Solved Problems Compilation Page 1

COMPILED SOLVED PROBLEMS FROM THIS CHAT

Prepared on 2026-05-01

This document compiles the numerical/problem-solving content completed in

this chat. It includes:

1. Oksendal Chapter 3 exercises solved in the chat

2. Oksendal Chapter 4 exercises solved in the chat
3. SCF 2026 Quiz 2 solutions

===========================================================================
SECTION A. OKSENDAL CHAPTER 3
===========================================================================

Exercise 3.1
Question:
Prove directly from the definition of Ito integrals that

integral_0^t s dB_s = t B_t - integral_0^t B_s ds.

Solution:
Take a partition

0 = t_0 < t_1 < ... < t_n = t,

with mesh going to 0, and define the left-step approximation of s by

f_n(s) = sum_{j=0}^{n-1} t_j * 1_{(t_j,t_{j+1}]}(s).

Since s -> s is continuous on [0,t], we have

integral_0^t |f_n(s) - s|^2 ds <= t * |pi_n|^2 -> 0.

Hence, by the definition of the Ito integral,

integral_0^t s dB_s
= lim_{n->infinity} sum_{j=0}^{n-1} t_j (B_{t_{j+1}} - B_{t_j})

in L^2.

Now use the discrete product rule:

t_{j+1} B_{t_{j+1}} - t_j B_{t_j}

= t_j (B_{t_{j+1}} - B_{t_j}) + B_{t_{j+1}} (t_{j+1} - t_j).

Summing over j = 0,...,n-1, the left side telescopes:

t B_t - 0 * B_0
= sum_j t_j (B_{t_{j+1}} - B_{t_j})
+ sum_j B_{t_{j+1}} (t_{j+1} - t_j).

Since B_0 = 0,

sum_j t_j (B_{t_{j+1}} - B_{t_j})

= t B_t - sum_j B_{t_{j+1}} (t_{j+1} - t_j).

Because Brownian paths are almost surely continuous, the second sum is an
ordinary right-endpoint Riemann sum:

sum_j B_{t_{j+1}} (t_{j+1} - t_j) -> integral_0^t B_s ds

almost surely.

Therefore

integral_0^t s dB_s = t B_t - integral_0^t B_s ds.

Solved Problems Compilation Page 2

Exercise 3.2
Question:
Prove directly from the definition of Ito integrals that

integral_0^t B_s^2 dB_s = (1/3) B_t^3 - integral_0^t B_s ds.

Solution:
Take a partition 0 = t_0 < ... < t_n = t, let Delta B_j = B_{t_{j+1}}-B_{t_j}.
By definition,

integral_0^t B_s^2 dB_s

= lim_{n->infinity} sum_{j=0}^{n-1} B_{t_j}^2 Delta B_j

in L^2.

Expand:

B_{t_{j+1}}^3 - B_{t_j}^3
= 3 B_{t_j}^2 Delta B_j + 3 B_{t_j} (Delta B_j)^2 + (Delta B_j)^3.

Hence

sum B_{t_j}^2 Delta B_j

= (1/3) B_t^3
- sum B_{t_j} (Delta B_j)^2
- (1/3) sum (Delta B_j)^3.

Now write

sum B_{t_j} (Delta B_j)^2

= sum B_{t_j} Delta t_j
+ sum B_{t_j} ((Delta B_j)^2 - Delta t_j).

The Riemann sum satisfies

sum B_{t_j} Delta t_j -> integral_0^t B_s ds

in L^2 by continuity of Brownian paths.

For A_n = sum B_{t_j} ((Delta B_j)^2 - Delta t_j), the cross terms vanish and

E[A_n^2]
= sum E[B_{t_j}^2] E[((Delta B_j)^2 - Delta t_j)^2]
= 2 sum t_j Delta t_j^2
<= 2 t^2 |pi_n| -> 0.

For C_n = sum (Delta B_j)^3,

E[C_n^2] = sum E[(Delta B_j)^6] = 15 sum Delta t_j^3 -> 0.

Therefore

sum B_{t_j}^2 Delta B_j

-> (1/3) B_t^3 - integral_0^t B_s ds,

so

integral_0^t B_s^2 dB_s = (1/3) B_t^3 - integral_0^t B_s ds.

Exercise 3.3
Question:
Let H_t^(X) = sigma{X_s : s <= t}.

(a) If X_t is a martingale with respect to some filtration {N_t}, show that
X_t is also a martingale with respect to its own filtration.
Solved Problems Compilation Page 3

(b) If X_t is a martingale with respect to H_t^(X), show that

E[X_t] = E[X_0] for all t >= 0.
(c) Give an example of a process satisfying E[X_t] = E[X_0] for all t >= 0
but not a martingale with respect to its own filtration.

Solution:
(a) Since a martingale is adapted, H_s^(X) subset N_s. Therefore

E[X_t | H_s^(X)]
= E(E[X_t | N_s] | H_s^(X))
= E(X_s | H_s^(X))
= X_s.

So X_t is a martingale with respect to its own filtration.

