Mathematics for Economics and Finance

Final Exam
Instructor: Norman Schürhoff

Date: 11.01.2019

Last Name, First Name:

Instructions:

• Exam paper is provided by the proctors. No own paper is allowed. You must turn
in all sheets at the end of the exam, including any scratch paper. All exam questions
must be returned.

• The exam includes 4 questions. Tell the proctor immediately if something is missing.
Start your answers to every question at the top of a new page in the ‘Solution’ part.

• The duration of the exam is 180 minutes. If you arrive late for the exam, you must
finish with all other students. No extension can be granted.

• This is a closed-book exam. You are allowed to use a black or blue ink pen, a
non-programmable calculator, and a dictionary only. The pen must not write
red or green. No pencils are allowed. The dictionary may not have any markings.
You are allowed no other material, including smartphones, computers, smart watches,
class material, books, cheat sheets.

• Show all derivations and computations. Numerical results without corresponding ana-
lytical derivations receive no points except if stated otherwise. The solution approach
has to be written legibly and clear to the grader.

• The total number of points is 100.

1
Question I (25 points)
1. Determine if the following statements are true or false and briefly justify your answer
for each statement.

(a) In any Lagrange optimization problem under equality constraints the Lagrange
multipliers are always nonnegative.
(b) An estimate is a random variable.
(c) For any random variable X and constant a it holds that Pr(X = a) = 0.
(d) The limit of f (x) as x → 0 exists and is equal to 0 if
(
1, when x ≥ 0
f (x) =
−1 , when x < 0.

(e) If x∗ is a local extremum of f (x), x ∈ A, then D(f (x∗ )) = 0.

(f) Any maximum likelihood estimator θb is asymptotically unbiased.
(g) If two events are mutually exclusive, then they are also independent.
(h) If an estimator is consistent, then it is also unbiased.

2. Compute the following expressions:

(a)
x+1
The derivative of f (x) = using the definition of a derivative,
2−x
(b)

x1
lim ex + x2 ,
x→∞

(c)
Z e2
1
√ dx.
0 x+ x

2
Question I – Solution
1. 2 points for each (correct statement + correct explanation)

(a) False. E.g. see section 6.6.3 in the Lecture Notes.

(b) False. An estimate is a realization of an estimator, which is a random variable.
(c) False. Consider any discrete distribution.
(d) False. The one-sided limits are not equal, meaning the limit doesn’t exist.
(e) False. x∗ may be a boundary point. Alternatively: gradient may not exist.
(f) True. MLE is asymptotically normal with mean equal to the true value of the
parameter.
(g) False. P (A ∩ B) = 0 if A and B are mutually exclusive. If they are independent,
then P (A ∩ B) = P (A)P (B), which is > 0 in general.
(h) False. Consider e.g. the ML estimator for σ 2 of N (µ, σ 2 ).

2. 3 points each

(a)
(x+h)+1
0 f (x + h) − f (x) 2−(x+h)
− x+1
2−x
f (x) = lim = lim
h→0 h h→0 h
(x + h − 1)(2 − x) − (x + 1)(2 − x − h) 1 3h 3
= lim = lim = .
h→0 (2 − x − h)(2 − x) h h→0 (2 − x − h)(2 − x)h (2 − x)2
(b)    h i
x 2

x1 0 1 x 2 ∞ H
lim e + x = [∞ ] = exp lim log(e + x ) = =
x→∞ x→∞ x ∞
!
1
(ex + 2x) ex +x ex + 2
h∞i   h i
2 H ∞ H
exp lim = = exp lim x = =
x→∞ 1 ∞ x→∞ e + 2x ∞
ex
   
1
exp lim x = exp lim = exp(1).
x→∞ e + 2 x→∞ 1 + 2/ex

(c)
 √ 
 t= x 
Ze2  dt = √1 dx  Ze
 
dx  2 x  2tdt
√ =  2tdt = dx  =
x+ x  t2 + t
 x = 1, t = 1 

1 1
 x = e2 , t = e 
Ze
dt
=2 = 2 ln |t + 1||e1
t+1
1
= 2 (ln (e + 1) − ln 2) .

3
Question II (25 points)
There are m different states of nature and n assets in the market. The market can be
represented by the payoff matrix X, where each row corresponds to a state and each column
shows how much an asset pays in each state:
 
x11 . . . x1n
X =  ... ..  .

. 
xm1 . . . xmn
We say that the market is complete if by means of existing assets one can construct a portfolio
that gives any desired combination of payoffs in different states. If there are more assets
than needed to construct such a portfolio, then the unused assets are called redundant.
1. Briefly answer the following questions. Justify your answers.
(a) If the market is complete, what is the relationship between m and n (that is,
n ≤ m, m ≤ n, or no relation between m and n)?
(b) What can you say about the rank of X if the market is complete?
(c) Provide a condition on X ensuring that there are no redundant assets.
(d) Suppose there are n assets. If two assets xa and xb are such that xa + xb =
(1, 1, . . . , 1)0 , with xi ∈ Rn , can the market be complete?
(e) Give an example of a payoff matrix characterizing a complete market in which
there are redundant assets.
Consider two markets U and Z with the following payoff matrices:
   
2 2 2 1 1
U= 1 −3 −2 , Z = 3 1 .
1 2 3 5 3
2. Show that market U is complete by using the eigenvalues of U . Can you say something
about the definiteness of U with no further computations? Hint: one of the eigenvalues
is equal to 1.
3. Your client asks you to construct a portfolio. You want to determine the allocation
in each asset in market U , denoted θU ∈ R3 given the target payoff b = (b1 ; b2 ; b3 )0 in
each state of nature. Provide the set of allocations that you can offer to your client.
4. Is the market Z complete? Are there any redundant assets? Justify your answer.
5. Suppose now that you want to determine the portfolio allocation in each asset in market
Z, denoted θZ ∈ R2 , given the target payoff b = (b1 ; b2 ; b3 )0 in each state of nature.
What is the set of payoffs B that you can provide to your client in this market?
Suppose now that the states have probabilities (1/2, 1/4, 1/4) in market Z.
6. For each asset in market Z, compute the expected payoff and variance of the payoffs
as well as the correlation coefficient of payoffs between all assets.
7. If you could only invest in one asset in market Z, which would you choose? Justify
your answer.

4
Question II – Solution
1. 1 point each

(a) If the market is complete, we need to have at least as many assets as there are
states in the economy (n ≥ m). Otherwise (if n < m) we are not able to replicate
any desired combination of payoffs. However, if there are only k, k < n linearly
independent columns, then we cannot build a basis and the market is incomplete,
no matter whether m > n, m = n, or m < n.
(b) If the market is complete we have rank(X) = m, even if n ≥ m. In case when
rank(X) < m, we are not able to replicate all combinations of assets. In other
words, rank(X) = m is a necessary and sufficient condition for the market to be
complete.
(c) There are no redundant assets if rank(X) = n.
(d) