ICEF master program. Maths for economists.

Exam collection.
Kirill Bukin, Dmitri Pervouchine, Boris Demeshev
November 4, 2024

Contents
1 2008-2009 3
1.1 Exam, 12.01.2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Exam, 12.01.2009, solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Retake, ⁇.⁇.2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Retake, ⁇.⁇.2010-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 2009-2010 8
2.1 Exam, 14.01.2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Exam-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 2010-2011 10
3.1 Exam, 12.01.2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Exam-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Retake, ⁇.01.2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Retake-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5 Reretake, 08.02.2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.6 Reretake, 08.02.2011, solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 2011-2012 15
4.1 Retake, 03.02.2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Retake-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

5 2012-2013 17
5.1 Midterm, 13.11.2012 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Exam, 10.01.2013 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 exam 10.01.2013-hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.4 Retake, 08.02.2013 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.5 retake — short hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

6 2013-2014 22
6.1 Stochastic calculus hometask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6.2 Solution for stochastic calculus hometask . . . . . . . . . . . . . . . . . . . . . . . . 22
6.3 Exam, 17.01.2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.4 marking scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.5 Retake, 08.02.2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1
 6.6 Answers and hints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

7 2014-2015 35
7.1 Practice problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.2 Practice-solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.3 Hometask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.4 Hometask-solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.5 Exam, 27.12.2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
7.6 Exam, 27.12.2014 - solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

8 2015-2016 41
8.1 Stochastic calculus hometask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2 Stochastic calculus hometask — Solution . . . . . . . . . . . . . . . . . . . . . . . . 42
8.3 Exam, 12.01.2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.4 Solution, exam 12.01.2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
8.5 Retake, 15.02.2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
8.6 Retake, solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

9 2016-2017 47
9.1 Hometask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
9.2 ht, solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
9.3 exam, 10.01.2017 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
9.4 exam, solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

10 2017-2018 53
10.1 Hometask . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
10.2 Hometask soltuion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
10.3 Exam 2017-12-29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
10.3.1 Dynamic optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
10.3.2 Stochastic Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
10.4 Exam, solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

11 Some junk 58
11.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
11.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

12 2018-2019 59
12.1 Stochastic calculus: home assignment . . . . . . . . . . . . . . . . . . . . . . . . . . 59
12.2 Exam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
12.3 Solution, stochastic calculus part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

13 2019-2020 61
13.1 Exam, 2019-12-23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
13.2 Stochastic calculus, hints and solutions . . . . . . . . . . . . . . . . . . . . . . . . . 62

14 2020-2021 63
14.1 Exam, 2020-12-29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
14.2 Marking scheme, exam 2020-12-29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

2
15 2021 Fall 64
15.1 Exam, 2021-12-21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
15.2 Marking scheme, 2021-12-21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

16 2022 Fall 65
16.1 Exam, 2022-12-29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
16.2 Marking scheme, 2022-12-29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

17 2023 Fall 68
17.1 Exam, 2023-12-27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
17.2 Marking scheme, 2023-12-27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

1 2008-2009
1.1 Exam, 12.01.2009
Final exam consists of the two parts: A and B. Part A lasts for 110 minutes. Upon completion of
that part the papers will be collected and the students will have 10 minutes break. Part B lasts for
70 minutes. Students should answer eight of the following nine questions: six from Part A and two
from Part B. Points will be deducted for the insufficient explanation within your answers.
Part A. Answer all SIX questions of this section. Each question is worth 10 points.
1. The joint distribution of vector (X, Y ) is given by P(X = i, Y = j) = 0.1 for 1 ⩽ i ⩽ j ⩽ 4.
Find E(Y | X).
2. The random variable X is exponentially distributed with parameter λ. The random variable
Y is exponentially distributed with parameter 1/X. Find E(Y | X), E(Y ) and Var(Y ).
3. Let Yt = Wt3 − 3tWt .
(a) Using Ito’s lemma find dYt .
(b) Using your previous result find E(Yt ) and Var(Yt ).
4. Using a current value Hamiltonian maximize the integral
Z 2  
−t 5 2
e x − x − 2y 2
dt
0 2

subject to the conditions ẋ = y − x/2, x(0) = 0, x(2) is free. Find x, y, µ.

5. Solve the bounded control problem
Z T
e−rt (1 − u)x dt → max
0

, subject to ẋ = xu, x(0) = 1, 0 ⩽ u ⩽ 1, where 0 < r < 1.

6. Consider the following optimization problem

X
∞
u(at ) → max
0
P∞
, subject to 0 at ⩽ s, s > 0, at ⩾ 0.
(a) Show that if u(a) is increasing and strictly concave this problem has no solution

3
 P∞
(b) What happens if the sum in the maximization problem is changed to 0 δ t u(at ) , where
0 < δ < 1?
Part B. Answer two questions out of the three from this section. Each question is worth 20 points.
Part B lasts for 70 minutes.
7. In the framework of the Black and Scholes model find the price at t = 0 of an asset that pays
max{0, ln ST } at time T , where ST denotes the price of one share at time T .
8. Let’s consider the following system of stochastic differential equations

dXt = aXt dt − Yt dWt
dYt = aYt dt + Xt dWt

with initial conditions X0 = x0 and Y0 = 0

(a) Find the solution of the form Xt = f (t) cos Wt and Yt = g(t) sin Wt
(b) Prove that for any solution Dt = Xt2 + Yt2 is nonstochastic
9. Consider the profit-maximizing problem for a representative competitive firm
Z ∞
(p − wn(t))q(t)e−rt dt → max
0

, subject to (*) ẋ = x(1 − x) − q, where the state variable x(t) represents

p a renewable stock
resource (fish) that evolves according to the equation (*) and q(t) = 2 x(t)n(t) is the ex-
traction rate. Here n(y) is a labor effort with a constant wage rate w. The price of fish is
assumed to be constant and equal p. The optimization problem is to choose n(t) to maximize
the discounted profits, 0 < r < 1.
(a) Derive necessary conditions
(b) Draw the phase diagram for this problem with the fish stock and the multiplier labeled
on the axes
(c) Show that if p/w is sufficiently large the fish stock will be driven to zero, while p/w is
low there is a steady-state with a positive stock

1.2 Exam, 12.01.2009, solutions

1. Joint distribution table of X and Y is:

X/Y 1 2 3 4
1 0.1 0.1 0.1 0.1
2 0 0.1 0.1 0.1
3 0 0 0.1 0.1
4 0 0 0 0.1

Fixing a certain value of X we find the conditional expectation of Y :

 1+2+3+4

 = 2.5, if x = 1
 4
 2+3+4 = 3, if x = 2
E(Y | X) = 3+43

 = 3.5, if x = 3

42
1
= 4, if x = 4

4
 Or, simply, E(Y | X) = 2 + X/2.
2. Variable X has exponential distribution, so

fX (x) = λe−λx

And we know conditional distribution of Y :

1 −1y
fY |X (x, y) = e x
x
So, Z Z  −1
+∞ +∞
1 1
y e− X y dy =
1
E(Y | X) = yfY |X (X, y)dy = =X
−∞ 0 X X
Expected value:
Z +∞ Z +∞
E(Y ) = E(E(Y | X)) = E(X) = xfX (x) dx = xλe−λx dx = λ−1
−∞ 0

Variance:

Var(Y ) = Var(E(Y | X)) + E(Var(Y | X)) =

 −2 !
1
= Var(X) + E = λ−2 + E(X 2 )
X

We know that E(X 2 ) = Var(X) + (E(X))2 , so:

Var(Y ) = λ−2 + λ−2 + λ−2 = 3λ−2

3. Using Ito’s lemma dYt = (3Wt2 − 3t) dWt + (−3Wt + 12 6Wt ) dt = (3Wt2 − 3t) dWt . Hence,
Yt is a martingale. Y0 = W03 − 3 · 0W0 = 0.
Now E(Yt ) = E(Y0 ) = 0 and using Ito’s isometry:
Z t Z t Z t
Var(Yt ) = E((3Ws − 3s) ) ds =
2 2
27s + 9s − 18s ds =
2 2 2
18s2 ds = 6t3
0 0 0

Here we have used the facts that E(Ws2 ) = s and E(Ws4 ) = 3s2 .
4.
5.
6.
7.
8. Using Ito’s lemma we find dXt and dYt

dXt = ft′ (t) cos Wt dt − f (t) sin Wt dWt − 0.5f (t) cos Wt dt =
= (ft′ (t) − 0.5f (t)) cos Wt dt − f (t) sin Wt dW t

dYt = gt′ (t) sin Wt dt + g(t) cos Wt dWt − 0.5g(t) sin Wt dt =

= (gt′ (t) − 0.5g(t)) sin Wt dt + g(t) cos Wt dW t

5
 Comparing these expressions with the system we receive:

ft′ (t) − 0.5f (t) = af (t)

f (t) = g(t)
gt′ (t) − 0.5g(t) = ag(t)
g(t) = f (t)
The general solution has the form

f (t) = g(t) = Ae(a+0.5)t , A∈R

From initial condition we get A = x0 and

f (t) = g(t) = x0 e(a+0.5)t

Answer:
Rt Rt
Xt = x0 + x0 0 ae(a+0.5)u cos Wu du − x0 0 e(a+0.5)t sin Wu dW u
Rt Rt
Yt = x0 0 ae(a+0.5)u sin Wu du + x0 0 e(a+0.5)t cos Wu dW u
Now we use Ito’s lemma once again:

dDt = 2Xt dXt + 2Yt dYt = 2Xt ((ft′ (t) − 0.5f (t)) cos Wt dt − f (t) sin Wt dW t)+
+ 2Yt ((gt′ (t) − 0.5g(t)) sin Wt ], dt + g(t) cos Wt dW t) =
= 2f (t)(ft′ (t) − 0.5f (t)) cos2 Wt dt − f 2 (t) cos Wt sin Wt dW t+
+ 2f (t)(ft′ (t) − 0.5f (t)) sin2 Wt dt + f 2 (t) cos Wt sin Wt dW t =
= 2f (t)(ft′ (t) − 0.5f (t) dt = 2af 2 (t) dt

Z t
2ax0 (2a+1)t
Dt = D0 + 2ax0 e(2a+1)u du = x20 + (e − 1)
0 2a + 1

1.3 Retake, ⁇.⁇.2009

Final exam consists of the two parts: A and B. Part A lasts for 110 minutes. Upon completion of
that part the papers will be collected and the students will have 10 minutes break. Part B lasts for
70 minutes. Students should answer eight of the following nine questions: six from Part A and two
from Part B. Points will be deducted for the insufficient explanation within your answers.
Part A. Answer all SIX questions of this section. Each question is worth 10 points.
1. In the first bag there balls numbered from 0 to 9, in the second bag there are balls numbered
from 1 to 10. Two balls were selected. You know that one ball was selected from the first bag
and one from the second one. You will select at random one ball from these two and you will
know only its number. Let’s denote its number by X and the number of the other of the two
balls by Y .
Find E(Y | X)
2. The random variable X is uniformly distributed on [0; a]. The random variable Y is uniformly
distributed on [0; X].
Find E(Y | X),  E(Y ) and Var(Y
 ).
a2
3. Let Yt = exp −aWt − 2
t .

6
 (a) Using Ito’s lemma find dYt
(b) Using your previous result find E(Yt ) and Var(Yt )
R1

4. Find extremals for the integral 0 12 ẋ2 + xẋ + x dt when x(0) = 0 and x(1) is chosen freely.
R2
5. Solve the bounded control problem 0 (2x − 3u) dt → max , subject to ẋ = x + u, x(0) = 5,
0 ⩽ u ⩽ 2, where x(2) is free.
P
6. Consider the following optimization problem: maximize ∞ 0 β u(ct ) , subject to ct + kt+1 =
t

f (kt ), 0 < β < 1, where both the utility function u(t) and the production function f (k) have
the standard properties of monotonicity and strict concavity. Let the state variable be k and
denote the next period value of k as k ′ . Substitute c = f (k) − k ′ into utility function and write
down the Bellman equation.
Using the formal differentiation of the Bellman equation with respect to k under the sign
of max, and drawing FOC from Bellman equation, exclude the value function and find the
equation which combines the values of u′ and f ′ at the adjacent time periods.
Part B. Answer two questions out of the three from this section. Each question is worth 20 points.
Part B lasts for 70 minutes.
7. In the framework of the Black and Scholes model find the price at t = 0 of an asset that pays
min{M, ln St } at time T , where ST denotes the price of one share at time T , M — arbitrary
constant, specified at the moment of the issue.
8. The price of a share in euros is driven by the equation dS = σSdW + αSdt, the dollar/euro
exchange rate is driven by the equation dU = bU dW + cU dt. Find the current price in dollars
of a European call option with maturity date T , strike price K.
R∞
9. Consider the profit-maximizing problem for a representative competitive firm 0 (p −
c(x(t)))q(t)e−rt dt → max , subject to (*) ẋ = 1 − x − q, where the state variable x(t) < 1
represents a nonrenewable stock resource (oil) that depletes according to the equation (*) and
q(t) is the extraction rate. Here c(x) is a cost function of the extraction that is defined by
c(x) = e−x . The price of oil is assumed to be constant and equal p where 1/e < p < 1. The
optimization problem is to choose q(t) to maximize the discounted profits, 0 < r < 1.

(a) Derive necessary conditions.

(b) Prove that the steady-state exists and is unique.

1.4 Retake, ⁇.⁇.2010-solutions

1. If X = 0 we know that that the chosen ball was in the first basket, consequently Y was in the
second one.
1 + 2 + · · · + 10
E(Y | X = 0) = E(Y | second basket ) = = 5.5
10
Similarly
0 + 1 + ··· + 9
E(Y | X = 10) = E(Y | first basket ) = = 4.5
10
Finaly,
1 1
E(Y | 0 < X < 10) = E(Y | first basket )+ E(Y | second basket ) = 0.5(5.5+4.5) = 5
2 2
7
 Answer: 

5.5, if X = 0
E(Y | X) = 5, if 0 < X < 10


4.5, if X = 10
2. E(Y | X) = X/2, E(Y ) = E(X/2) = a/4,

7a2
Var(Y ) = E(Var(Y | X)) + Var(E(Y | X)) = E(X /12) + Var(X/2) = 2
2
12
3. dYt = −aYt dWt , so Yt is a martingale, and E(Yt ) = Y0 = 1. To find variance one may use
Ito’s isometry.
4.
X0 = Ep̃ (exp (−rT ) XT | F0 )
5. We need to find a price of the classic European option in dollars, while stock prices are in
euros.

2 2009-2010
2.1 Exam, 14.01.2010
Final exam consists of the two parts: A and B. Part A lasts for 120 minutes. Upon completion of
that part the papers will be collected and the students will have 10 minutes break. Part B lasts for
60 minutes. Students should answer eight of the following eight questions: six from Part A and two
from Part B. Points will be deducted for the insufficient explanation within your answers.
Part A. Answer all SIX questions of this section. Each question is worth 10 points.
1. The joint distribution of the random vector (X, Y ) is given by its p.d.f
 x−y
ce , for 0 ⩽ x, y ⩽ 1
f (x, y) =
0, otherwise

where c is a normalization constant. Find E(X | Y ).

2. Let Yt = Wt3 − tWt4 . Find E(Yt ) and Var(Yt ). You don’t have to use Ito Lemma here.
3. Let Xn be a discrete time stochastic process that converges in probability to a random number
X as n → ∞. Does this condition imply that Xn converges to X in mean? Almost surely?
In distribution? Support at least one of your answers with a proof or counterexample. Give a
definition for each type of convergence.
R ln 2
4. Seek to optimize the
integral 0
2
(4u − u2 − x − 3x2 ) dt subject to the conditions ẋ = u + x,
x(0) = 5/8, x ln22 is free. Find x, u, λ. What kind of optimum did you find?
R1
5. Solve the bounded control problem 0 (2 − 5t)u dt , subject to ẋ = 2x + 4te2t u, x(0) = 0,
x(1) = e2 , −1 ⩽ u ⩽ 1.
P 3
t
6. Consider the following optimization problem: maximize ∞
√ 0 4
ln ct , subject to ct + kt+1 =
kt , k0 > 0. Let the state variable be k and denote the next period value of k as k ′ .
(a) Write down the Bellman equation for the value function V (k)
(b) Using method of undetermined coefficients find V (k)
(c) Find the optimal policy function k ′ = h(k)

8
Part B. Answer both questions from this section. Each question is worth 20 points. Part B lasts for
60 minutes.
7. Let Xt be a stochastic process such that dXt = X∞τ−Xt dt + σdWt , where X∞ and τ are non-
random constants, Wt is a Wiener process, and let Yt = Xt et/τ .
(a) Use Ito Lemma to find both differential and integral expressions for Yt and use them to
express Xt (a.s.) in terms of X∞ , X0 , τ , σ and t. Here X0 is the value of Xt at time t = 0.
(b) Find E(Xt ) and Var(Xt ). Sketch the graph of E(Xt ) as a function of t for X∞ = 1, τ = 1,
and X0 = 0, 1, and 2. Plot a possible trajectory of Xt in each case. Is there any name or
names associated with Xt ?
8. Utility U (C, P ) = U1 (C) + U2 (P ) increases with the consumption C and decreases with the
level of pollution P . For C > 0, P > 0 it is known that U1′ > 0, U1′′ < 0, U2′ < 0 and U2′′ < 0.
It is known that limC→0 U1′ (C) = ∞ and limP →0 U2′ (C) = 0.
Consumption lies within the range 0 ⩽ C ⩽ C̄. Consumption contributes to pollution, while
pollution control reduces it; moreover environment absorbs pollution at a constant rate b > 0.
Pollution dynamics is governed by equation Ṗ = C 2 − (C ∗ )2 − bP in which the first two
terms represent the net contribution to the pollution flow, where 0 < C ∗R< C̄.
∞
Consider the problem of maximizing the discounted (r > 0) utility stream 0 e−rt U (C, P ) dt →
max subject to Ṗ = C 2 − (C ∗ )2 − bP , P(0) = P0 > 0, P ⩾ 0, 0 ⩽ C ⩽ C̄.
(a) Derive necessary conditions, using the current value Hamiltonian.
(b) Sketch the phase diagram for this problem with the pollution and the consumption la-
beled on the axes.
(c) Find the condition under which the steady state solution (Ps , Cs ) exists and 0 < Cs < C̄.
(d) Explore the stability of the steady state. Hint. You may find the return to the variables
(P, m) easier for solving that part.

2.2 Exam-solutions
R1R1
1. The probability density should integrate to one, so from 0 0 f (x, y) dx dy = 1 we infer the
value of c, c = e/(e − 1)2 .
Using the formula f (x | y) = f (x, y)/f (y) we obtain f (x | y) = ex /(e − 1).
R1
We obtain marginal density, f (y) = 0 f (x, y) dx = e1−y /(e − 1) and hence the conditional
density f (x | y) = ex /(e − 1). As one may notice it does not depend on y. The reason is that
the random variables X and Y are independent and f (x, y) = fX (x)fY (y). So we may expect
that E(X | Y ) = E(X).
And coup de grâce
Z 1
E(X | Y ) = xf (x | Y ) dx = 1/(e − 1)
0

2. E(Yt ) = E(Wt3 − tWt4 ) = E(Wt3 ) − E(tWt4 ) = E(Wt3 ) − t E(Wt4 ) = 0 − 3t3 = −3t3 , as

Rt
E(Wt3 ) = 3(3−1) E(Ws3−2 )ds = 0
2 R 0
t Rt
E(Wt4 ) = 4(4−1)
2 0
E(W 4−2
s )ds = 6 0
sds = 3t2
Var(Yt ) = E(Yt ) − (E(Y ))
2 2

E(Yt2 ) = E(Wt3 − tWt4 )2 = E(Wt6 − 2tWt7 + t2 Wt8 ) = E(Wt6 ) − 2t E(Wt7 ) + t2 E(Wt8 ) =

15t3 + 105t6 , Rt
as E(Wt5 ) = 5(5−1)
2 0
E(Ws5−2 )ds = 0

9
 Rt Rt 2
E(Wt6 ) = 6(6−1) E(W 6−2
)ds = 15 3s ds = 15t3
2 R 0
t
s 0
E(Wt7 ) = 7(7−1) E(Ws7−2 )ds = 0
2 R 0
t
E(Wt8 ) = 8(7−1)
2 0
E(Ws8−2 )ds = 105t4
Var(Yt ) = E(Yt2 ) − (E(Y ))2 = 15t3 + 105t6 − (−3t3 )2 = 15t3 + 96t6
3.
4.
5.
6.
7.

1 t t 1 t X∞ − Xt t t
dYt = Xt e τ dt + e τ dXt = Xt e τ dt + e τ dt + σe τ dWt =
τ τ τ
X∞ t t
= e τ dt + σe τ dWt
τ
Z t Z t Z t
X∞ u u t u
Y t = Y0 + e τ du + σe dWu = X0 + X∞ (e − 1) +
τ τ σe τ dWu
0 τ 0 0

Z t Z t
− τt − τt t u
− τt − τt u−t
Xt = e Yt = e (X0 +X∞ (e −1)+ τ σe dWu ) = X0 e
τ +X∞ (1−e )+ σe τ dWu
0 0

E(Xt ) = X0 e− τ + X∞ (1 − e− τ )
t t

Z t Z t Z t
u−t u−t 2u−2t τ σ2 −2t
Var(Xt ) = Var( σe dWu ) =
τ E(σe ) du =
τ
2
σ 2 e τ du = (1 − e τ )
0 0 0 2

3 2010-2011
3.1 Exam, 12.01.2011
Notation: Wt is the standard Wiener process.
Part A (10 points each problem). Time allowed: 120 minutes.
1. The joint probability density function of X and Y is given by
(
x + y, x ∈ [0; 1], y ∈ [0; 1]
f (x, y) =
0, otherwise

Find E(Y | X) in terms of X, R t find the probability density function of E(Y | X)

2. Consider the process Xt = 0 sWs dWs . Find E(Xt ), Var(Xt ), Cov(Xt , Wt )
3. The process Yt is given by Yt = 2Wt + 5t. The stopping time τ is given by τ = min{t | Yt2 =
100}. Find the distribution of the random variable Yτ and the expected value E(τ ).
Hint: you may find the martingales aYt and Yt − f (t) useful
4. Find P(W2 − W1 > 2)
5. In the framework of Black and Scholes model find the price of an asset which gives you the
payoff of 1 rubble only if the final price St is at least two times bigger than the initial price S0
of the asset.

10
 6. Consider the free end problem, where T > 0 is not given
Z T
(ẋ2 − x + 1)dt → extr
0

At the left end x(0) = 0 Find the optimal T value and the extremal. Check that you solved
the optimality problem or show the opposite.
Part B (20 points each prolem). Time allowed: 60 minutes.
7. Consider the stochastic differential equation
p p
dXt = ( 1 + Xt2 + 0.5Xt )dt + 1 + Xt2 dWt , X0 = 0

(a) Suppose that Yt is another process that depens only on Xt , i.e. Yt = f (Xt ). Find dY
using the Ito’s lemma.
(b) Find such function f that the term before dW in dY is constant.
(c) Find Xt
(d) Sketch P(Xt > 0) as the function of t.
8. Consider the neoclassical optimal growth model
Z ∞  
−rt 1
e Ū − dt → max
0 c(t)

subject to k̇ = A ln(1 + k) − c − δk, where A > r > δ > 0, k(0) = k0 , Ū > 0.

(a) Derive necessary conditions, using the current value Hamiltonian
(b) Sketch the phase diagram for this problem with the capital intensity and consumption
labeled on the horizontal and vertival axes, respectively
(c) Check that the steady state solution exists. Provide explanation.
(d) Explore the stability of the steady state, using the Jacobian
(e) Why are you sure the found growth path maximizes the discounted stream of utility?

3.2 Exam-solutions
R1
1. Let’s start, f (x) = 0 f (x, y) dy = x + 1/2 for x ∈ [0; 1]. The conditional density, f (y | x) =
f (x, y)/f (x) = (x + y)/(x + 1/2) for x and y in [0; 1].
The conditional expected value:
Z 1
3X + 2 1 1
E(Y | X) = yf (y | X) dy = = +
0 6X + 3 2 12X + 6
We also need to find the density of Ŷ = 6X+3 3X+2
.
Let’s first find the probability distribution function:
   
1 1 2 − 3t
FŶ (t) = P(Ŷ ⩽ t) = P + ⩽t =P X⩾ = 1 − FX (a),
2 12X + 6 6t − 3
Ra
where a = 2−3t6t−3
and F X (a) = 0
f (t) dt.
Take the derivative with respect to t to obtain:

11
  
da 2 − 3t 1 −3
fŶ (t) = −fX (a) =− +
dt 6t − 3 2 (6t − 3)2
2. First, E(Xt ) = 0. Using Ito’s isometry:
Z t Z t
Var(Xt ) = E(s Ws ) ds =
2 2
s3 ds = t4 /4
0 0

Again using Ito’s isometry:

Z t Z t  Z t
Cov(Xt , Wt ) = Cov sWs dWs , 1 dWs = E(sWs · 1) ds = 0
0 0 0

3. First we find distribution of Yτ :

Yτ −10 10
p 1−p
We introduce Zt = aYt . To make Zt martingale we should eliminate dt part in dZt . Hence
ln a = −2.5. And Zt = e−2.5Yt is a martingale.
Zτ e25 e−25
p 1−p
From Doob’s theorem: E(Zτ ) = Z0 = 1. And we obtain p from linear equation

pe25 + (1 − p)e−25 = 1
The process Rt = Yt − 5t is a martingale. By Doob’s theorem E(Rτ ) = R0 = 0, but

E(Rτ ) = E(Yτ ) − 5 E(τ )

So,
E(τ ) = E(Ytau )/5 = −2p + 2(1 − p) = 2 − 4p
4. We know that W2 − W1 ∼ N (0; 1), therefore
P(W2 − W1 > 2) = 1 − Φ(2) = 1 − 0.9772 = 0.0228
5. (
1 if SST0 > 2,
XT =
0 otherwise
So,
      
−rT −rT ST −rT σ2
X0 = e EP̃ (XT | F0 ) = e P̃ > 2 | F0 =e P̃ exp r − T + σ W̃T > 2 | F0
S0 2
 
 2
 ln 2 − r − σ2
T
σ 2
σ W̃T > ln 2 − r − T ⇒ W̃T >
2 σ
√
We want standard normal distribution of W̃T , so we standardize by T:
 
σ2
W̃ ln 2 − r − 2
T
√T > √
T σ T

12
 So,
        
σ2 σ2
W̃T ln 2 − r − T ln 2 − r − T
X0 = e−rT P̃  √ >  = e−rT 1 − Φ  
2 2
√ √
T σ T σ T

3.3 Retake, ⁇.01.2011

Answer all SIX questions of this section. Each question is worth 10 points.
1. The joint distribution of the random vector (X, Y ) is given by its p.d.f
 x−y
ce , for 0 ⩽ x, y ⩽ 1
f (x, y) =
0, otherwise

where c is a normalization constant. Find E(X | Y ).

2. Let Yt = Wt3 − tWt4 . Find E(Yt ) and Var(Yt ). You don’t have to use Ito Lemma here.
3. Let Xn be a discrete time stochastic process that converges in probability to a random number
X as n → ∞. Does this condition imply that Xn converges to X in mean? Almost surely?
In distribution? Support at least one of your answers with a proof or counterexample. Give a
definition for each type of convergence.
R1 √
4. Solve the calculus of variations problem to optimize the integral 1/2 1 + ẋ2 /x dt → max
√
subject to the conditions x(1/2) = 3/2, x(1) = 1. Justify your answer referring to the
sufficiency conditions. What kind of extremum did you find?
5. Let x(t) represent the revenue of a firm. The fraction of it, namely xu(t), where 0 ⩽ u ⩽ 1,
is spent on investments allowing the revenue to grow according to the rule ẋ = αxu with
α = const > 0.
Another fraction of the revenue βx, β = const > 0 serves to reimburse the costs. Maximize
the profit of the firm over the finite time horizon
Z T
(1 − β − u)x dt → max
0

subject to ẋ = αxu, x(0) = x0 , 0 ⩽ u ⩽ 1.

6. Consider the following «cake-eating» problem: maximize

X
∞
β t ln ct
0

subject to Wt+1 = Wt − ct , Wt ⩽ W , W > 0, 0 < β < 1. Let the state variable be W and
denote the next period value of W as W ′ .
(a) Write down the Bellman equation for the value function V (W )
(b) Using method of undetermined coefficients find V (W )
(c) Find the optimal policy function c = h(W )
Part B. Answer both questions from this section. Each question is worth 20 points. Part B lasts for
60 minutes.
7. Let Xt be a stochastic process such that dXt = X∞τ−Xt dt + σdWt , where X∞ and τ are non-
random constants, Wt is a Wiener process, and let Yt = Xt et/τ .

13
 (a) Use Ito Lemma to find both differential and integral expressions for Yt and use them to
express Xt (a.s.) in terms of X∞ , X0 , τ , σ and t. Here X0 is the value of Xt at time t = 0.
(b) Find E(Xt ) and Var(Xt ). Sketch the graph of E(Xt ) as a function of t for X∞ = 1, τ = 1,
and X0 = 0, 1, and 2. Plot a possible trajectory of Xt in each case. Is there any name or
names associated with Xt ?
8. Consider the profit maximizing problem over the infinite time horizon
Z ∞ √
e−rt (2 K − cI) dt → max
0

where K is capital, I is investment, c = const is the unit cost of investment, r is discount rate.
Let K(0) = K0 > 0. The capital changes under the investment equation K̇ = I − bK, where
b > 0 is the depreciation rate. Investment is bounded 0 ⩽ I ⩽ I. ¯ Suppose the parameters of
the problem satisfy conditions
1 I¯
K0 < 2 <
c (r + b)2 b
(a) Derive necessary conditions, using the current value Hamiltonian
(b) Does the steady-state solution(s) (Ks , Is ) exist? Explain. If you answer is positive ex-
plore its stability.

3.4 Retake-solutions
1. Let’s go: (
ce(x−y) , 0 ⩽ x, y ⩽ 1
f (x, y) =
0, else
First, we need to find constant c.
R1R1 R1 1
R1 1
1 = 0 0 (cex−y )dxdy = 0 (−cex−y ) dx = 0 (−cex−1 + cex )dx = −cex−1 + +cex =
0 0
−c + ce + ec − c = 1
From this simple equation we get c = (e−1)
e
2

So the problem now looks like:

(
e
(e−1)2
e(x−y) , 0 ⩽ x, y ⩽ 1
f (x, y) =
0, else
R1
E(X | Y ) = 0 xf (x | y)dx, where f (x | y) = ff(x,y)
R1 −y
(y)
e
f (y) = 0 f (x, y)dx = (e−1)2 (e 1−y
−e )
ex
f (x | y) = e−1
R 1 xex R1 1
E(X | Y ) = 0 e−1 1
dx = e−1 0
1
xdex = e−1 (xex − ex ) = 1
e−1
0
So the final answer will be E(X | Y ) = e−1 1

2. E(Yt ) = E(Wt − tWt ) = E(Wt ) − E(tWt4 ) = E(Wt3 ) − t E(Wt4 ) = 0 − 3t3 = −3t3 , as

3 4 3
Rt
E(Wt3 ) = 3(3−1) E(Ws3−2 )ds = 0
2 R 0
t Rt
E(Wt4 ) = 4(4−1)
2 0
E(W 4−2
s )ds = 6 0
sds = 3t2
Var(Yt ) = E(Yt ) − (E(Y ))
2 2

14
 E(Yt2 ) = E(Wt3 − tWt4 )2 = E(Wt6 − 2tWt7 + t2 Wt8 ) = E(Wt6 ) − 2t E(Wt7 ) + t2 E(Wt8 ) =
15t3 + 105t6 , Rt
as E(Wt5 ) = 5(5−1)
2R
E(Ws5−2 )ds = 0
t
0 Rt
E(Wt6 ) = 6(6−1) E(Ws6−2 )ds = 15 0 3s2 ds = 15t3
2 Rt0
E(Wt7 ) = 7(7−1) E(Ws7−2 )ds = 0
2 R 0
t
E(Wt8 ) = 8(7−1)
2 0
E(Ws8−2 )ds = 105t4
Var(Yt ) = E(Yt2 ) − (E(Y ))2 = 15t3 + 105t6 − (−3t3 )2 = 15t3 + 96t6
3.

3.5 Reretake, 08.02.2011

Stochastic calculus part
1. (10 pts) The joint density of random variables X and Y is given by f (x, y) = x+y for x ∈ [0; 1]
and y ∈ [0; 1]. Find E(Y | X) and Var(Y | X)
2. (10 pts) The random variables Xi are independent and uniformly distributed on [0; 2011]. We
define Yn = (X1 · X2 · . . . · Xn )1/n . In what sense does the sequence Yn converge? What is
the limit? Find the expected value and the variance of the limit.
3. (10
√ pts) Let W1 (t) and W2 (t) be two independent Wiener processes. Consider Yt = aW1 (t) +
1 − a2 W2 (t). Is Yt a Wiener process?
4. (20 pts) Consider the Vasicek interest rate model, dRt = a(b − Rt ) dt + s dWt , where a, b and
s are positive constants.
(a) Using the substitution Yt = eat Rt find the solution of the stochastic differential equation;
(b) Find E(Rt ) and Var(Rt ).

3.6 Reretake, 08.02.2011, solution

1.
2. Consider Rn = ln Yn , then Rn is the average of ln X1 , …, ln Xn . By the law of large numbers
Rn converges to E(ln X1 ) in probability and almost surely.
3. Yes, Yt is a Wiener process. We just check all properties: independence of increments, station-
arity of increments, Y0 = 0, continuity of path, normality of increments.
4. Start with Ito’s lemma:

dYt = aeat Rt dt + eat dRt = aeat Rt dt + eat (a(b − R) dt + s dWt ) = eat (ab dt + s dWt )

So, Z Z
t t
au
Yt = Y0 + e abdu + seau dWu ;
0 0

4 2011-2012
4.1 Retake, 03.02.2012
Section A. 10 points for each problem.
1. Solve the bounded control problem with the free end value where T > 1 is specified in advance
but x(T ) is not.
Z T
(x − u) dt → max, x(0) = 1
0

15
 x′ = u, 0 ⩽ u ⩽ x. What guarantees that a maximizer would be found?
2. Two stochastic processes are defined by the system of SDE with the Brownian motions W1t ,
W2t independent of each other
(
dXt = (2 + 5t + Xt )dt + 3dW1t
dYt = 4Yt dt + 8Yt dW1t + 6dW2t

Calculate d(Xt Yt ) Rt
3. Find the value of the constant a such that the process Xt = Wt3 − a 0 Ws ds is a martingale.
4. The process Yt is given by Yt = 2Wt + 5t. The stopping time τ is given by τ = min{t | Yt2 =
100}. Find the distribution of the random variable Yτ and the expected value E(τ ).
Hint: you may find the martingales aYt and Yt − f (t) useful
5. What is the expected value and variance of Wt2 for t > s given that Ws = x?
6. In the framework of Black and Scholes model find the price at the time 0 of an asset which
gives you the payoff max{ln(St ), 0} at the time t. Here St is the price of the underlying asset.
Section B. 20 points for each problem.
7. The goal of this exercise is to solve the SDE
1
dXt = dt + Xt dWt
Xt
with initial condition X0 = 1.
(a) Apply the Ito’s lemma to the process Yt = exp(f (t) + g(Wt ))Xt2
(b) Find non-constant functions f and g such that the coefficient before dWt in the expres-
sion for dYt is zero.
(c) Find Xt . The final expression may contain a Riemann integral of some stochastic process.
8. A typical firm with the production function x = f (l) in an economy employs 1 unit of the
capital paying for it r̄. There are n identical firms in this economy where n(t) is a function of
time. Let the output of a firm be x(t). Then the aggregate supply equals nx. Assuming that
the market is in the equilibrium we use the inverse demand function p(nx), where p′ (y) < 0.
Equation defining the dynamics of labor is given by dt dl
= a(w − w̄), where a > 0 and w̄
is some equilibrium wage rate. The growth (or fall) in number of firms is governed by the
equation dn dt
= b(px − w̄l − r̄), where b > 0. Write down the system of ODE for the unknowns
l(t), n(t), using the first-order condition for the profit-maximizing firm, find and classify the
steady-state solutions (if any exist). Production function is twice differentiable and concave
everywhere.

4.2 Retake-solutions
1.
2.
1
d(Xt Yt ) = 0 dt + Xt dYt + Yt dXt + dXt dYt
2
Here W1t and W2t are independent and dW1t dW2t = 0.
3. Using Ito-Doeblin lemma, dWt3 = 3Wt2 dWt + 12 6Wt dt. Hence,

dXt = 3Wt2 dWt + (3 − a)Wt dt

16
 Ito’s integral is always a martingale, so a = 3.
4. The distribution of Yτ is given by the table
y −10 10
P(Yτ = y) 1 − p p
First we find two martingales, Zt = Yt − 5t, and Xt = exp(−2.5Yt ). Here Ito’s lemma is
useful.
According to Doob’s theorem E(Xt ) = X0 = 1 and we have an equation for p:

p · exp(−25) + (1 − p) · exp(25) = 1
We use once again Doob’s theorem, E(Zt ) = Z0 = 0, so

E(Yτ − 5τ ) = 0
The value of E(τ ) may be calculated as E(τ ) = E(Yτ )/5 = 10p−10(1−p)
5
.
5. Let’s first consider the unconditional version of the problem.

E(Wt2 ) = E((Wt − W0 )2 ) = V ar(Wt − W0 ) + E(Wt − W0 )2 = t − 0 = t

V ar(Wt2 ) = E(Wt4 ) − E(Wt2 )2 = E(Wt4 ) − t2
In order to compute E(Wt4 ) we use the following formula:
Z
k k(k − 1) t
E(Wt ) = E(Wsk−2 )ds
2 0

So, Z Z
t t
E(Wt4 ) =6 E(Ws2 )ds =6 sds = 3t2 ⇒ Var(Wt2 ) = 3t2 − t2 = 2t2
0 0
However we know that Ws = x. Using the properties of Wiener process we may assume that
the timer is started again from 0. So we define Wt′′ = Wt′ +s − x.
So,

E(Wt2 | Ws = x) = E((Wt′′ + x)2 ) = E(Wt′2′ + 2xWt′′ + x2 ) = t′ + x2 = (t − s) + x2

And

Var(Wt2 | Ws = x) = Var((Wt′′ + x)2 ) = Var(Wt′2′ + 2xWt′′ + x2 ) =

= Var(Wt′2′ ) + 4x2 Var(Wt′′ ) + 0 = 2(t − s)2 + 4x2 (t − s) (1)

5 2012-2013
5.1 Midterm, 13.11.2012
1. (25 points) Solve the calculus of variations problem (use necessary conditions)
Z π/2
(x2 + ẋ2 − xẋ + 4x cos2 t) dt → extr,
0

where x(0) = −6/5, x(π/2) = −4/5.

17
 What kind of a sufficient condition is applicable to this problem and why?
2. (25 points) Solve the control problem with the free end value
Z T
(−u2 /2 − x) dt → max,
0

where x(0) = x0 > 0, x(T ) is unknown, ẋ = u2 x.