(b) Take s = 0:

E[X_t | H_0^(X)] = X_0.

Taking expectation gives

E[X_t] = E[X_0].

(c) Take X_t = B_t^3. Then E[X_t] = 0 = E[X_0]. But since x -> x^3 is
invertible,

H_t^(X) = sigma(B_s^3 : s <= t) = sigma(B_s : s <= t).

For s < t,

E[X_t | H_s^(X)]
= E[B_t^3 | F_s]
= B_s^3 + 3(t-s) B_s
!= B_s^3 = X_s.

So X_t is not a martingale with respect to its own filtration.

Exercise 3.4
Question:
Check whether the following are martingales:

(i) X_t = B_t + 4t

(ii) X_t = B_t^2
(iii) X_t = t^2 B_t - 2 integral_0^t s B_s ds
(iv) X_t = B_1(t) B_2(t), where (B_1,B_2) is 2-dimensional Brownian motion.

Solution:
(i) Not a martingale:

E[X_t | F_s] = E[B_t + 4t | F_s] = B_s + 4t != B_s + 4s = X_s.

(ii) Not a martingale:

E[B_t^2 | F_s] = B_s^2 + (t-s) != B_s^2.

(iii) This is a martingale. For s <= t,

E[t^2 B_t | F_s] = t^2 B_s,

and

E[integral_s^t u B_u du | F_s]

= integral_s^t u E[B_u | F_s] du
= integral_s^t u B_s du
= ((t^2 - s^2)/2) B_s.
Solved Problems Compilation Page 4

Hence

E[X_t | F_s]
= t^2 B_s - 2 integral_0^s u B_u du - (t^2 - s^2) B_s
= s^2 B_s - 2 integral_0^s u B_u du
= X_s.

(iv) This is a martingale. Let Delta_i = B_i(t) - B_i(s). Then

E[B_1(t) B_2(t) | F_s]

= E[(B_1(s)+Delta_1)(B_2(s)+Delta_2) | F_s]
= B_1(s) B_2(s),

since the increments are centered and independent of F_s, and the two
components are independent.

Answers:
(i) No, (ii) No, (iii) Yes, (iv) Yes.

Exercise 3.5
Question:
Prove directly that M_t = B_t^2 - t is an F_t-martingale.

Solution:
For s <= t,

E[M_t | F_s]
= E[B_t^2 - t | F_s]
= E[(B_s + (B_t-B_s))^2 - t | F_s]
= B_s^2 + (t-s) - t
= B_s^2 - s
= M_s.

Hence B_t^2 - t is a martingale.

Exercise 3.6
Question:
Prove that N_t = B_t^3 - 3t B_t is a martingale.

Solution:
Let s <= t and Delta = B_t - B_s. Then

E[N_t | F_s]
= E[(B_s + Delta)^3 - 3t(B_s + Delta) | F_s]
= B_s^3 + 3 B_s E[Delta^2 | F_s] - 3t B_s
= B_s^3 + 3B_s(t-s) - 3t B_s
= B_s^3 - 3s B_s
= N_s.

So N_t is a martingale.

Exercise 3.7(a)
Question:
Verify that in each of the n iterated Ito integrals

n! integral_{0 <= u_1 <= ... <= u_n <= t} dB_{u_1} ... dB_{u_n}

the integrand satisfies the requirements in Definition 3.1.4.

Solution:
Define recursively

I_0(u) = 1,
I_k(u) = integral_0^u I_{k-1}(r) dB_r, k >= 1.
Solved Problems Compilation Page 5

Then the k-th integral has integrand I_{k-1}(u). We need adaptedness and
square integrability.

1. Adaptedness:
I_{k-1}(u) depends only on B_r for r <= u, hence it is F_u-measurable.

2. Square integrability:
By Ito isometry,

E[I_k(u)^2] = integral_0^u E[I_{k-1}(r)^2] dr.

Starting from I_0(u)=1, induction gives

E[I_k(u)^2] = u^k / k!.

Therefore

integral_0^t E[I_{k-1}(u)^2] du = t^k / k! < infinity.

Hence each stage is an admissible Ito integrand.

Exercise 3.7(b)
Question:
Verify formula (3.3.8) for n = 1,2,3.

Solution:
The formula is

n! integral_{0 <= u_1 <= ... <= u_n <= t} dB_{u_1} ... dB_{u_n}
= t^(n/2) h_n(B_t / sqrt(t)),

with h_1(x)=x, h_2(x)=x^2-1, h_3(x)=x^3-3x.

n = 1:

integral_0^t dB_u = B_t = t^(1/2) h_1(B_t/sqrt(t)).

n = 2:

2! integral_{0 <= u_1 <= u_2 <= t} dB_{u_1} dB_{u_2}

= 2 integral_0^t (integral_0^{u_2} dB_{u_1}) dB_{u_2}
= 2 integral_0^t B_s dB_s
= B_t^2 - t
= t h_2(B_t/sqrt(t)).

n = 3:

3! integral_{0 <= u_1 <= u_2 <= u_3 <= t} dB_{u_1} dB_{u_2} dB_{u_3}
= 6 integral_0^t (integral_{0 <= u_1 <= u_2 <= s} dB_{u_1} dB_{u_2}) dB_s
= 6 integral_0^t (1/2)(B_s^2 - s) dB_s
= 3 integral_0^t (B_s^2 - s) dB_s.

