”Crash course on Calculus”: from

Classical to Ito
based on
Stochastic models, Nottingham University
and
Financial Mathematics II, Leicester University

Ito Lemma

Classical. Consider a ”smooth” function x(t) (differentiable or twice differentiable, that is what-
ever we need we can assume). And let us derive the formula

(x2 )0 = 2xx0

from the rubbish rules.

Notation/convention Throughout this section, we will fix time/variable t and increment ∆t
and write

∆f = ∆f (t) = f (t + ∆t) − f (t)

df = df (t) = f (t + dt) − f (t).

Start with

x(t + ∆t) = x + ∆x
= x(t) + ∆x(t)

Then use (a + b)2 = a2 + 2ab + b2 to derive

x(t + ∆t)2 = [x + ∆x]2

= x2 + 2x∆x + (∆x)2
= x2 (t) + 2x(t)∆x(t) + (∆x(t))2

The key ingredient in the classical analysis is Classical assumption

(∆x(t))2 = o(∆t)

and so

x(t + ∆t)2 = [x + ∆x]2

= x2 + 2x∆x + (∆x)2
= x2 (t) + 2x(t)∆x(t) + o(∆t)

1
In particular,
∆x2 = [x + ∆x]2 − x2
= 2x∆x + o(∆t)
Proof 1: via derivative from which by SDL
dx2
= (x2 )0
dt
∆x2 (t)
= defn = lim
∆t→0 ∆t
2x(t)∆x(t)
= lim
∆t→0 ∆t
∆x(t)
= 2x(t) lim
∆t→0 ∆t
= defn = 2x(t)x0 (t)
Now, write it again
dx2 dx
= 2x
dt dt
by deleting dt
dx2 = 2xdx
and we (re)derive the chain rule for the example in the differential form
dx2 = 2xdx (∗)
Proof 2 of the differential form (*) without the derivative. Simply replace ∆· by d· and let
o(∆t) = 0, so
∆x2 = 2x∆x + o(∆t) =⇒
dx2 = 2xdx
Ito. Imagine that there exists a function Bt = B(t) such that
[∆B]2 = [∆Bt ]2 = ∆t + o(∆t)
Then, derivative B 0 does NOT make sense since roughly
√
∆Bt ∆t
∼ →∞
∆t ∆t
The argument of Proof 1 will not work, since derivative is not defined everywhere.
However, by applying argument of Proof 2 as before
2
Bt+∆t = [B + ∆B]2
= B 2 + 2B∆B + (∆B)2
as before
= B 2 (t) + 2Bt ∆Bt + ∆t + o(∆t)
with a NEW term

2
and so
∆B 2 = [B + ∆B]2 − B 2
= 2B∆B + ∆t + o(∆t)
resulting in
dB 2 = [B + dB]2 − B 2
= 2BdB + dt
Overall, we have shown that the the strange function B satisfies the chain rule with an extra
(so-called Ito) term
dB 2 = 2BdB + dt

Summarising via the abstract rules

Let us apply the trivial rule
(dt)2 = 0 and so if x is differentiable (dx)2 = (x0 (t)dt)2 = 0
In particular,
dx2 = (x + dt)2 − x2 = x2 + 2xdx + (dx)2 − x2 = 2xdx
However, for the strange VERY chaotic function B,
(dB)2 = dt
and so
dB 2 = (B + dt)2 − B 2 = B 2 + 2BdB + (dB)2 − B 2 = 2BdB

Ito lemma
Ito Lemma. Consider a twice continuously differentiable function f (x, t). Then,
df (t, Bt ) = fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt or
∂f ∂f ∂ 2f
df (t, Bt ) = dB + dt + (1/2) 2 dt
∂B ∂t ∂ B
∂f (t, Bt ) ∂f (t, Bt ) ∂ 2 f (t, Bt )
= dB + dt + (1/2) dt
∂B ∂t ∂ 2B
The two first terms are usual terms in the differential form of the Chain rule. The last one
(1/2)fBB (t, Bt )dt is the Ito term.
Proof. Start with the Taylor formula (B = Bt )
f (t + ∆t, Bt+∆t ) = f (t + ∆t, B + ∆B)
= f (t, B) + fB (t, B)∆B + ft (t, B)∆t
+(1/2)fBB (t, B)(∆B)2
+(1/2)ftt (t, B)(∆t)2 + fBt (t, B)∆B∆t + o([∆t]2 + [∆B]2 )