Can you apply any of the sufficiency theorems to prove that the maximizer has been found?
Why we are sure in the validity of the solution? P
3. (20 points) Consider an optimal growth problem ∞ t=0 5t ln ct → max, subject to kt+1 =
1
7/8
kt − ct , k(0) = k0 > 0.
(a) State the Bellman equation
(b) Use the iteration method to find the first two iterations of the value function. Set V0 (k) =
0.
(c) Use the guess-and-verify method to find the value function.
4. (30 points) Consider a profit-maximizing firm over the infinite horizon planning period
Z ∞
(pf (K, I) − cK − gI)e−rt dt → max,
0

subject to K̇ = I − δK, K(0) = K0 > 0.

The production function f (K, I) = (1 − αI 2 ) ln K, where parameter α is so small that f
takes only positive values within the reasonable values of the gross investment rate I which
can take any sign (investment/disinvestment). Moreover, the production function is concave
everywhere. Using the current value Hamiltonian derive the system of differential equations
in K(t) and I(t). Prove that the steady-state solution exists. By calculating the Jacobian at the
equilibrium prove that (Ks , Is ) is the saddle point. All exogeneous parameters in this model
are positive.

5.2 Exam, 10.01.2013

Optimal control part.

R1
1. [10 points] Solve the optimal control problem 0 (x + u) dt → max subject to ẋ = −x + u + t,
x(0) = 0, x(1) is free, and 0 ⩽ u ⩽ 1.
2. [10 points] Consider an intertemporal utility maximization problem over the finite horizon
in discrete time t = 1, . . . , T . The utility is u(c) = ln c, the initial wealth is w and if the
remaining wealth at time t was wt , then by the beginning of the next period it becomes wt+1 =
(1 + r)(wt − ct ), where r is the discount rate and ct is the consumption. To find the maximum
value of the utility stream one can use the finite version of the Bellman equation written in
the form Vt (w) = max{u(c) + Vt+1 ((1 + r)(w − c))} where the function within the braces is
maximized over the values of 0 ⩽ c ⩽ w. It is important to note that the value function Vt is
the maximum value of the utility stream from time t onwards.
(a) [3 points] Let VT = ln w. Find VT −1 .
(b) [7 points] Try to verify the conjecture that Vt = γt ln(1 + r) + (T − t + 1) ln T −t+1
w
where
γt is some (unknown) function of time.

18
 3. [20 points] A spill of toxic substance should be cleaned up by a company.
Rt If a cleanup rate
is u(t), then the area of the spill shrinks by the law x(t) = x0 − 0 u(s) ds , where x0 is the
initial area. According to the government contract by the time T (and not earlier) the area
of the spill should be reduced to xT , where xT < x0 . Clearly u(t) ⩾ 0, and by technology
requirement u(t) ⩽ ū, where ū > (x0 − xT )/T . Costs of cleaning are directly proportional to
u(t). Let the unit costs equal c > 0 and the costs are discounted with the discount rate r > 0.
(a) [5 points] Formulate the minimization problem with the bounded control. Hint: this is
a fixed endpoints problem.
(b) [15 points] By using the current value Hamiltonian set the system of equations and solve
it.

Stochastic calculus part. Here Wt always denotes the standard Wiener process.

4. [10 points] Waves are arriving on the seashore. The sizes of the waves are independent uniform
on [0; 1] random variables Xi . If the size of i-th wave is greater than the size of its neighboring
waves, Xi > max{Xi−1 , Xi+1 }, then we call it a «high» wave. Let Hi be the indicator of the
«high» waves: Hi = 1 if Xi > max{Xi−1 , Xi+1 } and Hi = 0 otherwise.
Find E(Xi | Hi ) and E(Hi | Xi )
5. [10 points] Find E(Ws · Wt ) and E(Wr R· Ws · Wt ) where r < s < t.
t
6. [10 points] Consider the process Xt = 0 s2 Ws dWs . Find E(Xt ), Var(Xt ), Cov(Xt , Wt )
7. [10 points] In the framework of Black and Scholes model find the price at the time 0 of an
asset which gives you the payoff ln(ST ) at the time T . Here St is the price of the underlying
asset at time t.
8. [20 points] The goal of this exercise is to solve the SDE

dXt = −9Xt2 (1 − Xt )dt + 3Xt (1 − Xt )dWt , X0 = 1/2

(a) Using Ito’s lemma find dYt for the process Yt = f (Xt )
(b) Find such a function f () that the term before dt in dYt is zero
(c) Obtain a simple differential equation for Yt . It should not contain Xt .
(d) Solve the stochastic differential equation for Yt
(e) Express Xt as a function of Wt and t.

5.3 exam 10.01.2013-hints

1. E(Hi | Xi ) = Xi2 ,
3
E(Xi | Hi = 1) = E(Xi | Xi is the biggest of three waves) =
4

1 1
E(Xi | Hi = 0) = E(Xi | Xi is the smallest of three waves)+ E(Xi | Xi is the median of three waves
2 2
E(Xi | Hi ) = (3 + 3Hi )/8

19
 2. E(Ws · Wt ) = s and
Let’s introduce deltas: D1 = Ws − Wr and D2 = Wt − Ws . The bonus of this notation is that
D1 , D2 and Wr are independent random variables. And

E(Wr ·(Wr +D1 )·(Wr +D1 +D2 )) = E(Wr3 +Wr2 D1 +Wr2 D2 +Wr2 D1 +Wr D12 +Wr D1 D2 ) = 0+0+0+0+0+

3. E(Xt ) = 0, Var(Xt ) = t6 /6, Cov(Xt , Wt ) = 0

4. C0 = e−rT (ln S0 + (r − σ 2 /2)T
5. Equation for f has the form f ′′ (x)(1 − x) − 2f ′ (x) = 0. Use the substituion f ′ (x) = h(x).
This ordinary differential equation has many solutions, one of them is f (x) = 1/(1 − x). We
may take any function f that makes initial stochastic differential equation less diffucult, so
we take f (x) = 1/(1 − x), and Yt = 1/(1 − Xt ).
And
dYt = 3Xt /(1 − Xt ) dWt
We may notice that Xt /(1 − Xt ) = 1/(1 − Xt ) − 1 = Yt − 1, so

dYt = 3(Yt − 1) dWt

We use the substitution Zt = Yt − 1, so dZt = 3Zt dWt and then the substitution Qt = ln Zt .

5.4 Retake, 08.02.2013

Stochastic calculus part. Here Wt denotes standard Wiener process

1. (10 points) Random variables X and Y are jointly normal with zero expected values, unit
variances and correlation ρ. Find E(Y | X)Rand E(Y 2 | X) Rt
t
2. (10 points) Consider the processes Xt = 0 s3 Ws dWs and Yt = 0 Ws dWs . Find E(Xt ),
Var(Xt ), Cov(Xt , Yt )
3. (10 points) The stochastic process Yt is given by equation Yt = Wt4 − 6tWt2 + 3t2 . Find dYt
and E(Y7 | Y2 )
4. (20 points) Consider the stochastic differential equation

dXt = Xt dt + Xt dWt , X0 = 1

(a) Solve this stochastic differential equation

(b) Let τ = inft>0 {t : Xt ⩾ R}. Find E(τ ) applying the optional stopping theorem to the
process Wt .

Optimal control part.

R1
5. (10 points) Solve the optimal control problem: 0 u2 dt → min, subject to ẋ = x + u, x(0) = 1.
6. (10 points) Consider a maximization problem over the finite horizon in discrete time t =
1, . . . , T :
X T  3 
t t2
+ a2t → max
t=1
3 2

20
 P
subject to constraint Tt=1 at = c, at ⩾ 0.
P
By introducing wt = ti=1 ai reduce that problem to dynamic programming (3 points) and
solve it (7 points).
7. (20 points) Consider the profit-maximizing problem for a representative competitive firm
Z ∞
(p − c(x(t)))q(t)e−rt dt → max
0

subject to (*) ẋ = 1 − x − q, where the state variable x(t) < 1 represents a nonrenewable
stock resource (oil) that depletes according to the equation (*) and q(t) is the extraction rate.
Here c(x) is a cost function of the extraction that is defined by c(x) = e−x . The price of oil
is assumed to be constant and equal p where 1/e < p < 1. The optimization problem is to
choose q(t) to maximize the discounted profits, 0 < r < 1.

(a) (10 points) Derive necessary conditions.

(b) (10 points) Prove that the steady-state exists.

5.5 retake — short hints

1. First we decompose Y as Y = aX +bZ, where Z is independent from X and has unit variance.
To obtain a and b we make system:
(
Var(Y ) = a2 Var(X) + b2 Var(Z) = 1
Cov(Y, X) = a Cov(X, X) = ρ
p p
So, a = ρ and b = 1 − ρ2 . The variable Y is decomposed as Y = ρX + 1 − ρ2 Z. Then
p
E(Y | X) = E(ρX + 1 − ρ2 Z | X) = ρX + 0
And
p
E(Y 2 | X) = E(ρ2 X 2 + (1 − ρ2 )Z 2 + 2ρ 1 − ρ2 XZ | X) = ρ2 X 2 + (1 − ρ2 ) · 1 + 2 · 0

2. E(Xt ) = E(Yt ) = 0, for variance and covariance we use isometry property

Z t  Z t Z t
t8
Var(Xt ) = Var 3
s Ws dWs = E(s Ws ) ds =
3 2
s7 ds =
0 0 0 8

Z t Z t  Z t Z t
t5
Cov(Xt , Yt ) = Cov 3
s Ws dWs , Ws dWs = E(s 3
Ws2 )ds = s4 ds =
0 0 0 0 5

3. Calculate dYt . We observe that dYt = (4Wt3 − 12tWt ) dWt , so Yt is a martingale and E(Y7 |
Y 2 ) = Y2 .
4. Use the substitution Yt = ln Xt . Then, using Ito’s lemma we obtain:

1 1
dYt = dXt − (dXt )2
Xt 2Xt2

21
 Initial SDE gives us dXt = Xt dt + Xt dWt , so we can compute also (dXt )2 and plug it in the
dYt .
dXt2 = (Xt dt + Xt dWt )2 = Xt2 dt
Recall that dWt · dWt = dt, dWt · dt = 0 and dt · dt = 0
Plugging in we get:

Xt dt + Xt dWt Xt2 dt dt dt
dYt = − 2
= dt + dWt − = dWt +
Xt 2Xt 2 2

The full form for Yt :

Z t Z t
ds t
Yt = Y0 + dWs + = Y0 + Wt + = ln Xt
0 0 2 2

Now we are ready to write the solution:

t t
X t = e Y 0 e Wt e 2 = X 0 e Wt + 2

6 2013-2014
6.1 Stochastic calculus hometask
1. Let X and Y be independent exponentially distributed with parameter λ. Find E(X + Y |
X − Y ).
2. Let Sn be symmetric random walk with S0 = 2013. The stopping time τ is the first moment
when |Sn | reaches the value 2014 or the value 2000. Find E(τ Sτ ).
Hint: You may construct a martingale of the form Mn = Sn3 − f (n)Sn for some function f .
3. Conditional variance is defined as Var(Y | X) = E(Y 2 | X) − (E(Y | X))2 . Find Var(Ws |
Wt ).
Hint: do not forget two cases, t > s and s < t. You may find inversion property of the Wiener
process useful.
4. Let f be a real function such that f ′′ is continuous. Find all such functions f that Xt =
exp(αt)f (Wt ) is a martingale.
5. Solve the stochastic differential equation
p
dXt = (Xt /t + t)dt + 2 tXt dWt
Hint: You may suppose without a proof that the solution has the form Xt = f (t)g(Wt ).
6. In the framework of Black and Scholes model find the price of the asset, which pays you ST
at the fixed moment of time T > 1 if ST −1 > 1 and 0 otherwise.

6.2 Solution for stochastic calculus hometask

Author: Vladimir Shmarov
Problem 1.
Denote ξ = X + Y and η = X − Y .
Then we are gonna find E(ξ | η).
Since X ∼ Expo(λ), it has a nice density: pX (u) = λe−λu · I{u⩾0} .
Similarly, pY (v) = λe−λv · I{v⩾0} .

22
Since X and Y are independent, the vector (X, Y ) has a two-dimensional density, which equals to
the product of densities of X and Y , i.e.

p(X,Y ) (u, v) = I{u⩾0,v⩾0} λ2 e−λ(u+v)

ξ+η ξ−η
We have X = X(ξ, η) = 2
and Y = Y (ξ, η) = 2
.
∂X ∂X
1 1
∂ξ ∂η =
Then J = ∂Y 2 2 = − 21 .
∂Y 1
2
− 12
∂ξ ∂η
Therefore
 
s+t s−t λ2 λ2
p(ξ,η) (s, t) = p(X,Y ) , · |J| = I{s+t⩾0} I{s−t⩾0} e−λs = I{s⩾|t|} e−λs
2 2 2 2

is a density of a vector (ξ, η).

Further, Z
λe−λ|t|
pη (t) = p(ξ,η) (s, t)ds =
R 2
Then the conditional density is

p(ξ,η) (s, t)
p(ξ|η) (s | t) = = λe−λ(s−|t|) · I{s−|t|⩾0}
pη (t)

And finally, denoting z = s − |t|, we get

Z Z
E(ξ | η) = s · p(ξ|η) (s | t)ds = (z + |t|) · λe−λz · I{z⩾0} dz =
R t=η R t=η
Z Z  
−λz −λz 1 1
= z · λe · I{z⩾0} dz + |t| ·λe · I{z⩾0} dz = + |t| = + |η| (2)
R t=η R t=η λ t=η λ

(the first integral is just the expectation of exponential distribution with parameter λ,
and the second is |t| times the integral of the density, which equals 1)

Answer: E(X + Y | X − Y ) = λ1 + |X − Y |
Problem 2 — part 1 of 2.
P
At first, Sn = S0 + ni=1 εi , where {εi } is the sequence of independent random variables with
P(εi = −1) = P(εi = 1) = 21 . Also denote Fn = σ(ε1 , . . . , εn ).
Observe that τ is really a stopping time, because at each moment of time we precisely know, are we
in one of the points {2000, 2014} or not.
Let’s prove that E(τ ) < ∞ (which will also imply that P(τ = +∞) = 0).
We shall obviously reach τ after any 14 consecutive steps in one direction. Then

P (τ ⩾ 14n) ⩽ P(At least one of ε1 , . . . , ε14 equals −1) ×. . .× P(At least one of ε14(n−1)+1 , . . . , ε14n equals −

n
= 1 − 2−14 = q n (3)

where q < 1.

23
Then
X
+∞ X
+∞ X
∞
E(τ ) = P (τ ⩾ n) ⩽ 14 P (τ ⩾ 14n + 1) ⩽ 14 q n = 14 · 214 < ∞
n=1 n=0 n=0

as required.
The sequence {St } is a martingale w.r.t. filtration {Fn }, because it is obviously {Ft }-adapted, has
a finite (zero) expectation and

E(St+1 | Ft ) = E(St + εt+1 | Ft ) = St + E(εt+1 | Ft ) = St

Further, Smin{t,τ } is bounded in t: it always lies between 2000 and 2014. The Doob’s theorem condi-
tions are satisfied, which means that

E(Sτ ) = S0 = 2013

But E(Sτ ) = 2000 · P(Sτ = 2000) + 2014 · P(Sτ = 2014), which after the substitution E(Sτ ) = 2013
gives us
P(Sτ = 2000) = 1
14
and P(Sτ = 2014) = 13
14

Further, let’s prove that the sequence {St3 − 3tSt } is a {Ft }-martingale . At first, it is obviously
adapted and |E(St3 − 3tSt )| < ∞
Then


E(St+1
3
− 3(t + 1)St+1 | Ft ) = E (St + εt+1 )3 − 3(t + 1)(St + εt+1 ) | Ft =
 
E St + 3St εt+1 + 3(εt+1 − 1)St − 3tSt + εt+1 − 3(t + 1)εt+1 | Ft = St3 − 3tSt (4)
3 2 2 3

(the underlined terms are going out after taking conditional expectation; (ε2t+1 − 1) goes out even
without it)
Denote Xt = St3 − 3tSt .
For t ⩾ τ we have |Xmin{t+1,τ } − Xmin{t,τ } | = 0, and for t < τ we have
|Xmin{t+1,τ } − Xmin{t,τ } | = |Xt+1 − Xt | = |3St2 εt+1 − (3t + 2)εt+1 | ⩽ 3 × 20142 + 3τ + 2, which
expectation is bounded by the finite number 3 × 20142 + 2 + 3E(τ ).


Then, the overall expectation E |Xmin{t+1,τ } − Xmin{t,τ } | ⩽ 3 × 20142 + 2 + 3E(τ ) for every t, i.e.
is bounded in t.
Problem 2 — part 2 of 2.
Then again, the conditions of the Doob’s theorem are satisfied, i.e.

E(Xτ ) = X0 = 20133

But also E(Xτ ) = E(Sτ3 ) − 3 E(τ Sτ ), i.e.

 
1
1 1 13
E(τ Sτ ) = E(Sτ ) − 2013 =
3 3
2000 + 2014 − 2013 = 13 × 2009 = 26117
3 3 3
3 3 14 14

Answer: E(τ Sτ ) = 26117.

Problem 3.
Case 1. Let s ⩾ t.

24
Then



E Ws2 | Wt = E Wt2 + 2Wt (Ws − Wt ) + (Ws − Wt )2 | Wt =


= Wt2 + 2Wt E(Ws − Wt ) + E (Ws − Wt )2 = Wt2 + (s − t)

Further,
E(Ws | Wt ) = Wt + E(Ws − Wt | Wt ) = Wt
Then we have Var(Ws | Wt ) = E(Ws2 | Wt ) − E(Ws | Wt )2 = (s − t).
Case 2. Let s < t.
By the inverse property, Vt = tW 1 is also a standard Wiener process.
t
From the case 1 we have
     
1 1 1 1 1 1 1
− = Var V 1 | V 1 = Var Ws | Wt = 2 Var Ws | Wt = 2 V ar(Ws | Wt )
s t s t s t s t s


the last equation is because obviously σ(Wt ) = σ 1t Wt .
2
Then Var(Ws | Wt ) = s − st .

Summing up these two cases, we get the

n o
s2
Answer: Var(Ws | Wt ) = max s − t, s − t
Problem 4.
Let g(t, Wt ) = Xt = eαt f (Wt ), then gt = αg, gWt = eαt f ′ (Wt ) and gWt Wt = eαt f ′′ (Wt ).
By the Ito’s lemma we have
1
dXt = αXt dt + eαt f ′ (Wt )dWt + eαt f ′′ (Wt )dt =
  2
1
= eαt αf (Wt ) + f ′′ (Wt ) dt + eαt f ′ (Wt )dWt
2

Therefore
Z t   Z t
1 ′′
Xt = f (0) + e αu
αf (Wu ) + f (Wu ) du + eαu f ′ (Wu )dWu
0 2 0

The last term in the right-hand side is a martingale, so Xt is a martingale if and only if the coefficient
near dt equals zero for all t, i.e.
1
αf (Wt ) + f ′′ (Wt ) = 0 a.s.
2
which (together with continuity of f ′′ ) implies
1
αf (x) + f ′′ (x) = 0 ∀x
2
We know how to solve the linear differential equations, so I will not go into details and I’ll just write
the answer.