Using Exercises 3.2 and 3.1:

integral_0^t B_s^2 dB_s = (1/3) B_t^3 - integral_0^t B_s ds

integral_0^t s dB_s = t B_t - integral_0^t B_s ds.

So

3 integral_0^t (B_s^2 - s) dB_s

= B_t^3 - 3t B_t
= t^(3/2) h_3(B_t/sqrt(t)).

Exercise 3.7(c)
Solved Problems Compilation Page 6

Question:
Use part (b) to give a new proof that B_t^3 - 3t B_t is a martingale.

Solution:
From part (b),

B_t^3 - 3t B_t
= 6 integral_{0 <= u_1 <= u_2 <= u_3 <= t} dB_{u_1} dB_{u_2} dB_{u_3}
= 3 integral_0^t (B_s^2 - s) dB_s.

The right-hand side is an Ito integral, hence a martingale. Therefore

B_t^3 - 3t B_t is a martingale.

Exercise 3.8(a)
Question:
Let Y be real-valued with E|Y| < infinity and define

M_t = E[Y | F_t].

Show that M_t is an F_t-martingale.

Solution:
M_t is F_t-measurable by definition of conditional expectation, and

E|M_t| <= E|Y| < infinity.

For s <= t,

E[M_t | F_s]
= E(E[Y | F_t] | F_s)
= E[Y | F_s]
= M_s.

Hence M_t is a martingale.

Exercise 3.8(b)
Question:
Conversely, let M_t be an F_t-martingale such that

sup_{t>=0} E|M_t|^p < infinity

for some p > 1. Show that there exists Y in L^1(P) such that

M_t = E[Y | F_t].

Solution:
Since the martingale is bounded in L^p for some p > 1, the standard
martingale convergence theorem gives a random variable Y in L^p(P) subset
L^1(P) such that

M_t -> Y

almost surely and in L^1 as t -> infinity.

For fixed t and every s >= t,

M_t = E[M_s | F_t].

Let s -> infinity. Since M_s -> Y in L^1, continuity of conditional

expectation in L^1 yields

M_t = E[Y | F_t].

Exercise 3.9
Solved Problems Compilation Page 7

Question:
Compute the Stratonovich integral

integral_0^T B_t o dB_t

and compare with the Ito integral.

Solution:
Let

S_pi = sum_j B_{t_j^*} Delta B_j, where t_j^* = (t_j + t_{j+1})/2.

Write

B_{t_j^*} Delta B_j

= ((B_{t_j} + B_{t_{j+1}})/2) Delta B_j + R_j.

But

((B_{t_j} + B_{t_{j+1}})/2) Delta B_j

= (1/2)(B_{t_{j+1}}^2 - B_{t_j}^2),

so

sum_j ((B_{t_j} + B_{t_{j+1}})/2) Delta B_j = (1/2) B_T^2.

The remainder sum tends to 0 in L^2. Hence

integral_0^T B_t o dB_t = (1/2) B_T^2.

Comparison:

integral_0^T B_t dB_t = (1/2)(B_T^2 - T),

so

integral_0^T B_t o dB_t

= integral_0^T B_t dB_t + (1/2) T.

Exercise 3.10
Question:
Assume

E(|f(s,.) - f(t,.)|^2) <= K |s-t|^(1+epsilon), 0<=s,t<=T.

Show that the Ito and Stratonovich integrals of f coincide.

Solution:
Let

I_pi = sum_j f(t_j) Delta B_j,

S_pi = sum_j f(t_j^*) Delta B_j.

Then

S_pi - I_pi = sum_j (f(t_j^*) - f(t_j)) Delta B_j.

The assumption gives

E|S_pi - I_pi|^2
<= C sum_j E|f(t_j^*) - f(t_j)|^2 Delta t_j
<= C sum_j (Delta t_j)^(2+epsilon)
<= C T |pi|^epsilon -> 0.

Therefore the midpoint and left-endpoint sums have the same L^2-limit, so
Solved Problems Compilation Page 8

integral_0^T f(t,omega) o dB_t(omega)

= integral_0^T f(t,omega) dB_t(omega).

Exercise 3.11
Question:
Let W_t satisfy:
(i) W_{t1} and W_{t2} are independent for t1 != t2,
(ii) the process is stationary,
(iii) E[W_t] = 0.
Show that W_t cannot have continuous paths unless it is trivial.

Solution:
For N >= 1 define the truncation

W_t^(N) = max(-N, min(N, W_t)).

If W_t had continuous paths, then W_t^(N) would also be continuous, so for
fixed t,

(W_t^(N) - W_s^(N))^2 -> 0 almost surely as s -> t.