3
Now,
√
∆B∆t = O( ∆t)∆t = o(∆t),
|∆B|2 = ∆t, ∆t2 = o(∆t)

and we derive

f (t + ∆t, Bt+∆t ) = f (t, B) + fB (t, B)∆B + ft (t, B)∆t

+(1/2)fBB (t, B)∆t + o(∆t)

implying

∆f (t, Bt ) = f (t + ∆t, Bt+∆t ) − f (t, B)

= fB (t, B)∆B + ft (t, B)∆t + (1/2)fBB (t, B)∆t + o(∆t)

and so

df (t, Bt ) = fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt

∂f (t, Bt ) ∂f (t, Bt ) ∂ 2 f (t, Bt )
= dB + dt + (1/2) dt
∂B ∂t ∂ 2B
Again, via the abstract rules
The semi-group commutative operations are

dtdBt = dBt dt = 0, (dt)2 = 0, (dBt )2 = dt, 1 · dt = dt · 1 = dt, 1 · dBt = dBt · 1 = dB

Hence, from the formal Taylor expansion

f (t + dt, Bt+dt ) = f (t + dt, B + dB)

= f (t, Bt ) + fB (t, Bt )dB + ft (t, Bt )dt
+(1/2)fBB (t, Bt )(dB)2
+(1/2)ftt (t, Bt )(dt)2
by the rules
= f (t, Bt ) + fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt

and so

df (t, Bt ) = f (t + dt, Bt+dt ) − f (t, Bt )

= f (t, Bt ) + fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt − f (t, Bt )
= fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt

Example. Let f (x, t) = exp(ax + bt) and let X = X(t) = f (t, Bt ). Then,

fx (x, t) = ∂ exp(ax + bt)/∂x = a exp(ax + bt) = af,

ft (x, t) = ∂ exp(ax + bt)/∂t = b exp(ax + bt) = bf,
fx,x (x, t) = ∂ 2 exp(ax + bt)/∂x2 = a2 exp(ax + bt) = a2 f .

4
So,

fB (t, B) = af = aX,
ft (t, B) = bf = bX,
fB,B (t, B) = a2 f = a2 X .

and thus

dX = df (t, Bt ) = fB (t, Bt )dB + ft (t, Bt )dt + (1/2)fBB (t, Bt )dt

=⇒ dX = aXdB + (b + a2 /2)Xdt

i.e. X satisfies the so-called stochastic differential equation (SDE).

5
Ito Process and Ito Martingale

Let {Bt ; t ≥ 0} be a standard Brownian motion.

Ito process

Defn 1 We say that {St ; t ≥ 0} is an Ito process if its Ito differential is written in the form

dSt = µt dt + σt dBt

where µt , σt are adapted processes, i.e. measurable with respect to the Brownian motion filtration,
Ft , or roughly aT is adapted if at = H(Bs , s ≤ t); µt is referred to as a drift, σt - volatility.

Main properties of Ito processes:

(i) at is adapted implies that f (t, at ) is also adapted;

(ii) Ito process St is adapted;
(iii) (dSt )2 = σt2 dt;
(iii) f (t, St ) is again an Ito process with the differential defined via the Ito formula

df (t, St ) = fS0 t dSt + ft0 dt + 21 fS00t ,St (dSt )2

= fS0 t [µt dt + σt rdBt ] + ft0 dt + 21 fS00t ,St σt2 dt

(iv)Ito process is identified by S0 , µt and σt . (i.e. two processes with same characteristics are
equal almost surely).
(v)(Defn 2) Stochastic integral representation. The equivalent form for the Ito process is
Z t Z t Z t
St = S0 + dSu = S0 + µu du + σu dBu .
0 0 0

(vi) Magic rules: dt2 = 0, dtdBt = 0 and (dBt )2 = dt.

(vii)!!! Bt is an Ito process;
(viii) f (t, Bt ) is an Ito process.

Ito martingales or Martingales among Ito processes

Lemma/defn 3 An Ito process {Mt ; t ≥ 0} is a martingale with respect to the Brownian motion
filtration Ft if and only if

dMt = σdBt

(i.e. the martingale has no drift µt ≡ 0).