25
 √ √
Answer: If α < 0, then f (x) = C1 et −2α + C2 e−t −2α , where C1 , C2 are any real constants;
if α = 0, then f (x) = C1 x + C√ 2; √
if α > 0, then f (x) = C1 cos(t 2α) + C2 sin(t 2α)
Problem 5 — part 1 of 2.
Assume for simplicity that Xt = f (t)g(Wt ), where f and g are kinda nice functions (defined
on [0, +∞) and continuously differentiable the required number of times on this interval; right-
differentiable at zero, of course).
By the Itô’s lemma we have
 
1
′
dXt = f (t)g(Wt ) + f (t)g (Wt ) dt + f (t)g ′ (Wt )dWt
′′
2

Substituting this into the condition of the problem and separating the parts with dt and dWt we get

 f ′ (t)g(W ) + 1 f (t)g ′′ (W ) = f (t) g(W ) + t (1)
t t t
2
p t

f (t)g ′ (Wt ) = 2 tf (t)g(Wt ) (2)

Both equations are satisfied for every t > 0 almost surely.

Step 1. Let’s prove that f (t) ̸= 0 if t > 0. Indeed, if f (t0 ) = 0, (1) gives us

f ′ (t0 )g(Wt0 ) = t0 a.s.

so
t0
g(Wt0 ) = ′
a.s.
f (t 0)

But g is continuous, so this gives us

g(x) ≡ C = Const

But then (2) gives

Xt ≡ 0 a.s.
which does not satisfy the condition of the problem.
Therefore, f (t) ̸= 0, when t ̸= 0.

Now (2) can be transformed into

f (t) ′
g (Wt )2 = g(Wt ) (3)
4t
Step 2. g ′ (x) cannot be zero in any interval.
Suppose that g ′ (x) = 0 for all x ∈ (a, b). From (3) we have g(x) = 0 almost everywhere at (a, b)
(because P(Wt ∈ (a, b)) ̸= 0 for t > 0), and also g ′ (x) = 0 at (a, b). But substituting these values to
(1) provides us contradiction.

Let A = {x ∈ R | g ′ (x) ̸= 0}. Since g ′ is continuous and is not identical zero (from the step 2), the

26
set A has a positive Lebesgue measure; therefore P(Wt ∈ A) ̸= 0 for all t > 0.
Problem 5 — pat 2 of 2.
Now fix some t0 > 0. From (3) we have that almost everywhere in the event {Wt0 ∈ A} the equality

g(Wt0 ) f (t0 )
=
g ′ (Wt0 )2 4t0

holds. Then for almost all x ∈ A we have g(x)

g ′ (x)2
=f (t0 )
4t0
. Now fix any other t1 > 0.
We have that almost everywhere in A the equality gg(x)′ (x)2 =
f (t1 )
4t1
holds. Since A has a positive
measure, we have
f (t0 ) f (t1 )
=
4t0 4t1
i.e. f (t) = Ct for all t > 0 and some C = Const. From the continuity f (t) = Ct for all t ⩾ 0

Substitute this to (1):

1
Cg(Wt ) + Ctg ′′ (Wt ) = Cg(Wt ) + t
2
2
g ′′ (Wt ) = (4)
C
(4) holds almost everywhere, but since g ′′ is continuous, we have g ′′ (x) ≡ 2
C

Then g ′ (x) = 2x
C
+ D and g(x) = x2
C
+ Dx + E for some constants D, E.
From (3) we have
f (t) ′ 2 C ′ 2
g(x) = g (x) = g (x)
4t 4
Substitute the values of g ′ (x) and g(x):
 2
x2 C 2x
+ Dx + E = +D
C 4 C

x2 x2 CD2
+ Dx + E = + Dx +
C C 4
CD 2
which gives us E = 4
and g(x) = (Ax + B)2 , where A2 = 1
C
and 2AB = D.

Answer: Xt = t(Wt + C)2 for some constant C.

(in our notation this constant equals B
A
) √
REMARK. Precisely speaking, this is not a correct answer, because tXt is always positive , so (2)
does not always hold, because the left side of (2), which is 2t(Wt + C), can be negative with nonzero
probability. In other words, we solved similar, but different equation
  p
Xt
dXt = + t dt ± 2 tXt dWt
t

and the original problem has no solutions of the type f (t)g(Wt ).

27
Problem 6 — part 1 of 4.
The risky asset St satisfies the stochastic differential equation

dSt = µSt dt + σSt dWt

which implies    
σ2
St = S0 exp t µ − + σWt
2
The risk-free asset satisfies the equation

dBt = rBt dt

which implies
Bt = B0 ert
Let Xt be our self-financing portfolio at time t, which contains ∆t shares of the risky asset.
The total amount of money in the risky asset is ∆t , in the risk-free asset — (Xt − ∆t St ).
Therefore
dXt = ∆t dSt + r(Xt − ∆t St )dt
Let W̃t = µ−r
σ
t + Wt , and P̃ is the probability measure, w.r.t. which W̃t is a standard Wiener process
(the existence of such measure is the statement of Girsanov’s theorem).
In class we explicitly derived that
Z t
−rt
e Xt = X0 + ∆u e−ru σSu dW̃u
0

The last term is a P̃ -martingale, which means that

EP̃ e−rT XT | F0 = X0 (1)

Now recall that

XT = ST I{ST −1 >1}
so we need to calculate
ZZ
EP̃ (ST I{ST −1 >1} ) = u · p̃(ST ,ST −1 ) (u, v)dudv
u>0,v>1

where p̃(ST ,ST −1 ) is a two-dimensional density of a corresponding vector w.r.t. measure P̃ .

Problem 6 — part 2 of 4.
But
ZZ Z +∞ Z +∞  Z +∞
u·p̃(ST ,ST −1 ) (u, v)dudv = u · p̃(ST ,ST −1 ) (u, v)du dv = p̃ST −1 (v)·EP̃ (ST | ST −1 =
u>0,v>1 1 0 1

Further,
       
σ2 σ2
ST −1 = S0 exp (T − 1) µ − + σWT −1 = S0 exp (T − 1) r − + σ W̃T −1
2 2

28
    
σ2
which means that with notation A(x) = S0 exp (T − 1) r − 2
+ σx we have

p̃W̃T −1 (x) = p̃ST −1 (A(x)) · σA(x)

    
We have x = σ −1 ln A(x)
S0
− (T − 1) r− σ2
2
. Therefore

   2 !
1 1 1 σ2
p̃ST −1 (v) = ·p exp − 2 ln(v) − ln(S0 ) + (T − 1) r − (2)
σv 2π(T − 1) 2σ (T − 1) 2

Very ugly expression! Let’s believe most of those terms will cancel out.
Now
      
σ2 σ2
ST = S0 exp T r − + σ W̃T = ST −1 · exp r − + σ W̃T − W̃T −1
2 2
 
and W̃T − W̃T −1 is independent of the σ-algebra σ(ST −1 ) ⊆ FT −1 .
Then
 
  
σ2
EP̃ (ST | ST −1 = v) = EP̃ ST −1 · exp r − + σ W̃T − W̃T −1 | ST −1 = v =
2
     
σ2
= v · exp r − EP̃ exp σ W̃T − W̃T −1 (3) (5)
2

Further, if ξ ∼ N (0, 1), then

Z Z
1 2
− x2 +σx σ2 1 (x−σ)2 σ2
E(e ) = √
σξ
e dx = e 2 ·√ e− 2 dx = e 2
2π R 2π R

which gives us
   
σ2 σ2
EP̃ (ST | ST −1 = v) = v · exp r − exp = ver (4)
2 2

which, I think, is a very nice formula (especially if we compare it with (2)).

Problem 6 — part 3 of 4.
Taking (2) and (4), we have
Z +∞
p̃ST −1 (v) · EP̃ (ST | ST −1 = v) dv =
2 !
1
Z +∞  2
 
1 1 1 σ
= er ·p exp − 2 ln(v) − ln(S0 ) + (T − 1) r − dv =
1 σ 2π(T − 1) 2σ (T − 1) 2
Z +∞    2
2 !
1 1 σ
= er es · p exp − 2 s − ln(S0 ) + (T − 1) r − ds
0 2πσ 2 (T − 1) 2σ (T − 1) 2
(6)

We substituted s = ln(v). Of course, dv = es ds.

29
  
σ2
Denote also B = ln(S0 ) + (T − 1) r − 2
.

Then our integral equals

Z  
+∞
1 s2 − 2Bs − 2σ 2 (T − 1)s + B 2
e r
p exp − ds =
0 2πσ 2 (T − 1) 2σ 2 (T − 1)
!Z !
2 2
B2 (B + σ 2 (T − 1)) +∞
1 (s − B − σ 2 (T − 1))
= exp r − 2 + p exp − ds =
2σ (T − 1) 2σ 2 (T − 1) 0 2πσ 2 (T − 1) 2σ 2 (T − 1)
!
B2 (B + σ 2 (T − 1))
2



= exp r − 2 + · P̃ η > − B + σ 2 (T − 1) (7)
2σ (T − 1) 2σ (T − 1)
2

where η ∼ N (0, σ 2 (T − 1)). The integral magically turns into the probability, because under the
integral there is a well-known density.
Therefore our integral finally equals
!  
2
(B + σ 2 (T − 1)) − B 2 η B + σ 2 (T − 1)
exp r + · P̃ √ >− √ =
2σ 2 (T − 1) σ T −1 σ T −1
  
    ln(S ) + (T − 1) r+ σ2
σ (T − 1)
2
B + σ (T − 1)
2 0
= eln(S0 )+rT ·Φ  
2
= exp r + B + Φ √ √
2 σ T −1 σ T −1
(8)

Rx  2

where Φ(x) is a distribution function of a standard normal random value, Φ(x) = √1
−∞ 2π
exp − t2 dt.

Problem 6 — part 4 of 4.
And now I’ll finish this huge problem.
We had
  2


ln(S0 ) + (T − 1) r + σ2
X0 = e−rT EP̃ ST I{ST −1 >1} = e−rT · eln(S0 )+rT Φ  √ =
σ T −1
  
σ2
ln(S0 ) + (T − 1) r +
= S0 Φ   (9)
2
√
σ T −1

This is our answer.

Answer: The price of our asset equals
  
σ2
ln(S0 ) + (T − 1) r +
S0 · Φ  
2
√
σ T −1

Rx  2
where Φ(x) is a distribution function of a standard normal random value, Φ(x) = √1
−∞ 2π
exp − t2 dt

30
6.3 Exam, 17.01.2014

Optimal control part

1. [10 points] A collector of wine is told by a doctor that he is going to die in T years and he
develops a consumption plan that maximizes the utility of consumption of wine over remain-
ing lifetime. At present the stock of wine equals w(0) = w0 > 0. It will be consumed fully
by time T , that is w(T ) = 0. The utility function is UR(c) = c, where c is consumption of
t
wine. The stock of wine follows the rule w(t) = w0 − 0 c(s) ds. Consumption is bounded
0 ⩽ c ⩽ c̄, where c̄ > w0 /T .
State the dynamic optimization problem, if the discount rate is r > 0. Introduce the current
value Hamiltonian for that problem. Write down the first-order conditions. Solve the system
of equations.
P∞
2. [10 points] Consider
√ the following optimization problem: maximize t=0 0.75 ln ct , subject
t

to ct + kt+1 = kt , where k0 > 0. Let the state variable be k and denote the next period value
of k as k ′ .

(a) Write down the Bellman equation for the value function V (k)

(b) Using the method of undetermined coefficients find V (k)

(c) Find the optimal policy function k ′ = h(k)

3. [20 points] Consider the nonlinear system of differential equations

(
ẋ = −y + ax(x2 + y 2 )
ẏ = x + ay(x2 + y 2 )
where a is a parameter.

(a) [5 points] Linearize the system and classify the steady-state solution

(b) [10 points] Restore the nonlinear system and by changing the Carthesian into the polar
coordinates show that the system becomes
(
ṙ = ar3
θ̇ = 1

(c) [5 points] Solve this system and draw the phase diagram for a < 0, a = 0, a > 0.

Stochastic calculus part

Here Wt always denotes the standard Wiener process.

31
 1. [10 points] It is known that Var(X) = 16, Var(Y ) = 9, Cov(X, Y ) = −1, E(X) = E(Y ) = 0.
Let’s denote Ŷ = E(Y | X) and X̂ = E(X | Y ). It is also known that E(X̂ 2 ) = 4 and
E(Ŷ 2 ) = 1. Find Var(Ŷ ), Cov(X̂, Y ), Cov(Ŷ , Y ).
2. [10 points] Let’s consider a Wiener process with drift, Xt = Wt + µt.
(a) Find a non-trivial martingale of the form Mt = eβXt
(b) Let τ be a stopping time, the first moment when Xt hits 2 or −1. Find the probability
P(Xτ = 2)
You may assume without proof that some version of Doob’s theorem may be applied here.
3. [10 points] Is Xt = cos Wt a martingale? If not, find any non-zero function f (t) such that
Yt = f (t)Xt is a martingale. Find the variance of Yt .
Hint: You may use the fact that E(cos Wt ) = e−t/2 .
4. [10 points] In the framework of Black and Scholes model find the price at the time 0 of an
asset which gives you the payoff ST2 at the time T . Here St is the price of the underlying asset
at time t.
5. [20 points] Consider the following stochastic differential equation

dXt = −0.5e−2Xt dt + e−Xt dWt ,

with deterministic initial value X0 .

(a) Is Xt a martingale?
(b) Using the substitution Yt = f (Xt ) solve this differential equation. Hint: try to find a
function f such that the term before dt cancels out.
(c) Sketch the possible path of the process Xt
(d) Using basic facts about Wiener process find the limit limt→∞ P(Xt ⩾ a) for all values of
the parameter a.

6.4 marking scheme

1. 3 pts:
Var(Ŷ ) = E(Ŷ 2 ) − (E(Ŷ ))2 = 1 − 02 = 1

32
 3 pts:

Cov(X̂, Y ) = E(E(X | Y )Y ) − E(X̂) E(Y ) = E(E(Y X | Y )) − E(X) E(Y ) = Cov(X, Y )

4 pts:
Cov(Ŷ , Y ) = . . . = Cov(Ŷ , Ŷ )
2. Mt = e−2µXt 5 pts, probability — 5 pts, p = 1−exp(2µ)
exp(−4µ)−exp(2µ)
3. Xt is not a martingale — 1 pt,
1
dYt = cos Wt f ′ (t)dt + f (t)(− sin Wt dWt + (− cos Wt )dt)
2
′
We obtain simple differential equation f (t)−f (t)/2 = 0, we choose any solution, for example,
f (t) = et/2 .
Yt = et/2 Xt or proportional — 4 pts, variance — 5 pts

Var(Yt ) = E(Yt2 ) − (E(Yt ))2

We note that Yt is a martingale by construction, so E(Yt ) = Y0 = 1.
So, E(et/2 cos Wt ) = 1. Hence E(cos Wt ) = e−t/2 .
Let’s recall from middle school that cos2 Wt = 1+cos2 2Wt .
We also note that 2Wt ∼ N (0; 4t) and W4t ∼ N (0; 4t). In any expectation we may safely
replace 2Wt by W4t .
 
t 1 + cos 2Wt et et
E(Yt ) = E(e cos Wt ) = E e
2 t 2
= + E(cos 2Wt )
2 2 2
Where E(cos 2W t) = E(cos(W4t )) = e−4t/2 = e−2t .
Finally we obtain,

Var(Yt ) = (et + e−t − 2)/2

4. Pricing formula — 2 pts, risk-neutral substitution
 — 2 pts, calculations — 6 pts.
−rT
Using the pricing formula X0 = EP̃ e XT :
 
X0 = EP̃ [exp(−rT )XT ] = exp(−rT ) EP̃ ST2

We substitute ST = S0 exp((r − σ 2 /2)T + σ W̃T ):

h i
X0 = exp(−rT ) EP̃ S02 exp(2(r − σ 2 /2)T + 2σ W̃T ) =
h i
= S02 exp(−rT ) exp(2(r − σ /2)T ) EP̃ e
2 2σ W̃T
(10)

We use the fact that E[exp(aWt )] = exp(a2 t/2), so:

X0 = S02 e−rT exp(2(r − σ 2 /2)T ) exp(4σ 2 T /2) = S02 exp((2r + σ 2 )T )

5. (a) Xt is not a martingale — 1 pt
(b) use of Ito’s formula — 5 pts, equation f ′ = f — 5 pts, solution Xt = ln(Wt + eX0 ). — 5
pts
(c) Key features: “jiggly” — 1 pt, blows down to minus infinity in finite time — 2 pts

33
 (d) The limit is zero.

6.5 Retake, 08.02.2014

Stochastic calculus part
1. [10 points] Is Xt = sin Wt a martingale? If not, find any non-zero function f (t) such that
Yt = f (t)Xt is a martingale. Find the expected value E(Xt )
2. [10 points] In the framework of Black and Scholes model find the price of the asset, which
pays you S2 /S1 at the fixed moment of time T = 2
3. [10 points] Let τ be a stopping time, the first moment when Wt hits 2 or −1.
(a) Find the probability P(Wτ = 2)
(b) Find a martingale of the form Xt = Wt2 + f (t).
(c) Find E(τ )
4. [10 points] The joint distribution of the random vector (X, Y ) is given by its probability density
function  x−y
ce , for 0 ⩽ x, y ⩽ 1
f (x, y) =
0, otherwise
where c is a normalization constant. Find E(X | Y ).
5. [20 points] Let’s consider the following system of stochastic differential equations

dXt = aXt dt − Yt dWt
dYt = aYt dt + Xt dWt

with initial conditions X0 = x0 and Y0 = 0

(a) Ignoring the initial conditions find solution of the form Xt = f (t) cos Wt and Yt =
g(t) sin Wt
(b) Modify your solution to take into account the initial conditions
(c) Prove that for any solution Dt = Xt2 + Yt2 is nonstochastic

6.6 Answers and hints

1. f (t) = et/2 or proportional, E(Xt ) = 0. 
2. Using the pricing formula X0 = EP̃ e−rT XT :

X0 = EP̃ [exp(−rT )XT ] = exp(−2r) EP̃ [S2 /S1 ]

We substitute ST = S0 exp((r − σ 2 /2)T + σ W̃T ):

h i
X0 = exp(−2r) EP̃ exp(r − σ 2 /2 + σ W̃2 − σ W̃1 ) =
h i
= exp(−2r) exp(r − σ /2) EP̃ exp(σ W̃2 − σ W̃1 )
2
(11)

The distribution of W̃2 − W̃1 is similar to the distribution of W̃1 and is N (0; 1). So, we simplify:
h i
X0 = exp(−2r) exp(r − σ 2 /2) EP̃ exp(σ W̃1 )

34
 We use the fact that E[exp(aWt )] = exp(a2 t/2), so:

X0 = exp(−2r) exp(r − σ 2 /2) exp(σ 2 /2) = exp(−r)

3. Using Doob’s theorem, E(Wτ ) = 0, so P(Wτ = 2) = 1/3. Using Ito’s lemma or otherwise
Xt = Wt2 − t is a martingale, so E(Xτ ) = 0 and E(τ ) = 2.
R1
4. E(X | Y ) = 0 xf (x | Y ) dx. Here f (x, y) maybe factored, so X and Y are independent, and
E(X | Y ) = const.
5. See the solution of problem 8 from exam of 2008-2009 on page 5

7 2014-2015
7.1 Practice problems
These problems were not graded.
1. An enemy submarine is somewhere on the number line. The initial coordinate of the subma-
rine is some unknown integer number. It is moving at some constant integer speed (units per
minute).
You can launch a torpedo each minute at any integer on the number line. If the the submarine
is there, you hit it and it sinks. You have infinite number of torpedoes. You must sink this
enemy sub.
(a) Draw a picture of number line and the submarine. Just for fun!
(b) Devise a strategy that is guaranteed to eventually hit the enemy submarine
2. Compare the power of two sets: A = Q — the set of all rational numbers, B = Q2 — the set
of all possible pairs of rational numbers.
3. Using the fact that the Borel σ-algebra B is the smallest σ-algebra containing all subsets of
the form (−∞; t] show that N ∈ B.
4. Let Ω = R. Find explicitely the smallest σ-algebra which contains the sets A = [0; 1] and
B = [10; 100].
5. Let X be uniform on [0; 1] and Y = 1X<0.7 + 1X>0.1 . Describe the σ-algebra σ(Y ). How many
events σ(Y ) contains? Is the set σ(X) countable?
6. How many different σ-algebras one may construct if Ω contains three elements? Four?
7. We throw a coin infinite number of times. Let’s define the sequence of random variables Xn
such that Xn is equal to 1, if the result of n-th throw is tail, and 0 otherwise. We also define a
bunch of σ-algebras: Fn := σ(X1 , X2 , . . . , Xn ), Hn := σ(Xn , Xn+1 , Xn+2 , . . .).
(a) Give two non-trivial (other than Ω and ∅) examples of event A, such that A ∈ F2014 .
(b) Give two non-trivial examples of event A, such that A ∈
/ F2014 .
(c) Give two non-trivial examples of event A, such that A belongs to every Hn .
(d) Which σ-algebras contains the event A = {X37 > 0}?
(e) Which σ-algebras contains the event B = {X37 > X2014 }?
(f) Which σ-algebras contains the event C = {X37 > X2014 > X12 }?
(g) Simplify where possible: F11 ∩ F25 , F11 ∪ F25 , H11 ∩ H25 , H11 ∪ H25 .
8. Veniamin throws a coin until three consecutive tails appear. Let the random variable T be the
number of throws. Let FT be the σ-algebra of all events distinguishable by Veniamin.