By bounded convergence,

E[(W_t^(N) - W_s^(N))^2] -> 0.

But for s != t, stationarity and pairwise independence imply W_s^(N) and

W_t^(N) are independent and identically distributed, so

E[(W_t^(N) - W_s^(N))^2]
= 2 E[(W_t^(N))^2] - 2 (E[W_t^(N)])^2,

which does not depend on s. Therefore it must be 0. Hence Var(W_t^(N)) = 0,

so W_t^(N) is almost surely constant.

Let N -> infinity. Then W_t^(N) -> W_t and E[W_t] = 0, so W_t = 0 almost
surely. Thus the only continuous-path possibility is the trivial process.

Exercise 3.12
Question:
Convert the following between Stratonovich and Ito form.

Solution:
Use

dX_t = b(t,X_t) dt + sigma(t,X_t) o dB_t

<=> dX_t = (b + (1/2) sigma_x sigma)(t,X_t) dt + sigma(t,X_t) dB_t.

(i)(a) dX_t = gamma X_t dt + alpha X_t o dB_t

Here sigma(x)=alpha x, sigma_x=alpha, so

dX_t = (gamma + (1/2) alpha^2) X_t dt + alpha X_t dB_t.

(i)(b) dX_t = sin X_t cos X_t dt + (t^2 + cos X_t) o dB_t

Here sigma(t,x)=t^2+cos x, sigma_x=-sin x. Therefore

Ito drift
= sin x cos x - (1/2) sin x (t^2 + cos x)
= (1/2) sin x (cos x - t^2).

So

dX_t = (1/2) sin X_t (cos X_t - t^2) dt + (t^2 + cos X_t) dB_t.
Solved Problems Compilation Page 9

For the reverse direction use

dX_t = a(t,X_t) dt + sigma(t,X_t) dB_t

<=> dX_t = (a - (1/2) sigma_x sigma)(t,X_t) dt + sigma(t,X_t) o dB_t.

(ii)(a) dX_t = r X_t dt + alpha X_t dB_t

becomes

dX_t = (r - (1/2) alpha^2) X_t dt + alpha X_t o dB_t.

(ii)(b) dX_t = 2 e^{-X_t} dt + X_t^2 dB_t

Here sigma(x)=x^2, sigma_x=2x, so (1/2) sigma_x sigma = x^3. Hence

dX_t = (2 e^{-X_t} - X_t^3) dt + X_t^2 o dB_t.

Exercise 3.13
Question:
A process X_t is continuous in mean square if E[X_t^2] < infinity and

lim_{s->t} E[(X_s - X_t)^2] = 0.

(a) Show Brownian motion is continuous in mean square.

(b) If f is Lipschitz, show Y_t = f(B_t) is continuous in mean square.
(c) Show that mean-square continuous processes can be approximated by simple
processes in the L^2 sense needed for Ito integration.

Solution:
(a)

E[B_t^2] = t < infinity,

E[(B_s - B_t)^2] = |s-t| -> 0.

So Brownian motion is continuous in mean square.

(b) If |f(x)-f(y)| <= C |x-y|, then

E[(f(B_s)-f(B_t))^2] <= C^2 E[(B_s-B_t)^2] = C^2 |s-t| -> 0.

Also |f(x)| <= |f(0)| + C|x|, so E[f(B_t)^2] < infinity.

(c) For the dyadic left-point approximations phi_n,

E[(integral_S^T (X_t - phi_n(t)) dB_t)^2]

= E[integral_S^T (X_t - phi_n(t))^2 dt]

by Ito isometry. Mean-square continuity implies the integrand tends to 0,

and boundedness on compact intervals gives dominated convergence. Hence

integral_S^T phi_n(t) dB_t -> integral_S^T X_t dB_t

in L^2.

Exercise 3.14
Question:
Show that h(omega) is F_t-measurable iff it is a pointwise a.e. limit of
sums of functions of the form

g_1(B_{t_1}) ... g_k(B_{t_k}), t_j <= t,

with the g_j bounded continuous.

Solution:
One direction is immediate: each such product is F_t-measurable, so finite
Solved Problems Compilation Page 10

sums and pointwise limits are F_t-measurable.

Conversely, let h be F_t-measurable. By truncation we may assume h is bounded.

Let H_n = sigma(B_{j 2^{-n}} : j 2^{-n} <= t). Then H_n increases to F_t, so

h = lim_{n->infinity} E[h | H_n] almost surely.

Each h_n := E[h | H_n] is H_n-measurable, so by Doob-Dynkin

h_n(omega) = G_n(B_{t_1}(omega),...,B_{t_k}(omega))

for some Borel G_n. Borel functions can be approximated pointwise a.e. by
continuous functions, and by Stone-Weierstrass such continuous functions can
be uniformly approximated on compacts by finite sums of product functions.
Passing to the limit gives the claim.

Exercise 3.15
Question:
Suppose f,g in V(S,T) and

C + integral_S^T f(t,omega) dB_t(omega)

= D + integral_S^T g(t,omega) dB_t(omega) almost surely.