6
Main properties of Ito martingales:

(i) E[MT |Ft ] = Mt (definition of martingales);

(ii) E[Mt ] = M0 ;
(iii) Stochastic integral representation
Z t Z t
Mt = M0 + dMu = M0 + σu dBu .
0 0

Ito integral, basically same as the Ito martingale

Main properties of the Ito integral (basically same as for martingales)

(i) For T > t


Z T Z t
E σu dBu |Ft ] = σu dBu
0 0

(ii)E[Ito] = 0 or


Z T
E σu dBu ] = 0
0

”Use/apply Ito calculus” methodology:

To find an integral E[f (Bt )] use Ito formula to reduce it to simpler calculations since the E[Ito] = 0
and that the second moment of Ito integral can be found via Ito isometry (to be applied in fm2).
To show that the Ito process St (e.g. f (t, Bt )) is a martingale check that there is no the drift, i.e.
µt ≡ 0.

7
SDE=Stochastic differential equations.
Now the Ito process is defined via the equation (SDE):

dSt = µ(t, St )dt + σ(t, St )dBt

We assume the following

Fact/defn 4. Suppose that

dXt = µ(t, Xt )dt + σ(t, Xt )dBt

and X0 = S0 , then the strong solutions are same, i.e. Xt = St almost surely.

We mostly consider solutions of the form St = f (t, Bt ).

Examples

Example 1. Show that Mt = Bt2 − t is a martingale via the Ito calculus.

Solution. By the Ito formula applied to the Ito process St = Bt and f (t, Bt ) = Bt2 − t, i.e.
f (t, x) = x2 − t

dMt = df (t, Bt ) = fB0 t dBt + ft0 dt + 12 fB00 t ,Bt dt

= 2Bt dBt + [−t]dt + 21 2dt = 2Bt dBt

hence is the martingale.

Example 2. Compute E[Bt2 ] via the Ito calculus.

Solution. By the Ito formula applied to f (t, Bt ) = Bt2 , i.e. f (t, x) = x2 (so ft0 = 0)

dBt2 = df (t, Bt ) = fB0 t dSt + 12 fB00 t ,Bt dt = 2Bt dBt + dt

Z t Z t
2 2
Bt = B0 + 2Bu du + du
0 0
Z t Z t
= 2Bu du + du since B0 = 0
0 0

Hence
Z t Z t
E[Bt2 ]

= E 2Bu du] + E du
0 0
= 0 + t since E[Ito] = 0

8
Ito integral

Defn. Assume that the process {X(t) ; t ≥ 0} is defined via the differential/sde type representa-
tion

dX(t) = a(t, Bt )dBt + µ(t, Bt )dt

Then (which is the equivalent representation)

Z t Z t
X(t) = X(t0 ) + a(s, Bs )dBs + µ(s, Bs )ds
t0 t0

In particular,
Z t Z t
X(t) = X(0) + a(s, Bs )dBs + µ(s, Bs )ds
0 0

The first integral is called Ito integral.

Properties of the Ito integral.
(i) (E[Ito] = 0) or
Z t 
E a(s, Bs )dBs = 0
t0

Roughly in all our examples,

Z t 
E whatever dBs = 0
t0

(ii) (Fubini)
Z t  Z t
E a(s, Bs )ds = E[a(s, Bs )]ds
t0 t0

(ii) (Ito isometry)

Z t Z t  Z t 
E a(s, Bs )dBs c(s, Bs )dBs = E a(s, Bs )c(s, Bs )ds
t0 t0 t0
with
Z t Fubini
= E[a(s, Bs )c(s, Bs )]ds
t0

Example. Find E[Bt3 ] by applying the Ito lemma.

Solution. From the Ito lemma

df (t, Bt ) = fB (t, Bt )dB + [ft (t, Bt ) + (1/2)fBB (t, Bt )]dt

9
applied to f (x, t) = x3 . So (with x = B = Bt )

fB (t, B) = fx (x, t) = 3x2 = 3B 2 = 3B 2 (t) ,

ft (t, B) = ft (x, t) = 0 ,
00
fBB (t, B) = fxx (x, t) = 6x = 6B = 6Bt .