35
 (a) Provide two non-trivial (not equal to Ω or ∅) examples of event A such that A ∈ FT but
A∈/ σ(T )
(b) Provide two non-trivial (not equal to Ω or ∅) examples of event A such that A ∈ σ(T )
(c) Is it true, that σ(T ) ⊂ FT ?

7.2 Practice-solutions
1.
2. The set A and B have the same power.
3.
4. The σ-algebra contains 8 = 23 events: all combinations of A, B and R/(A ∪ B).
5. σ(Y ) = {∅, Ω, {Y = 1}, {Y = 2}}, σ(X) is not countable.

7.3 Hometask
1. Researcher Veniamin throws a fair dice until 6 appears. Let denote by T the total number of
throws and by N the number of throws when 5 appeared. Find E(N | T ), Var(N | T ), E(N ),
Var(N ) and E(T | N ).
2. It is 25◦ C in Australia today. Each day the temperature goes one degree up or down with
equal probability. Each day I will put the sum equal to the temperature in my piggy bank.
So today I will put 25 roubles. I will stop my investment strategy when the temperature will
reach 15◦ C or 30◦ C for the first time.

What will be the expected value of my account in my piggy bank?

3. Veniamin and Varvara are managers of two gladiators-vampires teams. Veniamin has 4
gladiators-vampires with initial strengths 1, 2, 3 and 4. Varvara has 3 glagiators-vampires
with initial strengths 1, 3 and 5.
The competition between two teams is organised as a sequence of rounds. In each round two
gladiators-vampires (one from each team) will meet and fight to the death. Varvara always
selects the best gladiator from her team to fight in the next round.
When the gladiators of strengths a and b meet the first will win with probability a/(a + b),
the second — with probability b/(a + b). The gladiators are vampires, so the strength of the
winner will become a + b.
(a) Let’s denote by τ the number of the final round. What is the maximum value of τ ?
(b) Let’s denote by Xt the total strength of the Veniamin’s team. Is Xt a martingale?
(c) What is best strategy for Veniamin? What is the probability that Veniamin’s team will
win?
Rt Rt
4. Consider the processes Xt = 0 (sWs )3 dWs and Yt = 0 Ws dWs . Find E(Xt ), Var(Xt ),
Cov(Xt , Yt )
5. The process Xt is given by

dXt = tWt eWt dt + sin(tWt ) dWt , X0 = 1

36
 Using Ito-Doeblin lemma find Z0 and dZt if Zt = Xt2 + t cos(Xt )
6. Solve the stochastic differential equation:

dXt = Xt (1 − Xt ) dt + Xt dWt , X0 = 1

Some hints: you may find the substitution Yt = ln Xt useful.

7. In the framework of Black and Scholes model find the price at t = 0 of the asset, which pays
you 1$ if ST −1 > 1 and 0 otherwise. Here St denotes the price of a share at time t. The
payment is made at the fixed moment of time T > 1.

7.4 Hometask-solution
1. If the total number of throws it T then the last throw has value 6. Each of the preceeding
(T − 1) throws may have a value from 1 to 5. So if T is considered fixed, then N has binomial
distribution with n = T − 1 and p = 1/5. It follows that E(N | T ) = T −15
and Var(N | T ) =
4(T −1)
25
.
Using tower property of conditional expectation
 
T −1 E(T ) − 1
E(N ) = E(E(N | T )) = E =
5 5

Here T is geometric with p = 1/6, so E(T ) = 6 and E(N ) = 1.

We use the following decomposition of variance:

Var(N ) = E(Var(N | T )) + Var(E(N | T )) =

   
4(T − 1) T −1
=E + Var = (Var(T − 1) + 4 E(T − 1))/25 (12)
25 5

As T is geometric Var(T ) = (1 − p)/p2 = 30. And, hence, Var(N ) = 2.

If N is fixed, then the total time T is equal to the following sum: from start to first five, from
first five to second five, …, from N -th five to six, so E(T | N ) = aN + b. From the properties
of geometric distribution, b = 5 (no five) and a = 5 (no six).
2. X0 = 25, Xt+1 = Xt + Dt+1 — is a martingale. The random variable Xτ can take only the
values 30 and 15. Applying the Doob’s theorem we get E(Xτ ) = 25 and, hence, P(Xτ =
15) = 1/3.
Using the second martingale, Xt2 − t and again Doob’s theorem, E(Xτ2 ) − E(τ ) = X02 = 625,
and E(τ ) = 50.
P
Using the thirdPmartingale, Mt = ti=0 Xi − 25(t + 1). Here we have (t + 1) P as there are
(t+1) terms in ti=0 Xi . Using Doob’s theorem, E(Mτ ) = M0 = 0, and, finally, E( τi=0 Xi ) =
25 · 51 = 1275.
3. τ ∈ {3, 4, 5, 6}, Xt is a martingale. The random variable Xτ can take only two values, 0 or
1 + 2 + 3 + 4 + 1 + 3 + 5 = 19. According to the Doob’s theorem, E(Xτ ) = X0 = 10. And
hence P(Veniamin wins) = P(Xτ = 19) = 10/19. The strategy does not matter.
4. E(Xt ) = 0, Var(Xt ) = 3t1 0/2, and Cov(Xt , Yt ) = t6 /2.
The expected value
E(Wt2k ) = tk · 1 · 3 · 5 · . . . · (2k − 1)
may be useful with Ito’s isometry.

37
 5.

dZt = [cos Xt + 2Xt tWt eWt − t2 Wt eWt sin Xt + sin2 (tWt ) − 0.5t cos Xt sin2 (tWt )] dt+
[sin(tWt )(2Xt − t sin(Xt ))] dWt (13)

R t substitution Yt = ln Xt we obtain equation dYt = (0.5 − Xt ) dt + dWt . And

6. Using the
ln Xt + 0 Xs ds = 0.5t + Wt .
We exponentiate and then take integral
Z t  Z t 
Xs ds = ln 1 + exp(0.5s + Ws ) ds
0 0

Differentiate to obtain
exp(0.5t + Wt )
Xt = Rt
1 + 0 exp(0.5s + Ws ) ds
7.  
−rT ln S0 − (σ 2 /2 − r)(T − 1)
X0 = e F √ ,
σ T −1
where F () is the cumulative probability function for standard normal distribution.

7.5 Exam, 27.12.2014

Optimal control part
1. (10 points) Consider the nonlinear system of DE
(
ẋ1 = ex1 +x2 − x2
ẋ2 = −x1 + x1 x2

Determine the points of rest of the system, classify their type and stability and draw the local
phase diagram at each of them with the found eigenvectors.
2. (10 points) Solve the optimal control problem
Z T  
pK(t) − wK(t) − cI 2 (t) dt → max
0

subject to K̇(t) = I(t), K(0) = K0 > 0, K(T ) not given, p > w > 0, c > 0.
Also show that K(t) > 0 for t ∈ [0; T ] and K̇(T ) = 0 (4 points for that part).
3. (20 points) Consider the optimal problem for the infinite horizon
Z ∞  
−rt 1 2 1 2
e − u + xu − x dt → max
0 2 2

subject to ẋ = u, x(0) = x0 > 0 and 1 < r < 2.

(a) (5 points) Write down the system of equations based on the current value Hamiltonian
(b) (3 points) Is the integrand a concave or a strictly concave function?
(c) (5 points) Reduce the necessary conditions down to a pair of linear DE involving only x
and λ

38
 (d) (3 points) Find all steady-states of the system
(e) (4 points) Classify them

Stochastic calculus part

Here Wt always denotes the standard Wiener process.

1. [10 points] Let τ denote the first moment of time when |Wt | = 100.
(a) What is the distribution of Wτ ?
(b) Are Wτ and τ independent?
(c) Assuming that some version of Doob’s theorem may be applied find E(e−2τ ).
Hint: maybe exp(2Wt − 2t) will help?
2. [10 points] The random variables X1 , X2 , …, Xn , …are independent uniformly distributed on
[0; 1]. I am summing them until the first Xi greater than 0.5 is added. After this term I stop.
Let’s denote by S the total sum and by N — the number of terms added. Find E(S | N ),
Var(S | N ), E(S)
Hints: If U is uniform on [a; b] then Var(U ) = (b − a)2 /12. If G has geometric distribution
(the number of throws to get the first success) then E(G) = 1/p where p is the probability of
success. R 
t
3. [10 points] Find Var 0 Ws ds .
You may use the following guiding steps:
(a) Find d(tWt ) in short and full form
 Rt 
(b) Find E 2tWt 0 s dWs
 2 
Rt
(c) Find E 0
s dWs
R 
t
(d) Find E 0
Ws ds

(e) (a − b)2 = a2 − 2ab + b2 :)

4. [10 points] The risk-free interest rate is equal to 0.1. The volatility of the share is equal to
σ = 1. You have an option to receive 1$ two years later if the percentage change of price of
the share, St , during the first year is less than during the second year. Assume the framework
of the Black and Scholes model. What is the fair price of this option?
5. [20 points] Solve the stochastic differential equation

dXt = t dt + Xt dWt , X0 = 1

You may use the following guiding steps:

(a) Solve a less difficult stochastic differential equation, dYt = −Yt dWt , Y0 = 1
(b) Find dAt , where At = Xt Yt
(c) Find any deterministic function f (t) such that d(f (t)At ) contains neither Xt nor At
(d) Using full form of d(f (t)At ) find Xt . It may depend on some integrals :)

39
7.6 Exam, 27.12.2014 - solution
1. Wτ takes values 100 and −100 with equal probabilities. Yes, independent. E (e−2τ ) =
2
e200 +e−200
2. E(S | N ) = 0.25N + 0.5, Var(S | N ) = 0.25N /12, E(N ) = 1/p = 2, E(S) = E(E(S |
N )) = 1. One may also find the variance, Var(S) = 1/8 + 1/24 = 1/6.
3. By hints:
(a) d(tWt ) = t dWt + Wt dt
(b) Using Ito’s isometry:

 Z t  Z t Z t  Z t Z t 
E 2tWt s dWs = 2t E 1 dWs s dWs = 2t Cov 1 dWs , s dWs =
0 0 0 0 0
Z t
= 2t E(1 · s) ds = t3 (14)
0

 2 
Rt
(c) E 0
s dWs = t3 /3 using Ito’s isometry
R 
t
(d) E Ws ds = 0
0
R 
t
(e) Var 0 Ws ds = t3 /3
n o
4. Event S1S−S
0
0
< S2 −S1
S1
simplifies to S12 < S0 S2 and further simplifies to 2W̃1 − W̃2 < 0.
And
X0 = e−0.2 P̃ (2W̃1 − W̃2 < 0) = e−0.2 /2
5. By steps:
(a) Using the substitution Ut = ln Yt , one obtains
1
dUt = −dWt − dt
2
And the solution Yt = exp(−Wt − t/2)
(b) As At = Xt Yt we have:
1
dAt = Xt dYt + Yt dXt + 2dXt dYt
2
40
 Finally
dAt = Yt (t − Xt )dt
(c)
d(f (t)At ) = f ′ (t)At dt + f (t)dAt = f ′ (t)Xt Yt dt + f (t)Yt (t − Xt )dt
To kill Xt in this expression we need to find a function f such that f ′ (t) − f (t) = 0. For
example, f (t) = et .
(d) From previous point we have
d(et At ) = et Yt tdt
And Z t
t 0
e At = e A0 + es Ys s ds
0

From initial condition A0 = 1 and finally

Rt
1+ 0
s exp(−Ws + s/2) ds
Xt =
exp(−Wt + t/2)

8 2015-2016
8.1 Stochastic calculus hometask
1. Find the expected value of E(exp(aWt )), E(1Wt ⩽b ), E(exp(aWt )·1Wt ⩽b ) and E(Wt 1Wt ⩽b ). Nat-
urally, you may use the standard normal cumulative distribution function F in your answer
2. It is known that E(Y | X) = 0. Which of the following quantities must be zero: E(Y )? E(X)?
Cov(X, Y )? Cov(X 2 , Y )? Cov(X, Y 2 )? Prove or provide a counter-example.
3. Alisa and Bob throw a fair coin until either the sequence THTH or the sequence HTHH ap-
pears. Alisa wins if the sequence THTH appears first and Bob wins if the sequence HTHH
appears first.
(a) What is the probability that Alisa will win?
(b) What is the expected duration of the game in tosses?
Hint: you may introduce a martingale from lecture or solve this problem without martingales
at all
4. Consider the process Yt = exp(2Wt − 2t).
(a) Find dYt
Rt
(b) Find 0 Yu dWu
(c) Find E(Yt ) and Var(Yt )
5. Consider stochastic differential equation

dXt = (a − bXt )dt + cdWt , X0 = x0

(a) Solve this differential equation1 .

(b) Find E(Xt ) and Var(Xt )
1
The answer may contain an Ito integral that cannot be simplified.

41
 6. In the framework of Black and Scholes model find the price at t = 0 of the classic European
сall option by calculating corresponding expected value.
European call option with strike price K is the right to buy at time t one share at price K. So,
at time t it pays you St − K if St > K and zero otherwise.

8.2 Stochastic calculus hometask — Solution

√
1. Let’s remark that Wt ∼ N (0; t) and Z = Wt / t ∼ N (0; 1).
√ Z ∞ √ 2
E(exp(aWt )) = E(exp(a tZ)) = exp(a tz)f (z)dz = ea t/2
−∞
√
E(1Wt ⩽b ) = F (b/ t), E(exp(aWt ) · 1Wt ⩽b ) = ..., E(Wt 1Wt ⩽b ) = ...
2. Short answers:

(a) E(Y ) = E(E(Y | X)) = 0.

(b) The value E(X) can be non-zero, counterexample: X ∼ N (42; 42), Y = 0 = const.
(c) Cov(Y, f (X)) = E(f (X)Y ) − E(Y ) E(f (X)) = E(f (X)Y ).

E(f (X)Y ) = E(E(f (X)Y | X)) = E(f (X) E(Y | X)) = E(0) = 0

So, Cov(Y, X) = Cov(Y, X 2 ) = 0

(d) Cov(X, Y 2 ) may be non-zero. Let Y take the values −1, 1, −2 and 2 with equal proba-
bility. And X = Y 2 . In this case Cov(X, Y 2 ) = Var(X) > 0.

3. Let’s denote the moment of time when the sequences A and B appear by NA and NB corre-
spondingly. The game ends at moment N = min{NA , NB }.
We need to find pA = P(N = NA ), pB = P(N = NB ) and E(N ). We have one equation,
pA + pB = 1 so we need two more equations:
We build them from nothing :)
(
E(NA ) = E(N ) + E(NA − N )
E(NB ) = E(N ) + E(NB − N )

Here E(NA ) = 16 + 4 = 20, E(NB ) = 16 + 2 = 18.

Or
(
E(NA ) = E(N ) + pA E(NA − N | N = NA ) + pB E(NA − N | N = NB )
E(NB ) = E(N ) + pA E(NB − N | N = NA ) + pB E(NB − N | N = NB )
(
E(NA ) = E(N ) + pB E(NA − N | N = NB )
E(NB ) = E(N ) + pA E(NB − N | N = NA )
…
Based on http://projecteuclid.org/download/pdf_1/euclid.aop/
1176994578
There is also a solution using Markov chains.

42
8.3 Exam, 12.01.2016
Stochastic calculus part
Here Wt always denotes the standard Wiener process.

1. [10 points] You throw a fair coin until «head» appears. Let’s denote the result of the first
toss by Y1 (0 for tail and 1 for head) and the total number of throws by N . Find E(Y1 | N ),
Var(Y1 | N ) and E(N | Y1 )
2. [10 points] Consider τ , the first moment of time when the standard Wiener
√ process will touch
the barrier 4y = x + 1, or formally, τ = inf{t | t ⩾ 0, |Wt | = 0.5 t + 1}. Find E(τ ).
2

Hint: you may find the process Mt = Wt2 − t useful, you may also suppose that technical
conditions of Doob’s theorem are satisfied
3. [10 points] Let  Z 
t
Yt = exp −6t3 + f (s) dWs ,
0

where f is some deterministic function.

(a) Using Ito’s lemma find dYt
(b) Find at least one function f such that Yt is a martingale
4. [10 points] The risk-free interest rate is equal to 0.1. The volatility of the share is equal to
σ = 1. You have an option to receive 1$ two years later if the price growth during the second
year is higher than 5%. Assume the framework of the Black and Scholes model. What is the
fair price of this option?
5. [20 points] Consider the stochastic differential equation

dXt = Xt3 dt + Xt2 dWt ,

(a) Apply Ito’s lemma to Yt = f (Xt )

(b) Find all the functions f that makes Yt a martingale
(c) Find all the solutions of the stochastic differential equation
(d) Find the solution such that X0 = 2.
(e) Find the probability that the sample path of Xt will be continuous for t ∈ [0; ∞)

Optimal control part

6. (10 points) Given the system of differential equations
(
ẋ = sin(x + y)
ẏ = sin(x − y)

explore the behavior of its solutions in the neighborhood of the 2 points: A(π, π) and
B(3π/2, 3π/2). Draw the phase portraits near A and B based on the knowledge of their
eigenvalues, eigenvectors where possible.
7. (10 points) Solve the bounded control problem: maximize
Z 1
(2x − u2 /2) dt
0

43
 subject to constraints ẋ = u − x + t2 , x(0) = 0, −1 ⩽ u ⩽ 0. Verify that the maximizer has
been found by applying one of the sufficient conditions.
8. (20 points) Solve so-called “limit pricing problem”. A homogeneous product is produced by
a dominant firm along with the “competitive fringe” consisting of the x(t) identical firms (x
is a continuous variable). Demand on good is given by f (p) ∈ C 2 , where p(t) is the price,
f ′ (p) < 0 for p > 0. Let the dominant firm has a constant returns to scale technology with
the marginal costs c = const > 0. Then (p−c)f (p) is a strictly concave function. The problem
of the firm is to maximize the discounted stream of profits
Z ∞
e−rt (p − c)(f (p) − x) dt
0

(each fringe firm produces only one unit of good) subject to constraint ẋ = k(p − p̄) where k
is some number, p̄ is the equilibrium price and r > 0. Also x(0) = x0 .
(a) Application of the current value Hamiltonian is required here. Derive the system of the
first-order conditions.
(b) By eliminating the Lagrange multiplier reduce the system to 2 equations with respect to
(x, p). Check that Hpp < 0.
(c) Show that if the dominant firm operates at the price level p̄ = c then the fringe firms
supply the entire market in the equilibrium.
s
(d) Let the equilibrium price p̄ be slightly above c. By evaluating the derivative ∂x
∂ p̄
at p̄ = c
show that the dominant firm’s market share becomes positive if p̄ is slightly above c (xs
is the equilibrium number of the fringe firms).