Show that C=D and f=g a.e.

Solution:
Subtract:

(C-D) + integral_S^T (f-g) dB_t = 0 almost surely.

Take expectation. Ito integrals have mean zero, so C-D=0. Thus

integral_S^T (f-g) dB_t = 0 almost surely.

Square and take expectation:

0 = E[(integral_S^T (f-g) dB_t)^2]

= E[integral_S^T (f-g)^2 dt]

by Ito isometry. Therefore (f-g)^2 = 0 a.e., so f=g a.e.

Exercise 3.16
Question:
Let E[X^2] < infinity and H subset F be a sigma-algebra. Show

E[(E[X|H])^2] <= E[X^2].

Solution:
By Jensen's inequality for the convex function u -> u^2,

(E[X|H])^2 <= E[X^2 | H] almost surely.

Take expectation:

E[(E[X|H])^2] <= E[X^2].

Exercise 3.17
Question:
Let G be a finite sigma-algebra with atoms G_1,...,G_n, and X in L^1.

(a) Show E[X|G] is constant on each atom.

(b) Show that on G_i,

E[X|G](omega) = ( integral_{G_i} X dP ) / P(G_i).

Solved Problems Compilation Page 11

(c) Show this agrees with the elementary finite-valued conditional expectation.

Solution:
(a) Every G-measurable function is constant on each atom, so E[X|G] is
constant on each G_i.

(b) Define

Y(omega) = sum_{i=1}^n
( (1/P(G_i)) integral_{G_i} X dP ) * 1_{G_i}(omega).

Then Y is G-measurable. For any A in G, A is a union of some atoms G_i, and

integral_A Y dP
= sum_{G_i subset A} integral_{G_i} X dP
= integral_A X dP.

Hence Y = E[X|G], so on G_i,

E[X|G](omega) = ( integral_{G_i} X dP ) / P(G_i).

(c) If X takes values a_1,...,a_m, then

integral_{G_i} X dP
= sum_{k=1}^m a_k P(X=a_k, G_i).

Dividing by P(G_i),

E[X|G_i]
= sum_{k=1}^m a_k P(X=a_k | G_i)
= E[X|G](omega) on G_i.

===========================================================================
SECTION B. OKSENDAL CHAPTER 4
===========================================================================

Exercise 4.1
Question:
Use Ito's formula to write the following in the standard form

dX_t = u(t,omega) dt + v(t,omega) dB_t.

Solution:
(a) X_t = B_t^2

dX_t = dt + 2 B_t dB_t.

(b) X_t = 2 + t + e^{B_t}

dX_t = (1 + (1/2)e^{B_t}) dt + e^{B_t} dB_t.

(c) X_t = B_1(t)^2 + B_2(t)^2

dX_t = 2 dt + 2 B_1 dB_1 + 2 B_2 dB_2.

(d) X_t = (t_0 + t, B_t)

dX_t = (1,0)^T dt + (0,1)^T dB_t.

(e) X_t = (B_1 + B_2 + B_3, B_2^2 - B_1 B_3)

dX_t^(1) = dB_1 + dB_2 + dB_3

dX_t^(2) = dt - B_3 dB_1 + 2B_2 dB_2 - B_1 dB_3.
Solved Problems Compilation Page 12

Exercise 4.2
Question:
Use Ito's formula to prove

integral_0^t B_s^2 dB_s = (1/3) B_t^3 - integral_0^t B_s ds.

Solution:
Apply Ito to g(x)=x^3/3:

d((1/3)B_t^3) = B_t^2 dB_t + B_t dt.

Integrating gives

(1/3) B_t^3
= integral_0^t B_s^2 dB_s + integral_0^t B_s ds.

Rearrange.

Exercise 4.3
Question:
For real Ito processes X_t, Y_t prove

d(X_t Y_t) = X_t dY_t + Y_t dX_t + dX_t . dY_t,

and deduce integration by parts.

Solution:
Write

dX_t = u_t dt + v_t dB_t,

dY_t = alpha_t dt + beta_t dB_t.

Then

dX_t dY_t = v_t beta_t dt.

Hence

d(X_t Y_t) = X_t dY_t + Y_t dX_t + dX_t dY_t.

Integrating,

X_t Y_t - X_0 Y_0

= integral_0^t X_s dY_s
+ integral_0^t Y_s dX_s
+ integral_0^t dX_s . dY_s.

Therefore

integral_0^t X_s dY_s

= X_t Y_t - X_0 Y_0
- integral_0^t Y_s dX_s
- integral_0^t dX_s . dY_s.

Exercise 4.4
Question:
For

Z_t = exp( integral_0^t theta(s,omega) dB(s)

- (1/2) integral_0^t theta^2(s,omega) ds ),

prove:
(a) dZ_t = Z_t theta(t,omega) dB(t)
(b) deduce Z_t is a martingale under the stated integrability condition.
Solved Problems Compilation Page 13

Solution:
Let

M_t = integral_0^t theta_s dB_s - (1/2) integral_0^t theta_s^2 ds.