And so,

dBt3 = 3B 2 (t)dBt + 6Bt dt

Thus,
Z t Z t
Bt3 3
= B (0) + 3B (s)dBs + 2
6Bs ds
0 0
Z t Z t
Bt3 = 2
3B (s)dBs + 6Bs ds
0 0

because B(0) = 0.
By taking expectation,
Z t  Z t 
E[Bt3 ] = E 2
3B (s)dBs + E 6Bs ds
0 0
Z t 
= E 6Bs ds
0
since E[Ito] = 0
Z t
= E[6Bs ]ds
0
Fubini
Z t
= 6 E[Bs ]ds
0

Now, to find E[Bs ] we do the whole proof again.

From the Ito lemma

df (t, Bt ) = fB (t, Bt )dB + [ft (t, Bt ) + (1/2)fBB (t, Bt )]dt

applied to f (x, t) = x, with x = B = Bt so

fB (t, B) = fx (x, t) = 1 ,
ft (t, B) = ft (x, t) = 0 ,
00
fBB (t, B) = fxx (x, t) = 0 .

And so we deduce (as expected)

dBt = dBt

10
Thus,
Z t
Bt = B(0) + 1dBs
0
Z t
Bt = dBs
0

because B(0) = 0.
By taking expectation,
Z t 
E[Bt ] = E dBs
0
= 0
since E[Ito] = 0.
Going back,
Z t
E[Bt3 ] = 6 E[Bs ]ds
0
Z t
= 6 0ds
0
= 0
Calculating the integral via modified Fundamental theorem of calculus. Assume that
u(s) is a good differentiable function. The Newton (Newton-Leibniz) formula (refered to as the
Fundamental theorem of calculus) states
Z b
u0 (x)dx = u(b) − u(a)
a

Let us write it in a differential form, i.e. using

du(x) = u0 (x)dx
that is
Z b
du(x) = u(b) − u(a)
a

This form is actually MUCH more powerful and holds basically regardless WHAT is the differential.
Classical calculus. Take u(x) = g(x)2 where again g differentiable and so by (either chain or
product rule)
[g(x)2 ]0 = 2g(x)[g(x)]0 or
dg(x)2 = 2g(x)dg(x) (more interesting!)
Then, we can write
Z b Z b
2g(x)dg(x) = dg 2 (x)
a a
= g (b) − g 2 (a)
2

11
In particular, if a = 0, b = t, g(0) = 0,
Z t
2g(x)dg(x) = g 2 (t) − g 2 (0) = g 2 (t)
0

Ito calculus. Now, take u(x) = B(x)2 . Then, by the Ito lemma

dBt2 = 2Bt dBt + dt

and so,
Z t
Bt2 = Bt2 − B(0) =2
dBt2
Z t Z 0t
= 2Bs dBs + ds
0 0
Z t
= 2Bs dBs + t
0

from which we derive

Z t
2Bs dBs = Bt2 − t
0
comparing with
Z t
2g(x)dg(x) = g 2 (t)
0

we again get the extra term.

12
Some Justification
For a given partition t0 = s0 < s1 < . . . < sK+1 = t1 , let
X
X({sj }) = X = a(B(sj ), sj )[B(sj+1 ) − B(sj )] ,
j
X
Y ({sj }) = Y = c(B(sj ), sj )[B(sj+1 ) − B(sj )] .
j
P PK
where j = j=0 .

Fact. Then,
X
X = aj Z j ,
j
X
Y = cj Zj
j

where Zj ⊥ aj , cj , . . . , a0 , c0 , Zj−1 , . . . , Z0 and Zj ∼ N (0, sj+1 − sj ) so in particular,

E[Zj ] = E[N (0, sj+1 − sj )] = 0 , E[Zj2 ] = E[N (0, sj+1 − sj )2 ] = sj+1 − sj .
Then,
E[X({sj })] = E[X]
X 
= E aj Zj
j
X
= E[aj Zj ]
j
X
= E[aj ]E[Zj ] since Zj ⊥ aj
j
X
= E[aj ]0 since E[Zj ] = 0
j
E[X({sj })] = 0
And hence, in the limit (understood in the L2 convergence sense) over partitions with ∆({sj }) =
maxj (sj+1 − sj ) → 0,
 Z t1 
E a(s, Bs )dBs = E lim X({sj })
t0 ∆({sj })→0

= lim E[X({sj })]

∆({sj })→0
= lim 0
∆({sj })→0
= 0
which proves that
Z t1 
E[Ito] = 0 or E a(s, Bs )dBs = 0
t0

13