8.4 Solution, exam 12.01.2016

1. If we know the value N then we know the value of Y1 , so Var(Y1 | N ) = 0 and
(
0, if N > 1;
E(Y1 | N ) =
1, if N = 1
(
1, if Y1 = 1;
E(N | Y1 ) =
1 + 2, if Y1 = 0
Or, simply, E(N | Y1 ) = 3 − 2Y1
2. We check that Mt is a martingale: dMt = −dt + 2Wt dWt + 12 2 dt = 2Wt dWt . And we apply
Doob’s theorem:
E(Wτ2 − τ ) = E(Mτ ) = E(M0 ) = 0
So, E(Wτ2 ) = E(τ )
From the definition of τ , Wτ2 = (τ + 1)/4. We get the equation

E(τ ) = E(τ + 1)/4

And E(τ ) = 1/3

Based on http://www-stat.wharton.upenn.edu/~shepp/publications/
14.pdf.

44
 Rt
3. Let’s write the process Yt as Yt = exp(Zt ) where Zt = −6t3 + 0
f (s) dWs . So, dYt =
exp(Zt ) dZt + 12 exp(Zt ) (dZt )2 .
We find dZt :
dZt = −18t2 dt + f (t) dWt
So, (dZt )2 = f 2 (t) dt and
1
dYt = exp(Zt )(−18t2 dt + f (t) dWt ) + exp(Zt )f 2 (t) dt
2
To cancel the term before dt one should have
1
−18t2 + f 2 (t) = 0
2
Possible solutions are f (t) = 6t and f (t) = −6t
4. We should find the discounted expected price:

X0 = e−2r P̃(S2 /S1 > 1.05) = e−2r P̃(exp((r − σ 2 /2) + σ(W̃2 − W̃1 )) > 1.05)
We remark that W̃2 − W̃1 ∼ N (0; 1) under P̃. So

X0 = e−2r P̃(N (0; 1) > (ln 1.05−r+σ 2 /2)/σ) = e−2r F ((r−σ 2 /2−ln 1.05)/σ) = e−0.2 F (−0.45)

5. Step by step
(a) Apply Ito’s lemma:
1 1
dYt = f ′ (Xt ) dXt + f ′′ (Xt ) (dXt )2 = f ′ · (Xt3 dt + Xt2 dWt ) + f ′′ · Xt4 dt
2 2

(b) To cancel the term before dt the following condition must be satisfied
1
f ′ · Xt3 + f ′′ · Xt4 = 0
2

Let’s denote f ′ by g, so g + g ′ · x/2 = 0. This equation may be solved by separation of

variables:

dx dg
−2 =
x g
The general solution is g(x) = cx−2 , so the general solution for f is

f (x) = c1 x−1 + c2

(c) To find all the solutions we use the simpliest possible f , f (x) = x−1 . In this case

dYt = 0 dt + f ′ · Xt2 dWt = −dWt

So, Yt = Y0 − Wt and
1
Xt =
c − Wt

45
 (d) If X0 = 2 than Xt = 1/(0.5 − Wt )
(e) The probability that Wt will forever stay below 0.5 is zero, so the probability that Xt will
have continuous sample path for all t is also zero.

8.5 Retake, 15.02.2016

Stochastic calculus part
Here Wt always denotes the standard Wiener process.

1. [10 points] You throw a fair coin infinite number of times. Let’s denote the result of the second
toss by Y2 (0 for tail and 1 for head) and the number of throws to get the first «head» by N .
Find E(Y2 | N ), Var(Y2 | N ) and E(N | Y2 )
2. [10 points] The process Yt is given by Yt = 2Wt + 5t. The stopping time τ is given by τ =
min{t | Yt2 = 100}. Find the distribution of the random variable Yτ and the expected value
E(τ ).
Hint: you may find the martingales aYt and Yt − f (t) useful
3. [10 points] Let X0 = 2016 and dXt = 2t dt + t2 dWt . Find E(Xt ) and Var(Xt ).
4. [10 points] The risk-free interest rate is equal to 0.1. The volatility of the share is equal to
σ = 1. The price of a share at t = 0 is S0 = 100. You have an option to receive 1$ two years
later if the price of the share after one year is more than 105. Assume the framework of the
Black and Scholes model. What is the fair price of this option?
5. [20 points] Consider the stochastic differential equation

dXt = 8Wt2 Xt dt + 4Wt Xt dWt , where X0 = 1

(a) Apply Ito’s lemma to Yt = ln Xt

(b) Find the solution of the initial stochastic differential equation

Optimal control part

6. Find extremals that provide the highest or lowest values of the following integral
Z e 
1 ′ 2 2yy ′ y 2
J(y) = x(y ) + − 2 dx
1 2 x x
with the boundary values y(1) = 1, y(e) = 2.
(a) (10 points) Find the extremal(s).
(b) (10 points) Let ỹ(x) be the extremal you have found. Let h(x) ∈ C 1 and h(1) = h(e) = 0
(h(x) is not identically zero). Prove that J(ỹ + h) − J(ỹ) > 0.
R∞
7. Consider Ramsey’s model. Maximize the integral I = 0 [u(c) − B]dt subject to k̇ = f (k) −
c − δk, k(0) = k0 . Function u(c) monotonically increases and tends to B at the infinity,
moreover I converges.
d ′
(a) (5 points) Derive Ramsey’s Law dt
u (c) = u′ (c)[δ − f ′ (k)].
(b) (5 points) Let ( 2
2c − cB , for c ⩽ B
u(c) =
B, otherwise

46
 Production function f (k) = αk and α > δ. Find the optimal solutions c∗ , k ∗ . Do these
solutions necessarily have an economic sense?
RT
8. (10 points) Solve the problem on the bounded optimal control 0 (1 − β − u)xdt → max,
subject to ẋ = αxu, x(0) = x0 , 0 ⩽ u ⩽ 1. In this problem α > 0 and 0 < β < 1.

8.6 Retake, solution

1. 

0.5, if N = 1;
E(Y2 | N ) = 1, if N = 2;


0, if N > 2;
(
0.25, if N = 1;
Var(Y2 | N ) =
0, if N > 1;
(
0.5 · 1 + 0.5 · (2 + 1/0.5), if Y2 = 0
E(N | Y2 ) =
0.5 · 1 + 0.5 · 2, if Y2 = 1
2. We find that Z t
2
Xt = 2016 + t + u2 dWu
0

So E(Xt ) = 2016 + t and Var(Xt ) = t /5.

2 5

9 2016-2017
9.1 Hometask
1. Consider the two independent Brownian motions, (Wt ) and (Vt ). Which of the following
processes is Brownian motion:
(a) Zt = 12 Wt + 12 Vt
(b) Qt = √12 Wt + √12 Vt
Rt
2. Let Yt = 0 (Wu + u)2 dWu . Find E(Yt ) and Var(Yt ).
3. James Bond plays in a casino. At every bet he wins one pound with probability 0.5, looses
one pound with probability 0.4 or wins nothing with probability 0.1. His initial fortune is
X0 = 10 pounds. He stops playing if he goes bankrupt or if he achieves the fortune of 300
pounds (airline ticket price from London to Moscow).
(a) Find the constant a such that Mt = aXt is a martingale.
(b) What is the probability that James Bond will win enough to buy the ticket?
4. Let Rt be the exchange rate at time t. We suppose that dRt = µRt dt + σRt dWt . Consider
the inverse exchange rate It = 1/Rt . Find the expression for dIt . The expression should not
contain Rt .
5. It is known that Mt is a martingale. We also know that in short-hand notation (dMt )2 = dt.
What can we say about the process Yt = Mt2 − t + 2017?

47
 6. Solve the stochastic differential equation
−1
dXt = Xt dt + dWt , X0 = 0 (15)
1−t
You may use or not use the following hints:
Rt
(a) The correct answer will contain the integral 1
0 1−u
dWu that cannot be simplified.
(b) Solve the ordinary differential equation
−1
dYt = Yt dt, Y0 = 1 (16)
1−t

(c) Represent Xt as Xt = Yt · Zt and find the equation for dZt . Find the expression for Zt .
7. Consider the framework of the Black and Scholes model. Let P be the original probability
measure and P̃ be the risk-neutral probability measure. Provide an example of three events A,
B and C such that P(A) > P̃(A), P(B) < P̃(B), P(C) = P̃(C).
8. Consider the framework of the Black and Scholes model. The asset pays you 1 dollar at fixed
time T if and only if the price of a share ST is above the strike-price K. Find the current price
X0 of this asset.

9.2 ht, solution

Solution by Anastasia Andreeva:
1. (a)
1 1
Zt = W t + V t
2 2
Var(Zt −Zs ) = Var[0.5·(W
t −Ws +Vt −Vs )] = 14 Var(Wt −Ws )+ 14 Var(Vt −Vs ) = 21 (t−s)
Thus, Zt − Zs ∼ N 0, t−s2
, therefore, Zt is not a Brownian motion.
(b)
1 1
Qt = √ Wt + √ V t
2 2
Let’s check all the conditions:
i. Q0 = √1 (W0
2
+ V0 ) = 0

ii. P(Qt trajectory is continuous) = 1 since Wt and Vt correspond to this condition

iii. Qt − Qs = √12 (Wt − Ws ) + √12 (Vt − Vs ) is independent on Fs since Wt − Ws and
Vt − Vs are independent on Fs

iv. E(Qt − Qs ) = E(Wt − Ws ) + √12 E(Vt − Vs ) = 0

√1
2 
Var(Qt − Qs ) = Var 2 · (Wt − Ws + Vt − Vs ) = 12 Var(Wt − Ws ) + 12 Var(Vt −
√1

Vs ) = t − s Thus, Qt − Qs ∼ N (0, t − s)

v. Qt is measurable relative to Ft
Thus, Qt is a Brownian motion.

48
2. Z t
Yt = (Wu + u)2 dWu
0

Yt is a martingale because it is a stochastic integral, so E(Yt ) = E(Y0 ) = 0.

E(Wu4 ) = 3u2 , E(Wu2 ) = u, E(Wu3 ) = E(Wu ) = 0

Z t Z t
Var(Yt ) = E(Wu + u) du =
4
E(Wu4 + 4Wu3 u + 6Wu2 u2 + 4Wu u3 + u4 )du =
0 0
Z t t
3u4 u5 3t4 t5
= 2 3 4
(3u + 6u + u )du = u +3
+ 3
=t + + (17)
0 2 5 0 2 5

3. (a) E(Mt+1 | Ft ) = E(aXt+1 | Ft ) = 0.5aXt +1 + 0.4aXt −1 + 0.1aXt = aXt (0.5a + 0.4a−1 +

0.1)
0.5a + 0.4a−1 + 0.1 = 1 ⇒ a2 − 1.8a + 0.8 = 0 ⇒ a1 = 1, a2 = 0.8
a ̸= 1 otherwise Mt = 1 for all t. So, a = 0.8.
(b) James Bond will get 300 with probability p and lose all money with probability (1 − p).
τ is the stopping moment (win or lose).
According to Doob’s theorem, E(Mτ ) = E(M0 ) = 0.810 .
At the same time E(Mτ ) = E(0.8Xτ ) = p · 0.8300 + (1 − p) · 0.80 = 0.810
1−0.810
Then p = 1−0.8 300 ≈ 0.89

Thus, the answer is 89%.

4. I will use the Ito’s formula:
(dRt )2 = σ 2 Rt2 dt

1 1 1 σ2
dIt = − dR t + (dR t ) 2
= − (µR t dt + σR t dW t ) + dt =
Rt2 Rt3 Rt2 Rt
1
= ((σ 2 − µ)dt − σdWt ) = It ((σ 2 − µ)dt − σdWt ) (18)
Rt
5.
1
dYt = 2Mt dMt − dt + 2dt = 2Mt dMt
2
Since Mt is a martingale dMt = At dWt . So dYt = 2Mt At dWt .
Since the expression does not include dt, Yt is a martingale.
6.
1
dXt = − Xt dt + dWt , X0 = 0
1−t
(a)
1
dYt = − Yt dt, Y0 = 1
1−t
dYt dt
=−
Yt 1−t
ln|Yt | = ln|1 − t| + C0
Yt = C(1 − t), Y0 = C = 1 ⇒ Yt = 1 − t

49
 (b)
X t = Yt Z t
Xt Z t Yt Xt
dXt = − dt + dWt = Zt dYt + Yt dZt = − dt + Yt dZt = − dt + (1 − t)dZt
1−t 1−t 1−t
Z t
dWt dWu
dWt = (1 − t)dZt ⇒ dZt = ⇒ Zt = Z 0 +
1−t 0 1−u

(c)  Z 
t
dWu
Xt = Yt Zt = (1 − t) Z0 + , X 0 = 0 = Z0
0 1−u
Z t
dWu
Xt = (1 − t)
0 1−u

7. Wt is a Brownian motion for P . Yt = Wt + µ−rσ

t is a Brownian motion for P̃ .
If µ > r, then Yt > Wt and P̃(Yt > 0) = P(Wt > 0) = P(Yt > µt) < P(Yt > 0).
If µ < r, then Yt < Wt and P̃(Yt > 0) > P(Yt > 0).
If µ = r, then Wt = Yt and P(Wt > 0) = P̃(Wt > 0).
8. (
1, if ST > K
XT =
0, otherwise

X0 = e−rT EP̃ (XT | F0 ) = e−rT P̃(ST > K | F0 ) = e−rT P̃(S0 e(r−0.5σ

2 )T +σ W˜T
> K | F0 )
 
σ2
ln S0 − r − 2 T
K

W̃T >
σ
Further, we standardize W̃T :
 
σ2
W̃ ln K
S0
− r− 2
T
√T > √
T σ T
Thus,
        
σ2 σ2
W̃T ln K
− r− T ln − r −
K
T
X0 = e−rT P̃  √ >  = e−rT 1 − Φ  S0 
S0 2 2
√ √
T σ T σ T

9.3 exam, 10.01.2017

Stochastic calculus part

Here Wt always denotes the standard Wiener process.

1. (10 points) You throw a standard fair die n times. Let X be the number of «fives» in the first
(n − 1) throws and Y — the number of «fives» in the last (n − 1) throws.
For n ⩾ 3 calculate E(Y | X), E(X | Y ), Var(Y | X).
Rt
2. (10 points) Find a constant b such that Yt = 0 Wu3 du + bWt5 is a martingale.

50
 3. (10 points) The process Xt evolves according to the formula
Z t
1
Xt = 1 − t + (1 − t) dWu
0 1−u

(a) Is Xt a martingale?
(b) Find E(Xt ), Var(Xt ) and Cov(Wt , Xt )
(c) Draw E(Xt ) and Var(Xt ) as functions of t
(d) Draw two possible trajectories of Xt
4. (10 points) Find the price of the «Asset-or-nothing» call option at time t = 0 in the framework
of Black and Scholes model. The risk-free interest rate is equal to r. The volatility of the share
is equal to σ. The current share price is S0 . The «Asset-or-nothing» call option pays you at
fixed time T the sum ST if ST is higher than the strike price K or nothing otherwise.
Hint: the correct answer will contain the normal cumulative distribution function F ().
5. (20 points) Interest rate evolves according to the stochastic differential equation

drt = −λ(rt − c) dt + σdWt

where r0 , c, λ and σ are some positive constants.

Solve this stochastic differential equation.
You may use or not use the following hints:
(a) Obtain a stochastic differential equation for Xt = rt − c.
(b) Solve the ordinary differential equation dYt = −λYt dt, Y0 = 1
(c) Represent Xt as Xt = Yt · Zt and find the equation for dZt .
(d) Solve the equation for Zt .
(e) The correct answer for rt will contain an Ito’s integral that cannot be simplified.

Optimal control part

6. (10 points) Solve the bounded control problem:
Z T
(u − x)2 dt → min
0

subject to ẋ = 12 (u − x), x(0) = 1, |u| ⩽ 1, T > 0.

7. (10 points) Find the extremals for the calculus of variations problem:
Z 1
1 2
ẋ + xẋ + x dt
0 2

when both endpoint values x(0) and x(1) can be chosen freely.
8. (20 points) Costs of the farmer’s production C(x, y) depends on the output y(t) and the fertility
of soil x(t) by the formula C = y 2 + √1x . This industry is perfectly competitive and the price of
the harvest per unit is p. Fertility changes over time by the dynamic law ẋ = B − αy, where
pα > 2B > 0, and x(0) > 0. Constant B is defined by the government subsidy. Profit of the
firm is discounted with the rate r > 0.

51
 (a) Write down the problem of maximizing the stream of discounted profit over the infinite
time horizon.

(b) Write down the system of first order conditions using the current value Hamiltonian.
Draw the phase portrait in (x, y) coordinate system, in which x — axis is drawn horizon-
tally. Find the steady-state (xs , ys ) if it exists, classify it using Jacobian. With the help
of the arrows provide the sketch of the dynamic paths of the system.

(c) Find the sign of ∂xs /∂p. What justification of this sign can you provide from the eco-
nomic point of view?

9.4 exam, solution

1.
n−2 1
E(Y | X) = X+
n−1 6
n−2 1
E(X | Y ) = Y +
n−1 6
2. Take dYt , kill term before dt, b = −0.1.

dYt = (Wt3 + 10bWt3 ) dt + 5bWt4 dWt

3. The process Xt is not a martingale as E(Xt ) = 1 − t.

Z t
1
Var(Xt ) = (1 − t) E(I ) = (1 − t)
2 2 2
du = t(1 − t)
0 (1 − u)
2

Z t Z t  Z t 
1 1
Cov(Wt , Xt ) = (1−t)· dWu · dWu = (1−t) E du = (t−1) ln(1−t)
0 0 1−u 0 1−u

4.
X0 = e−rT Ẽ(ST · 1ST >K )
5.
dXt = drt = −λXt dt + σdWt
Solving ODE we obtain Yt = e−λt .

dXt = Yt dZt + Zt dYt + dYt dZt

Equation for Zt :
dZt = eλt σdWt
Just recover the full form: Z t
Zt = Z0 + σ eλu dWu
0

Finally,  Z 
t
−λt
rt = e r0 − c + σ e λu
dWu +c
0

52
10 2017-2018

10.1 Hometask
1. Let Y = 1W3 ⩽2 . Find E(W5 · Y ), Var(W5 · Y ), Cov(W5 , Y ), Cov(W5 , W5 · Y ).
2. At time n = 0 I have one apples and one banana. At each moment of time I choose one of all
my fruits randomly with equal probabilities. I eat it and by two more of the same kind. For
example, if I choose banana I will eat it and buy two bananas more. I repeat the process of
choosing, eating and buying fruits indefinitely. Let An and Bn be the number of apples and
bananas after n rounds.

(a) Find non-trivial function f such that Mn = f (n) · An is a martingale.

(b) Let τ be the moment when the first banana is chosen. Assuming that Doob’s theorem is
applicable find E(1/(τ + 2)).
Rt Rt
3. Find all constants a such that Yt = 42 + Wt7 + a 0 Wu5 du + 0 u2 cos Wu dWu is a martingale.
Find E(Yt ).
4. Consider the framework of the Black and Scholes model. The asset pays you 1 dollar at fixed
time T if and only if
ST ST /2
> .
ST /2 S0
Find the current price X0 of this asset.
5. Solve the stochastic differential equation and find E(Xt ) and Var(Xt )

dXt + aXt dt = bdt + cdWt ;

You are free to use or not to use the following guiding steps:

(a) Find dYt for Yt = exp(γt)Xt ;

(b) Find the value of γ such that in dYt the term before dt is non-random.