Then

dM_t = theta_t dB_t - (1/2) theta_t^2 dt,

(dM_t)^2 = theta_t^2 dt.

Apply Ito to e^{M_t}:

dZ_t = Z_t dM_t + (1/2) Z_t (dM_t)^2 = Z_t theta_t dB_t.

Hence

Z_t = 1 + integral_0^t Z_s theta_s dB_s.

If Z_t theta_k(t,omega) is an admissible Ito integrand for each k, then the

right-hand side is an Ito integral, so Z_t is a martingale.

Exercise 4.5
Question:
Let beta_k(t) = E[B_t^k]. Prove

beta_k(t) = (1/2) k(k-1) integral_0^t beta_{k-2}(s) ds, k>=2,

and find E[B_t^6].

Solution:
Apply Ito to B_t^k:

d(B_t^k) = k B_t^{k-1} dB_t + (1/2) k(k-1) B_t^{k-2} dt.

Take expectation:

beta_k(t) = (1/2) k(k-1) integral_0^t beta_{k-2}(s) ds.

Now beta_0 = 1, beta_2(t)=t. Then

beta_4(t) = 6 integral_0^t s ds = 3 t^2,

beta_6(t) = 15 integral_0^t 3s^2 ds = 15 t^3.

So E[B_t^6] = 15 t^3.

Exercise 4.6
Question:
For the given exponential processes, compute dX_t.

Solution:
(a) X_t = e^{c t + alpha B_t}

dX_t = (c + (1/2) alpha^2) X_t dt + alpha X_t dB_t.

(b) X_t = exp(c t + sum_{j=1}^n alpha_j B_j(t))

dX_t
= (c + (1/2) sum alpha_j^2) X_t dt
+ X_t sum alpha_j dB_j(t).

Exercise 4.7
Question:
If
Solved Problems Compilation Page 14

X_t = integral_0^t v_s dB_s,

show X_t^2 need not be a martingale, and prove

M_t = X_t^2 - integral_0^t |v_s|^2 ds

is a martingale when v is bounded.

Solution:
(a) Take v_s = 1. Then X_t = B_t and X_t^2 = B_t^2, which is not a martingale.

(b) By Ito,

d(X_t^2) = 2 X_t dX_t + (dX_t)^2

= 2 X_t v_t dB_t + |v_t|^2 dt.

Hence

dM_t = 2 X_t v_t dB_t.

If v is bounded, then 2X_t v_t is admissible, so M_t is an Ito integral and

hence a martingale.

Exercise 4.8
Question:
Prove the Brownian-motion form of Ito's formula, and extend it to
g in C^1 with bounded g'' away from finitely many points.

Solution:
(a) For Brownian motion in R^n, use multidimensional Ito on f(B_t):

df(B_t) = grad f(B_t) . dB_t + (1/2) Delta f(B_t) dt.

Integrating:

f(B_t) = f(B_0)
+ integral_0^t grad f(B_s) . dB_s
+ (1/2) integral_0^t Delta f(B_s) ds.

(b) Approximate g by smooth functions f_k in C^2 with

f_k -> g, f_k' -> g' uniformly, |f_k''| bounded, and f_k'' -> g'' away from
the exceptional points. Apply part (a) to f_k and pass to the limit using
uniform convergence and dominated convergence.

Exercise 4.9
Question:
Explain why one may assume g, g_x, g_xx are bounded in the proof of Ito's
formula.

Solution:
Let tau_n = inf{s>0 : |X_s| >= n}, and choose g_n in C^2 with g_n = g on the
region reached by X up to time t ^ tau_n, and with bounded derivatives.
Apply Ito's formula to g_n stopped at t ^ tau_n. Since g_n and g coincide on
the stopped path, the formula for g_n gives the formula for g up to tau_n.
Then let n -> infinity and use P(tau_n > t) -> 1.

Exercise 4.10
Question:
Derive Tanaka's formula for |B_t|.

Solution:
Define

g_epsilon(x) = |x|, if |x| >= epsilon,

Solved Problems Compilation Page 15

= (1/2)(epsilon + x^2/epsilon), if |x| < epsilon.

Then

g_epsilon'(x) = sign(x) for |x| >= epsilon,

= x/epsilon for |x| < epsilon,
g_epsilon''(x) = (1/epsilon) 1_{|x|<epsilon}.

Apply the generalized Ito formula:

g_epsilon(B_t)
= g_epsilon(B_0)
+ integral_0^t g_epsilon'(B_s) dB_s
+ (1/(2 epsilon)) * |{s in [0,t] : |B_s| < epsilon}|.

As epsilon -> 0, the middle integral converges in L^2 to

integral_0^t sign(B_s) dB_s,

and g_epsilon(B_t) -> |B_t|. Therefore

|B_t|
= |B_0| + integral_0^t sign(B_s) dB_s + L_t,

where

L_t = lim_{epsilon downarrow 0}

(1/(2 epsilon)) * |{s in [0,t] : |B_s| < epsilon}|

is the local time at 0.

