the differential notation of Ito’s formula:
1
df (t, Xt ) = ft (t, Xt )dt + fx (t, Xt )dXt + fxx (t, Xt )dXt dXt
2
1 2
(4.12) = ft (t, Xt ) + fx (t, Xt )µ(t) + fxx (t, Xt )σ(t) dt + fx (t, Xt )dBt
2
4
5 Girsanov theorem
Definition 5.1 (Radon-Nikodym derivative process). Let P be a probability measure and
(Ft )(t≥0) a filtration. If Z is an almost surely positive random variable satisfying E [Z] = 1
then the martingale process
(5.1) Zt = E [Z|Ft ] , ∀0 ≤ t ≤ T
define a Radon-Nikodym derivative process.
The Radon-Nikodym derivative process permits changes of probability measures techniques
due to two important properties that we recall below.
Proposition 5.1 (Probability measure induced by Radon-Nykodim derivative). Let (Zt )(t≥0)
a Radon-Nykodim process. We then have a new probability measure P̃ defined by
Z
(5.2) P̃(A) = ZdP ∀ ∈ F
A
If Y is an Ft −measurable random variable, then for any t ≤ T
(5.3) EP̃ [Y ] = EP [Y Zt ]
1
(5.4) EP̃ [Y |Fs ] = EP [Y Zt |Fs ] .
Zs
Theorem 5.1 (Girsanov’s theorem). Let (Bt )(t≥0) be a Brownian motion with respect to
a probability measure P and (Ft )(t≥0) its associated filtration. If (θt )(t≥0) is an adapted
process satisfying the condition
Z T
(5.5) E θu2 Zu2 du < ∞,
0
then the exponential martingale process (Zt )(t≥0) given by
Z t
1 t 2
Z
(5.6) Zt = exp − θu dBu − θ du
0 2 0 u
is a Radon-Nykodym process. Under the probability measure P̃ obtained by the change of
measure
P̃
(5.7) = Zt ,
P
the process (B̃t )(t≥0) defined by
Z t
(5.8) B̃t = Bt + θu du,
0
is a Brownian motion under the probability measure P̃.