(c) Solve the equation for Yt and then find Xt ;

(d) Do not forget about E(Xt ) and Var(Xt ).

10.2 Hometask soltuion

Solution by Evgeny Zalyubovsky

1.
Z
e−x /62 2

E(W5 ·Y ) = E(W5 | W3 ⩽ 2) = E(W5 −W3 | W3 ⩽ 2)+E(W3 | W3 ⩽ 2) = 0+ x √ dx =

−∞ 6π
Z 0 Z 4  √ r
1 6
3 −2/3
= √ e−z/6 dz + e−z/6 dz = − √ 1 + e−4/6 − 1 = − e
2 6π +∞ 0 2 π 2π

53
 Var(W5 · Y ) = E(W52 · Y ) − (E(W5 · Y ))2 = E((W5 − W3 )2 | W3 ⩽ 2)+
3e−4/3
+ 2 E(W5 · (W5 − W3 ) | W3 ⩽ 2) + E(W32 | W3 ⩽ 2) − = 3 E((W5 − W3 )2 )+
2π
Z 2 −x2 /6
2e 3e−4/3
+ 2 E(W3 · (W5 − W3 ) | W3 ⩽ 2) + x √ dx − =
−∞ 6π 2π


= 3 Var(W5 − W3 ) + (E(W5 − W3 ))2 + 2 E(W3 | W3 ⩽ 2) E(W5 − W3 )+
 Z +∞ Z 2  Z +∞
1 2 −x2 /6 2 −x2 /6 3e−4/3 1
x2 e−x /6 dx−
2
+√ − xe dx + xe dz − = 3·(2+0)+0− √
6π 0 0 2π 6π 2
−4/3 −4/3 Z +∞
3e 3e 1
x2 e−x /6 dx,
2
− =6− −√
2π 2π 6π 2
where
Z Z ! Z +∞
1 +∞
1 +∞  x  3xe−x /6
2 +∞
1
−x2 /6
3x − e−x /6 dx = 3e−x /6 dx
2 2
−√ 2
xe dx = √ √ −√
6π 2 6π 2 3 6π 2 6π 2
e−2/3  √ 
= −p − 3 + 3Φ 2/ 3 ⇒
π/6

3e−4/3 e−2/3  √  3e−4/3 e−2/3 p 

⇒ Var(W5 ·Y ) = 6− − p −3+3Φ 2/ 3 = 3− − p +3Φ 4/3
2π π/6 2π π/6

Cov(W5 , Y ) = E((W5 − E(W5 )) (Y − E(Y ))) = E(W5 (Y − P(W3 ⩽ 2))) =

p  r r
3 −2/3 3 −2/3
= E(W5 · Y ) − Φ 4/3 E(W5 ) = − e −0=− e
2π 2π

Cov(W5 , W5 · Y ) = E(W5 (W5 · Y − E(W5 · Y ))) = E(W52 · Y ) − E(W5 · Y ) E(W5 ) =

e−2/3 p 
= Var(W5 · Y ) + (E(W5 · Y ))2 − 0 = 3 − p + 3Φ 4/3
π/6
q p 
−4/3 e−2/3
E(W5 · Y ) = − 2π 3 −2/3
e ; Var(W5 · Y ) = 3 − 3e2π − √ + 3Φ 4/3 ; Cov(W5 , Y ) =
q  p  π/6
e−2/3
− 2π3 −2/3
e ; Cov(W5 , W5 · Y ) = 3 − √ + 3Φ 4/3 .
π/6

2. Consider a function f such that

E(Mn+k | Fn ) = Mn . (19)

Obviously, as An is always finite, so are f (n) and E(Mn ). Moreover, An is always adapted
to Fn while f (n) is a constant, so Mn is also always adapted to Fn . Therefore, condition (19)
is both necessary and sufficient for Mn to be a martingale. Before using that to obtain f (n),

54
 let’s derive a formula for E(An+k | Fn ):

An An An n+3
E(An+1 | Fn ) = An + = An + = An + = An ⇒
An + B n A0 + B 0 + n 2+n n+2
(n + k − 1) + 3
⇒ E(An+k | Fn ) = E(E(An+k | Fn+k−1 ) | Fn ) = E( An+k−1 | Fn ) =
(n + k − 1) + 2
(n + k − 1) + 3 Yn+i+3
k−1
n+k+2
= · E(An+k−1 | Fn ) = . . . = E(An | Fn ) · = An .
(n + k − 1) + 2 i=0
n + i + 2 n + 2

n+k+2
(19) ⇒ E(Mn+k | Fn ) = E(f (n+k)·An+k | Fn ) = f (n+k) An = Mn = f (n)An ⇒
n+2
n+k+2 1
⇒ f (n + k) = f (n) ⇒ f (n) =
n+2 n+2

First, note that An + Bn = n + 2 ∀n, so if n+2 1

An is a martingale, then so are 1 − n+2 1
Bn and
1
B . What is more, τ = minn {n | Bn = 2} ⇒ Bτ = 2 ∀τ . According to Doob’s theorem,
n+2 n
we have that E( τB+2
τ
) = E( 0+2
B0
) = 12 , but E( τB+2)
τ
= E( τ +2
2
) = 2 E( τ +2
1
), hence E( τ +2
1
) = 14 .
3. dYt = 7Wt6 dWt + 12 42Wt5 dt + aWt5 dt + t2 cos Wt dWt = (7Wt6 + t2 cos Wt ) dWt + (21 +
a)Wt5 dt ⇒ in order to have no terms with dt, we need that 21 + a = 0 ⇔ a = −21.
As for the ultimate question, due to Yt being a martingale, E(Yt ) = Y0 = 42.
4.

XT 1S0 ST >ST2 /2
−rT e


X0 = B0 Ẽ( | F0 ) = B0 Ẽ( | F 0 ) = e P S 0 S T > S 2
T /2 | F 0 =
BT B0 erT
 

e S 2 e(µ−σ2 /2)T +σWT > S 2 e(µ−σ2 /2)T +2σWT /2 | F0 = e−rT P
= e−rT P e eWT > e2WT /2 | F0 =
0 0
   
−rT e µ−r µ−r e W̃T > 2W̃T /2 =
= e P W̃T + T > 2W̃T /2 + T | F0 = e−rT P
σ σ



e W̃T − W̃T /2 > W̃T /2 = e−rT P
= e−rT P e N (0, T /2) > N (0, T /2) = 1 e−rT
2
5.

dYt = exp(γt) dXt + γ exp(γt)Xt dt + γ exp(γt) dXt dt =

= exp(γt) (b − aXt ) dt + c dWt + γ exp(γt)Xt dt =
= c exp(γt) dWt + exp(γt)(b − aXt + γXt ) dt

That term will be constant iff (a − γ)Xt ≡ 0 ⇔ γ = a.

dYt = c exp(at) dWt + b exp(at) dt ⇒

Z t Z t
⇒ Yt = X 0 + c exp(au) dWu + b exp(au) du ⇒
0
 Z t 0
Z t 
⇒ Xt = exp(−at) X0 + c exp(au) dWu + b exp(au) du
0 0

55
  Z t  Z t
E(Xt ) = exp(−at) X0 + b exp(au) du + c exp(−at) E( exp(au) dWu ) =
0 0
b
= X0 exp(−at) +
a
Var(Xt ) = E(Xt ) − (E(Xt ))2 =
2

Z t 2
= E( exp(au) dWu ) =
0
Z t
= E(exp(2au)) du =
0
exp(2at)
=
2a

10.3 Exam 2017-12-29

10.3.1 Dynamic optimization
1. [10 points] Consider a system of differential equations
(
ẋ = −2y + x(x2 + y 2 − 1);
ẏ = 2x + y(x2 + y 2 − 1).

(a) [5 points] Check that (0, 0) is the only point of rest of this system. Change the variables
from the Cartesian to the polar (r, φ). Rewrite the system in the polar coordinates.
(b) [5 points] Draw two time paths of the system for r0 > 1, where r0 is the radius of the
initial point and for 0 < r0 < 1. Justify your sketch. What if r0 = 1?
2. [10 points] Solve the calculus of variations problem
Z 2 2 
2y ′ 2
+ (y ) dx → max / min
1 x2

with the fixed endpoints y(1) = 0, y(2) = 72 .

Hint: for solving Euler’s equation try to find solutions in the form of xa . Check all available
second-order conditions.
3. [20 points] Road construction costs minimization.
Let the terrain profile be represented by the function
(
3 − 3|t|, if |t| ⩽ 1
y(t) =
0, otherwise.

The contractor minimizes the excavation costs given by the formula

Z c
(x(t) − y(t))2 dt,
−c

where x(t) is the road profile we need to find, c > 2 and [−c, c] — road section where the
excavation takes place (c is not set). Allowable grade of the road satisfies |ẋ| ⩽ 1.

56
 (a) [5 points] Form Hamiltonian and derive first-order conditions taking into account that
multiplier λ satisfies transversality conditions λ(−c) = λ(c) = 0.
(b) [15 points] Find x(t) on the section [−c, c].

10.3.2 Stochastic Calculus

Standard Wiener process is denoted by Wt .
1. [10 points] Let Y be equal to 1 if W2 > 0 and 0 otherwise.
Find E(Y | W1 ), Var(Y | W1 ), E(Y | W12 ).
2. [10 points] James Bond flips a biased coin until the sequence Head-Tail-Head-Tail appears.
The probability of «Head» is equal to p. Using Doog’s theorem and an appropriate martingale
find the expected number of coin flips.
3. [10 points] The process Xt is given by
Z t
2 2
Xt = 2017 + t Wt + u dWu
0

(a) Find dXt ;

(b) Is Xt a martingale?
(c) Find E(Xt ).
4. [10 points] Consider the framework of the Black and Scholes model. You agreed with Warren
Buffett that at fixed time T he will pay you the strange sum

XT = ln ST · ln ST /2 ,

where St is the price of a share.

What is the non-arbitrage price X0 of this agreement?
5. [20 points] Solve the stochastic differential equation

dYt = (Yt3 − Yt ) dt + Yt2 dWt

You are free to use or not to use the following guiding steps:
(a) [5 points] Consider Zt = Ytn and find dZt ;
(b) [3 points] Find a constant n such that the term before dWt in dZt is non-random.
(c) [2 points] Write down the equation for dZt in terms of t and Zt only: you should get rid
of Yt .
(d) [8 points] Solve the equation for dZt .
Hint: It’s just the particular case of the equation in your hometask :) Do you remember?
Multiply Zt by some exponent :)
(e) [2 points] Finally, find Yt .

10.4 Exam, solution

1.

E(Y | W1 ) = P(W2 > 0 | W1 ) = P(W2 − W1 > −W1 | W1 ) = 1 − F (−W1 ) = F (W1 )

57
 Var(Y | W1 ) = F (W1 )(1 − F (W1 ))
E(Y | W12 ) = P(W2 > 0 | W12 ) = 0.5F (−|W1 |) + 0.5F (|W1 |) = 0.5
2. We organise a casino. Each entrant brings one dollar into the casino. And bets all his fortune
every time. Each player bets on head, tail, head and then tail consecutively. To make the
fortune of each player a martingale we will pay him 1/p if he guesses head correctly and
1/(1 − p) if he guesses tail correctly.
The total wealth of all players, St , is not a martingale because E(St ) = t. But the process
Mt = St − t is a martingale. Let τ be the first moment when the sequence HTHT appears.
We note that M0 = 0 and Mτ = p1 1−p 1
+ p1 1−p
1 1 1
p 1−p
− τ.
According to Doob’s theorem E(Mτ ) = E(M0 ) = 0.
So
1 1 1 1 1 1
E(τ ) = +
p1−p p1−pp1−p
3.
dXt = (2tWt2 + t2 ) dt + (2t2 Wt + t) dWt
Xt is not a martingale because the term before dt is not zero.

E(Xt ) = 2017 + E(t2 Wt2 ) = 2017 + t3

4.
3
X0 = exp(−rT )Ẽ(XT | F ) = . . . = exp(−rT )(ln2 S0 + ln S0 (r−σ 2 /2)T +(r−σ 2 /2)2 T 2 /2+σ 2 T /2)
2
5.
n − 1 n+2
dZt = n(Ytn+2 − Ytn + Yt ) dt + nYtn+1 dWt
2
There are two values of n, n = 0 and n = −1. The value n = 0 is useless. Consider n = −1.

dZt = Zt dt − dWt

Let’s introduce Xt = e−t Zt :

dXt = . . . = − exp(−t) dWt

So
1
Yt = e−t Rt
1/Y0 − 0
exp(−u) dWu

11 Some junk
11.1 Problems
1. Let Yt = Wt + µt. Find all constants b such that Zt = ebYt is a martingale.

11.2 Solutions
1. Using Ito’s lemma  
bYt 1
dZt = e b dWt + µdt + b dt
2

58
 So b = 0 and b = −2µ are ok.

12 2018-2019
12.1 Stochastic calculus: home assignment
1. I throw a fair die until first six appears. Let’s denote the total number of throws by X and the
number of odd integers thrown by Y .
Find E(Y | X), E(X| Y ) and Var(Y | X).
Rt
2. Let Xt = exp(−αt) 1 + 0 exp(αu) dWu .
Rt
(a) Simplify the expression Xt + α 0
Xu du.
(b) Find E(Xt ) and Var(Xt ).
3. Today the price of a share is S0 = 100 roubles. Each day the price St goes up by one rouble
with probability p ∈ (0; 1) or by two roubles with probability 1 − p.
(a) Find a number a such that Mt = aSt is a martingale.
(b) Let τ be the first moment of time when the price will be greater or equal to 200 roubles.
Find E(τ ).
4. In the framework of Black and Scholes model find the price at time t = 0 of an asset that pays
you at time T the amount
XT = min{ST , ST2 }.
5. Consider the stochastic differential equation

dXt = 2Wt exp(−Xt ) dWt + 2t exp(−2Xt ) dt

Find at least one solution of the form Xt = h(f (Wt ) + g(t)).

12.2 Exam
Optimal Control
1. Search costs for a good c are uniformly distributed on the segment c ∈ [0; z]. Every consumer
forms her expectation about the reservation price on this good while solving the minimization
problem
π(R) = p(R) + σ(R)c min,
R>0

where R is the reservation price, p(R) is the market price of the good, σ(R) — the length of
search for this good.
(a) (6 points) Assume the problem is solved, then R is the function depending on c. Prove
that dπ/dc = σ.
(b) (7 points) Let the monopolistic producer solve the problem of setting the optimal price
Z z
pX(π, c) dc → max,
0

59
 where π satisfies the original equation and reservation price is optimal. Form the Hamil-
tonian to derive the first-order conditions for the optimal control problem, where π is
the state variable, σ the control variable and X is the demand on good.
(c) (7 points) Let X = α − βπ 2 + γc, where α > 0, β > 0 and γ are fixed parameters. Use
the Hamiltonian system derived earlier to show that γ > 0.
2. (10 points) Find the steady-state solutions of the system of differential equations:
(
ẋ = ln(2 + xy − x − y);
.
ẏ = e2x+3y−5 − 1

On the portrait plot the equilibrium points, sketch the paths after finding the eigenvalues and
eigenvectors of the linearized system.
3. (10 points) Solve the calculus of variations problem with the free-end value
Z π/2
((y ′ )2 − 2yy ′ + y 2 − 2y ′ cos2 x) dx → max / min, y(0) = 0.
0

Check the sufficiency condition in order to classify the extremal.

Stochastic Calculus
Standard Wiener process is denoted by Wt .
4. [10 points] Consider a well-shuffled deck of 52 cards. You open the cards one by one until the
second King appears. Calculate the unconditional probability that
(a) the next card will be Queen of Diamonds;
(b) the next card will be King of Diamonds.
Hint: a good martingale may be very useful here :)
5. [10 points] Find E(W2019
3
| W2018 ) and E(W2018
3
| W2019 ).
6. [10 points] Suppose Xt satisfies the stochastic differential equation

dXt = 2018Xt dt + Xt2019 dWt

Determine constants a, b and c such that Yt = exp(aXtb + ct) is a martingale.

7. [10 points] Consider the framework of the Black and Scholes model. The asset X will pay you
one share at fixed time T if ST > 2ST /2 , where St is the price of a share.
What is the non-arbitrage price X0 of this asset?
8. [20 points] Consider the stochastic differential equation

dYt = (2Yt /t + 3(t4 Yt )1/3 ) dt + 3(tYt )2/3 dWt , Y0 = 0.

(a) [15 points] Solve the stochastic differential equation.

(b) [5 points] Find E(Yt ) and Var(Yt ).
You are free to use or not to use the following guiding steps:
(c) [5 points] Suppose that Yt = g(Wt )h(t). Find dYt .
(d) [5 points] By looking at the term before dWt provide the equations for g(Wt ) and h(t).
(e) [5 points] Find g(Wt ) and h(t) and check your solution.

60
12.3 Solution, stochastic calculus part
1. Non-martingale solution: let’s bet not on the next card, but on the last one. At every moment
of time we have the same information about the next and the last cards. But for last card the
probability is always 1/52.
Martingale solution. Consider the martingale studied in the lectures: Mn — indicator whether
the Queen of Diamonds is still in the remaining part of the deck divided by the number of re-
maining cards. And stopping time, τ — moment of the second King plus one. Doob’s theorem
is satisfied as τ ⩽ 52. So E(Mτ ) = E(M0 ) = 1/52.
2. Use the decomposition W2019 3
= (W2018 + (W2019 − W2019 ))3 . For the second question first
use inversion.
E(W2019
3
| W2018 ) = W2018
3
+ 3W2018


E(W2018
3
| W2019 ) = 20183 (W2019 /2019)3 + 3W2019 /(2018 · 20192 )
3. Use Ito’s lemma. Do not forget that dXt also contains dt term.

c + 0.5ab(abXt2b−2+4038 + (b − 1)Xtb−2+4038 + 4036Xtb ) = 0

Solutions are: c = 0, a = 0, any b; c = 0, b = 0, any a; b = −4036, a = 1, c = −0.5ab(b − 1).

4.
e−rT E∗ (ST · 1ST >2ST /2 )
5.
Yt = Wt3 t2
E(Yt ) = 0; Var(Yt ) = t4 E(Wt6 ) = 15t7

13 2019-2020
13.1 Exam, 2019-12-23
Optimal Control

1. Consider a profit maximization problem when a perfectly competitive firm spends u(t) on
advertising: Z ∞
e−rt (P x(t) − u(t)) dt → max,
0

where x(t) is the share of the market, P — price of sales per unit time, r > 0 — the discount
rate. Maximization is subject to ẋ = u(1 − x) − x/2, x0 ⩾ 0 and 0 ⩽ u ⩽ 1.
(a) (5 points) Using current value Hamiltonian state the system of first-order conditions.
(b) (15 points) Find the bounds on price Pmin < P < Pmax for which the steady-state solution
(xs , us ) with xs < 1 and us < 1 exists and find it. Draw in (x, u)-plane the paths of most
rapid approach to steady-state for xs ̸= x0 .
2. (10 points) Consider a problem in discrete time

X
∞
β t U (ct ) → max
t=0

61
 √
with respect to ct subject to kt+1 = kt − ct , where k0 is given and 0 < β < 1. Suppose there
is a constant δ > 0 such that U (c) ⩽ δc for all c ⩾ 0.
(a) Write down Bellman’s equation for this problem.
(b) Prove that under assumption that Bellman’s equation has a unique solution the value
function of the problem satisfies the estimate V (k) ⩽ δk + C for some C > 0 and
provide lower bound for C.
R4
3. (10 points) Solve bounded control problem 0 (u2 + x) dt → min subject to dx/dt = u, x(0) =
0 and |u| ⩽ 1.