Exercise 4.11
Question:
Show the following are martingales:

e^{t/2} cos B_t,

e^{t/2} sin B_t,
(B_t + t) e^{-B_t - t/2}.

Solution:
(a) For X_t = e^{t/2} cos B_t,

dX_t = - e^{t/2} sin B_t dB_t.

(b) For Y_t = e^{t/2} sin B_t,

dY_t = e^{t/2} cos B_t dB_t.

(c) Let U_t = B_t + t and V_t = e^{-B_t - t/2}. Then dU_t = dt + dB_t,
dV_t = -V_t dB_t. Product rule gives

d(U_t V_t) = e^{-B_t - t/2} (1 - B_t - t) dB_t.

Each process is therefore an Ito integral, hence a martingale.

Exercise 4.12
Question:
If

dX_t = u(t,omega) dt + v(t,omega) dB_t

and X_t is an F_t-martingale, prove u = 0 a.e.

Solution:
For s >= t,
Solved Problems Compilation Page 16

0 = E[X_s - X_t | F_t]

= E[integral_t^s u(r) dr | F_t],

because the Ito integral from t to s has conditional expectation 0. Hence

integral_t^s E[u(r)|F_t] dr = 0

for all s >= t. Therefore E[u(s)|F_t]=0 for a.e. s>t. Let t increase to s and
use the continuity property of conditional expectations to conclude

u(s)=0 a.s. for a.e. s.

Exercise 4.13
Question:
For

dX_t = u_t dt + dB_t,

M_t = exp( - integral_0^t u_r dB_r - (1/2) integral_0^t u_r^2 dr ),
Y_t = X_t M_t,

prove Y_t is a martingale.

Solution:
From the exponential martingale formula,

dM_t = -u_t M_t dB_t.

Using the product rule,

dY_t = X_t dM_t + M_t dX_t + dX_t dM_t

= -u_t X_t M_t dB_t + M_t(u_t dt + dB_t) - u_t M_t dt
= M_t(1 - u_t X_t) dB_t.

Thus Y_t is an Ito integral and hence a martingale.

Exercise 4.14
Question:
Find f(t,omega) in V[0,T] such that

F = E[F] + integral_0^T f(t,omega) dB_t

for each listed F.

Solution:
(a) F = B_T

B_T = integral_0^T 1 dB_t.

So f_t = 1.

(b) F = integral_0^T B_t dt

integral_0^T B_t dt = integral_0^T (T-t) dB_t.

So f_t = T-t.

(c) F = B_T^2

B_T^2 = T + integral_0^T 2 B_t dB_t.

So E[F]=T and f_t = 2 B_t.

(d) F = B_T^3
Solved Problems Compilation Page 17

E[B_T^3 | F_t] = B_t^3 + 3(T-t) B_t,

so

d E[B_T^3 | F_t] = 3(B_t^2 + T - t) dB_t.

Hence E[F]=0 and f_t = 3(B_t^2 + T - t).

(e) F = e^{B_T}

E[e^{B_T} | F_t] = e^{B_t + (1/2)(T-t)},

so

d E[e^{B_T} | F_t] = e^{B_t + (1/2)(T-t)} dB_t.

Hence E[F] = e^{T/2} and

f_t = e^{B_t + (1/2)(T-t)}.

(f) F = sin B_T

E[sin B_T | F_t] = e^{-(1/2)(T-t)} sin B_t,

so

d E[sin B_T | F_t] = e^{-(1/2)(T-t)} cos B_t dB_t.

Hence E[F]=0 and

f_t = e^{-(1/2)(T-t)} cos B_t.

Exercise 4.15
Question:
Let

X_t = (x^(1/3) + (1/3) B_t)^3, x > 0.

Show that

dX_t = (1/3) X_t^(1/3) dt + X_t^(2/3) dB_t, X_0 = x.

Solution:
Set

Y_t = x^(1/3) + (1/3) B_t,

X_t = Y_t^3.

Then dY_t = (1/3) dB_t. By Ito,

dX_t = d(Y_t^3)
= 3Y_t^2 dY_t + (1/2)*6Y_t (dY_t)^2
= Y_t^2 dB_t + (1/3) Y_t dt.

Since Y_t = X_t^(1/3),

dX_t = X_t^(2/3) dB_t + (1/3) X_t^(1/3) dt.

Also X_0 = x.

===========================================================================
SECTION C. SCF 2026 QUIZ 2 SOLUTIONS
===========================================================================

Q1
Solved Problems Compilation Page 18

Question:
Prove/Disprove that the following are Brownian motions, where c is a real
constant:

(a) -B_t
(b) c B_{t/c^2} (the stacked notation in the PDF was interpreted this way)

Solution:
(a) X_t = -B_t

1. X_0 = -B_0 = 0.
2. X_t has continuous paths because B_t does.
3. For 0=t_0<...<t_n,

X_{t_k} - X_{t_{k-1}}
= -(B_{t_k} - B_{t_{k-1}}),

so the increments are independent.

4. Each increment is normal with mean 0 and variance t_k-t_{k-1}.

Therefore -B_t is a Brownian motion.