Stochastic Calculus
Standard Wiener process is denoted by Wt .
1. [10 points] Find a function f (t) such that Mt = exp(Wt2 /(1 + 2t))/f (t) is a martingale.
2. [10 points] Find the limit of Xn in mean squared when n tends to infinity:

X
n
Xn = (Wti/n − Wt(i−1)/n )3
i=1

3. [10 points] In the framework of the Black and Scholes model find the price at t = 0 of an asset
that pays XT = ST2 /ST /2 at time T . Here St denotes the price of one share at time t.
4. [10 points] Solve the following stochastic differential equation.

Xt
dXt = dt + dWt , X0 = 0
t−1
Rt
Hint: the solution has the form Xt = a(t) · 0 b(u)dWu for some deterministic functions a(t)
and b(t). You only need to find a(t) and b(t) :)
5. [20 points] Let τ be the first moment of time when the Wiener process Wt touches the line
y(t) = 10 − t.
The goal of this exercise is to find E(τ ) and Var(τ ).
You are free to use or not to use the following hints:
(a) Check whether the process Rt = exp(σWt − σ 2 t/2) is a martingale.
(b) Apply the Doob’s theorem and find the function m(λ) = E(exp(−λτ )). You don’t need
to check all the conditions of the Doob’s theorem.
(c) Find m′ (0), find m′′ (0).
(d) Find E(τ ), Var(τ ) using m′ (0) and m′′ (0). You can do the last point without any previous
point.

13.2 Stochastic calculus, hints and solutions

1. Using Ito’s lemma take dMt . And obtain expression of the form dMt = At dt + Bt dWt . If
Mt is a martingale then At = 0. I believe you can be attentive and obtain the equation
f ′ /f = 1/(1 + 2t).
2. We check that E(Xn ) = 0, Var(Xn ) → 0, so the limit of Xn in mean squared is zero.
3. Just calculate E∗ (exp(−rT )XT ).

62
 4. Let’s denote Xt = a(t)Yt . We note that dYt = b(t)dWt . No second order terms as dtdYt = 0.
By Ito’s Lemma

dXt = a′ (t)Yt dt + a(t)dYt = Xt a′ (t)/a(t)dt + a(t)b(t)dWt

Hence we get the system (

a′ (t)/a(t) = 1/(t − 1),
a(t)b(b) = 1
5. m′ (λ) = E(−τ exp(−λτ )), so m′ (0) = E(−τ ), by the same reasoning m′′ (0) = E(τ 2 ).

14 2020-2021
14.1 Exam, 2020-12-29
Short rules: online with proctoring, 120 minutes, A4 cheat sheet and calculator are allowed.
1. (10 points) Ded Moroz would like to receive K roubles at time T = 2 if S2 < 0.5S1 and zero
otherwise.
Assume the framework of Black and Scholes model, St is the share price, r is the risk free rate,
σ is the volatility.
How much Ded Moroz should pay now at t = 0?
2. (10 points) Simplify as much as possible the integral
Z t
exp(−Wu − u/2)dWu .
0

3. (10 points) The process Yt is defined by

dYt = Wt2 dt + Wt dWt , Y0 = 0.

(a) (6 points) Find E(Yt ), E(Yt Wt ), E(Yt Wt2 ).

(b) (4 points) Find Var(Yt ).
4. (10 points) Find at least one solution of the stochastic differential equation

dRt = 4Rt dt + 7dWt , R0 = 1.

You are free to use the following steps:

(a) Solve deterministic equation dQt = 4Qt dt.
(b) Now you need to «remove» this deterministic solution Qt from Rt . To accomplish this
goal represent Rt as Rt = Qt Bt and find a very simple equation for Bt .
(c) Solve the equation for Bt .
(d) Finalize your solution.
5. (10 points) The variables X and Y are independent and are exponentially distributed with
mean 1. Let L = min{X, Y } and R = max{X, Y }.
Find E(L | R) and E(R | L).

Be brave‼ One more problem is waiting to kill you‼ The final boss of this game‼

63
 6. (20 points) Consider a process St = Z1 + Z2 + . . . + Zt with S0 = 0. Increments Zt are
identically distributed and take values (−1) and 1 with equal probabilities.
Let τ be the first moment of time when |St | = 100.
(a) Find f (λ) such that Mt = f (λ)t exp(λSt ) is a martingale.
(b) Are Sτ and τ independent?
(c) Using Doob’s optional stopping time theorem obtain q(λ) = E(f (λ)τ ).
(d) By explicitely finding G(f ) = E(f τ ) or otherwise find G′ (1).
(e) Express E(τ ) in terms of G′ (1).
Hint: you may not check technical conditions of Doob’s optional stopping time theorem.

14.2 Marking scheme, exam 2020-12-29

1. formula for St with Wt∗ and r = 1 point formula for X0 with E ∗ = 1 points from expectation
to K exp() P∗ (event) = 4 points from P∗ (event) to normal cdf = 4 points
2. guessed answer = 2 points, guess check = 8 points
3. three expected values = 2 points * 3 = 6 points, variance = 4 points
4. 4a = 2 points, 4b = 3 points, 4c = 3 points, 4d = 2 points
5. joint density of min and max = 4 points, E(min | max) = 3 points E(max | min) = 3 points
6. (a) f (λ) = 2/(exp(λ) + exp(−λ)).
(b) Yes. Possible argument: if you multiply St by (−1) the distribution of τ will be unaf-
fected, but Sτ will be multiplied by (−1).
(c) G′ (1) = E(τ f τ −1 ) = E(τ ).

15 2021 Fall
15.1 Exam, 2021-12-21
1. (10 points) Ded Moroz would like to receive one share at time T if ST > ST /2 and nothing
otherwise.
Assume the framework of Black and Scholes model, St is the share price, r is the risk free rate,
σ is the volatility.
How much Ded Moroz should pay now at t = 0?
2. (10 points) Let X0 = 0.2021. Next values of Xt and Yt are defined recursively. The filtration
is given by Ft = σ(X0 , . . . , Xt ).
The random variable Yt given Ft−1 has Bernoulli distribution with P(Yt = 1 | Ft−1 ) = Xt−1 .
And Xt = (Yt + Xt−1 )/2.
(a) Check whether Xt is a martingale.
(b) The sequence (Xt ) converges almost surely (you don’t need to prove this). What is the
distribution of the limit?
Rt Rt
3. (10 points) Consider two integrals: It = 0 Wu2 dWu and Jt = 0 Wu2 du.
(a) Find E(It ) and E(Jt ).
(b) Find Var(It ) and Var(Jt ).

64
 4. (10 points) You pick up cards one by one from a well shuffled deck of 52 cards. Let Xi be an
indicator variable, equal to one if the card number i is a Queen, and zero otherwise.
Let Yi = E(X52 | X1 , X2 , . . . , Xi ).
Find the covariance Cov(Y50 , Y51 ).
5. (10 points) Imagine the share price satisfies the equation
p
dSt = St dt + St dWt , S0 = 1.

Find all transformations Yt = h(St ) such that Yt is a martingale.

6. (20 points) Consider two stopping times: U = min{t | Wt = 2} and D = min{t | Wt = −1}.
The ultimate goal of this problem is to find M (s) = E(e−sU | U < D) and explain how to
obtain E(U | U < D) and Var(U | U < D).
You may use or not use the following guiding steps.
(a) (3 points) Is Wt a martingale? Using optional stopping time theorem find P(U < D).
(b) (3 points) Is exp(aWt −a2 t/2) a martingale? State the equation that follows from optional
stopping time theorem.
(c) (3 points) From the equation obtained
√ previously
√ get a system of two equations by con-
sidering two values of a: a = 2s and a = − 2s.
(d) (5 points) Find E(e−sU · I(U < D)) where I is indicator of the corresponding event.
(e) (3 points) Find the function M (s) = E(e−sU | U < D).
(f) (3 points) Explain how to find E(U | U < D) and Var(U | U < D) using M (s). (You
don’t need to do actual calculations).
You may not check technical conditions of Doob’s optional stopping time theorem.

15.2 Marking scheme, 2021-12-21

1. One dollar and not one share has been considered: -3 points. The random variable ST present
in final answer: -5 points. Variables ST and indicator assumed independent: -3 points.
2. 2a = 5 points, 2b = 5 points.
3. E(It ) = 2 points, E(Jt ) = 3 points, Var(It ) = 2 points, Var(Jt ) = 3 points.
4. • Possible values of Y51 : 2 points;
• Possible values of Y50 : 2 points;
• Table of probabilities: 3 points;
• E(Y51 ): 1 point;
• E(Y51 ): 1 point;
• Covariance: 1 point.
5. • correct differential equation for h: 6 points;
• solution of differential equation: 4 points;
• common error, 1/2 instead of 2: -2 points.
6. Points assigned in the text.

16 2022 Fall
16.1 Exam, 2022-12-29
Rules: 120 minutes, A4 cheat sheet and calculator are ok, Wt denotes a Wiener process, 10 points
for each problem.

65
1. Consider Xt = exp(−2Wt − 2t).
(a) Find dXt . Write the corresponding full form. Is Xt a martingale?
(b) Find E(Xt ) and Var(Xt ).
Rt
(c) Find 0 Xu dWu .
2. Let (Wt ) be a Wiener process.
(a) Find E(W5 W4 | W4 ), Var(W5 W4 | W4 ).
(b) Find E(W5 W4 W3 | W4 ), Var(W5 W4 W3 | W4 ).
3. Winnie-the-Pooh starts at zero of the real line. Every minute he moves left by one with prob-
ability 0.2, right by one with probability 0.2 or does not move and eats a honey-pot with
probability 0.6.
Let Xt be the coordinate of Winnie at time t. Rabbit’s hole has x-coordinate 6 and Owl lives
at (−4), let τ be the first moment when Winnie visits Owl or Rabbit, τ = min{t | Xt =
6 or Xt = −4}.
(a) Is Xt a martingale?
(b) Find a constant a such that Yt = Xt2 − at is a martingale.
(c) Find P(Xτ = 6) and E(τ ).
Hint: you don’t need to check technical conditions of Doob’s optional stopping theorem in
point (c).
4. Consider the Vasicek interest rate model

dRt = (0.05 − Rt )dt + dWt , R0 = 0.07.

(a) Write the same models in full form with integrals.

(b) Let bt = E(Rt ). Find bt and sketch it.
Hint: to find bt take expected value of both sides of the full form and solve the ordinary
differential equation.
5. Consider two-period binomial model with initial share price S0 = 600, Up and down multi-
pliers are u = 1.2, d = 0.8, risk-free interest rate is r = 0.05 per period.
Consider an option that pays you X2 = 100 at T = 2 if S2 > S1 and nothing otherwise.
(a) Find the risk neutral probabilities.
(b) Find the current price X0 of the asset.
(c) How much shares should I have at t = 1 in the «down» state of the world to replicate
the option?
6. Consider Black and Scholes model in continuous time with risk-free interest rate r, volatility
σ, initial share price S0 and exponential growth rate of expected share price µ.
(a) Let’s denote by P the real probability and by P∗ the risk-neutral probability. Find P(S2 >
S1 ) and P∗ (S2 > S1 ).
√
(b) An option pays you the sum X2 = S1 at T = 2. Find the non-arbitrage price X0 of
the option.
Hint: the answers may contain the standard normal cumulative distribution function F ().

66
16.2 Marking scheme, 2022-12-29
1. Process Xt is a martingale as

dXt = −2 exp(−2Wt − 2t)dWt .

Full form: Z t
Xt = 1 + (−2) exp(−2Wu − 2u)dWu ,
0

Hence, E(Xt ) = 1. Use Ito’s isometry for variance:

Z t Z t
Var(Xt ) = E((−2) exp(−4Wu − 4u))du =
2
4 exp(4u)du = exp(4t) − 1.
0 0
Z t
1 1
exp(−2Wu − 2u)dWu = − exp(−2Wt − 2t).
0 2 2
Marking: 3 points for dXt , 1 point for martingale, 1 point for full form, 1 point for E(Xt ), 2
points for variance and 2 points for integral.
2. 2 points for each expected value, 3 points for each variance: E(W5 W4 | W4 ) = W42 ,
Var(W5 W4 | W4 ) = W42 .
For point (b) remark that W5 = W4 + X and W3 = 43 W4 + Y , where X, Y and W4 are
independent normal with Var(X) = 1, Var(Y ) = 3 − 94 = 34 .
E(W5 W4 W3 | W4 ) = 34 W43 , Var(W5 W4 W3 | W4 ) = 21
16
W44 + 43 W42 .
3. Process Xt is a martingale (2 points); for α = 0.4 (3 points) Yt = Xt2 − 0.4t is a martingale.
Using Doob’s theorem and two martingales we get two equations, E(Xτ ) = X0 = 0, E(Xτ2 ) =
0.4 E(τ ).
Hence, P(Xτ = 6) = 0.4 (2 points) and E(τ ) = 60 (3 points).
4. 2 points for full form Z t
Rt = 0.07 + (0.05 − Ru )du + Wt .
0
Integral equation for bt : Z t
bt = 0.07 + (0.05 − bu )du
0
Differential equation:
b′t = 0.05 − bt
And bt = 0.05 + 0.02 exp(−t).
To find bt one may also solve the SDE and take the expectation.
Marking: 6 points for bt and 2 points for a sketch.
5. 2 points for probabilities p∗u = 5/8, p∗d = 3/8; 4 points for
5 1
X0 = 100 · · ≈ 56.7
8 (1 + r)2

4 points for
X2du − X2dd 100
α= = ≈ 0.52
S 2 − S2
du dd 192

67
 6. 2 points for each probability, 6 points for X0 ;
µ σ  r σ
P(S2 > S1 ) = F − , P∗ (S2 > S1 ) = F − ;
σ 2 σ 2
p
X0 = S0 exp(−3r/2 − σ 2 /8);

17 2023 Fall
17.1 Exam, 2023-12-27
Rules: 120 minutes, A4 cheat sheet and calculator are ok, (Wt ) denotes a Wiener process.
1. [10 points] Winnie-the-Pooh starts at zero of the real line. Every minute he moves left by one
with probability 0.2, right by one with probability 0.8. Let Xt be the coordinate of Winnie at
time t.
The stopping time τ is the first moment when Winnie visits (−4) or 6, τ = min{t | Xt =
6 or Xt = −4}.
(a) [5] Find all martingales of the form Mt = exp(bXt ) where b is a constant.
(b) [5] Using Doob’s theorem or otherwise find P(Xτ = 6).
Hint: You may not check all the conditions of the Doob’s theorem.
2. [10 points] Consider the process dXt = Wt dWt + Wt dt with X0 = 2024.
(a) [3] Find E(Xt ). Is (Xt ) a martingale?
(b) [3] Find dYt for Yt = Xt Wt .
(c) [4] Find E(Yt ).
3. [10 points] Solve the stochastic differential equation
p
dXt = (2Xt + 1)dt + 2 Xt dWt , X0 = 1.

You are free to use or to completely ignore the following hints:

√
(a) [3] Simplify the equation using substitution Yt = Xt .
(b) [7] Solve the simplified equation using substitution Zt = h(t)Yt with reasonable choice
of h.
Rt
4. [10 points] Consider the process Yt = 0 sign(cos u)dWu .
(a) [5] Find the covariance Cov(Y2024π , W2024π ).
(b) [5] Is (Yt ) a Wiener process?
Hint: sign c = 1 if c > 0, sign c = −1 if c < 0 and 0 otherwise.
5. [10 points] You have two correlated Wiener processes, (At ) and (Bt ), with Corr(At −As , Bt −
Bs ) = ρ for all t > s.
Split the time interval [0; t] into n small segments of equal length. Let ∆A
i be the increment of
the Wiener process (At ) on the i-th small segment, i.e. ∆A i = A(it/n) − A((i − 1)t/n). Let
∆i be the increment of the Wiener processP(Bt ) on the i-th small segment.
B

Consider the sum of cross-products, Sn = ni=1 ∆A i ∆i .

B

(a) [3] Find E(Sn ).

(b) [5] Find limn→∞ Var(Sn ).

68
 (c) [2] Find the mean square limit of (Sn ).
(d) [1] How would you write the limit in (c) using short hand notation with dAt and dBt ?
6. [10 points] Consider Black and Scholes model in continuous time with risk-free interest rate
r, volatility σ, initial share price S0 and exponential growth rate of expected share price µ.
An option A pays you 1 dollar at fixed time moment T if ST /ST /2 > ST /2 /S0 and 0 otherwise.
Find the non-arbitrage price X0 of the option A.
Hint: the answer may contain the standard normal cumulative distribution function F ().

17.2 Marking scheme, 2023-12-27

1. [10 points]
(a) [5] Let’s denote Xt − Xt−1 by Zt . The process (Xt ) should be a martingale,

E(exp(bXt + bZt+1 ) | Ft ) = exp(bXt ).

We obtain an equation [+2 points]

E(exp(bZt+1 )) = 1.

That is 0.8 exp(b) + 0.2 exp(−b) = 1 [+2 points] and b = −2 ln 2 or b = 0.

(b) [5] From Doob’s theorem we have E(Mτ ) = 1. The equation for p = P(Xτ = 6) is

p exp(−12 ln 2) + (1 − p) exp(−8 ln 2) = 1.

Hint: You may not check all the conditions of the Doob’s theorem.
2. [10 points]
(a) [3] Z  Z
t t
E(Xt ) = 2024 + E Wu dWu + E(Wu )du = 2024.
0 0

However (Xt ) is not a martingale as it has non-zero dt term.

(b) [3]

d(Xt Wt ) = Wt dXt + Xt dWt + 0.5 · 2dWt dXt = Wt2 dWt + Wt2 dt + Xt dWt + Wt dt

(c) [4] Z t
E(Xt Wt ) = 0 + E(Wu2 + Wu )du = t2 /2.
0

3. [10 points]
(a) [3] Using Ito’s lemma we obtain

dYt = 0.5Xt−0.5 dXt − 0.125Xt−1.5 (dXt )2 = Xt0.5 dt + dWt

Hence, the equation for Yt is dYt = Yt dt + dWt .

(b) [7] Using Ito’s lemma once again we obtain

dZt = h′ (t)Yt dt + h(t)dYt = (h′ (t)Yt + h(t)Yt )dt + h(t)dWt .

69
 If we choose h′ (t) = −h(t) then the dt term will be zero. Let’s do so, h(t) = exp(−t).
For Zt = exp(−t)Yt we have dZt = exp(−t)dWt .
Z t
Zt = Z0 + exp(−u)dWu
0

From X0 = 1 we get Z0 = 1. Finally,

 Z t 2
Xt = Yt2 = exp(2t)Zt2 = exp(2t) 1 + exp(−u)dWu .
0

4. (a) [5] One way to find Cov(Y2024π , W2024π ) is to use Ito’s isometry,
Z t Z t Z t
Cov( Au dWu , 1dWu ) = E(sign(cos u))du
0 0 0

The value 2024π contains 1012 full 2π periods. Hence the integral is zero.
(b) [5] Yes, (Yt ) is a Wiener process.
5. [10 points]
(a) [3]
E(Sn ) = n E(∆A
i ∆i ) = n Cov(∆i , ∆i ) = nρ(t − 0)/n = ρt
B A B

(b) [5]
lim Var(Sn ) = 0.
n→∞

(c) [2] The random sequence (Sn ) tends to ρt in mean square and in probability.
(d) [1] dAt dBt = ρdt.
6. [10 points] Recall that St = S0 exp((r − σ 2 /2)t + σWt∗ ). The event

B = ST /ST /2 > ST /2 /S0

may be written as B = {WT∗ > 2WT∗/2 } [5 points].

Hence,
X0 = exp(−rT ) P∗ (WT∗ > 2WT∗/2 ) = exp(−rT )/2.

70