(b) X_t = c B_{t/c^2}, c != 0

1. X_0 = cB_0 = 0.
2. X_t has continuous paths.
3. For 0=t_0<...<t_n,

X_{t_k} - X_{t_{k-1}}
= c (B_{t_k/c^2} - B_{t_{k-1}/c^2}),

so increments are independent.

4. Mean is 0 and variance is

c^2 * ((t_k-t_{k-1})/c^2) = t_k - t_{k-1}.

Thus c B_{t/c^2} is a Brownian motion for c != 0.

Q2
Question:
Prove/Disprove that the following are martingales:

(a) B_t^2
(b) B_t^3

Solution:
(a) For 0<=s<=t,

E[B_t^2 | F_s]
= E[(B_s + (B_t-B_s))^2 | F_s]
= B_s^2 + (t-s) != B_s^2.

So B_t^2 is not a martingale.

(b) For 0<=s<=t,

B_t^3
= B_s^3 + 3B_s^2(B_t-B_s) + 3B_s(B_t-B_s)^2 + (B_t-B_s)^3.

Taking conditional expectation:

E[B_t^3 | F_s] = B_s^3 + 3(t-s) B_s != B_s^3.

So B_t^3 is not a martingale.

Remark:
Solved Problems Compilation Page 19

The corrected martingales are B_t^2 - t and B_t^3 - 3tB_t.

Q2 (second occurrence on the sheet)

Question:
State and prove Ito's Lemma (Ito or Ito-Doeblin Formula).

Statement:
Let

dX_t = a(t,omega) dt + b(t,omega) dB_t,

where X_t is an Ito process, and let f=f(t,x) belong to C^{1,2}. Then

df(t,X_t)
= [ f_t(t,X_t) + a(t,omega) f_x(t,X_t)
+ (1/2) b^2(t,omega) f_xx(t,X_t) ] dt
+ b(t,omega) f_x(t,X_t) dB_t.

Equivalently,

f(t,X_t) - f(0,X_0)
= integral_0^t [ f_t(s,X_s) + a(s,omega) f_x(s,X_s)
+ (1/2) b^2(s,omega) f_xx(s,X_s) ] ds
+ integral_0^t b(s,omega) f_x(s,X_s) dB_s.

Proof sketch:
Take a partition 0=t_0<...<t_n=t and apply second-order Taylor expansion to

f(t_{k+1}, X_{t_{k+1}}) - f(t_k, X_{t_k}).

Using

Delta X_k ~ a(t_k,omega) Delta t_k + b(t_k,omega) Delta B_k,

we get

(Delta X_k)^2
~ a^2 (Delta t_k)^2 + 2ab Delta t_k Delta B_k + b^2 (Delta B_k)^2.

In the limit only the term b^2 (Delta B_k)^2 survives, because

(Delta B_k)^2 ~ Delta t_k,

Delta t_k Delta B_k -> 0,
(Delta t_k)^2 -> 0.

Summing and passing to the limit yields the stated formula.

Q3
Question:
A large institutional investor under the National Pension System is forced
to sell a large quantity of shares. Let F_t be the fundamental price and
L_t the temporary liquidity discount. The observed traded price is

S_t = F_t - L_t.

Assume:
1. F_t follows a geometric Brownian motion with drift mu and volatility sigma.
2. L_t is positive immediately after the sale and decays at a rate
proportional to its current size.
3. L_0 = ell > 0 and S_0 = s_0.
4. mu, sigma, lambda > 0 are constants.

(a) Form a deterministic differential equation for L_t and solve it.
(b) Form an SDE for S_t = F_t - L_t and solve it.
Solved Problems Compilation Page 20

Solution:
Since the liquidity discount decays at a rate proportional to its current
size,

dL_t/dt = -lambda L_t, L_0 = ell,

or

dL_t = -lambda L_t dt.

Solving:

L_t = ell e^{-lambda t}.

Now the fundamental price follows a GBM:

dF_t = mu F_t dt + sigma F_t dB_t.

Since S_t = F_t - L_t,

dS_t = dF_t - dL_t

= mu F_t dt + sigma F_t dB_t + lambda L_t dt.

Using F_t = S_t + L_t,

dS_t
= [ mu(S_t + L_t) + lambda L_t ] dt
+ sigma(S_t + L_t) dB_t.

Substitute L_t = ell e^{-lambda t}:

dS_t
= [ mu(S_t + ell e^{-lambda t}) + lambda ell e^{-lambda t} ] dt
+ sigma(S_t + ell e^{-lambda t}) dB_t,
with S_0 = s_0.

To solve it, note that F_0 = S_0 + L_0 = s_0 + ell. Therefore

F_t = (s_0 + ell)

exp( (mu - (1/2)sigma^2)t + sigma B_t ).

Hence

S_t = F_t - L_t

= (s_0 + ell) exp( (mu - (1/2)sigma^2)t + sigma B_t )
- ell e^{-lambda t}.

===========================================================================
END OF COMPILATION
===========================================================================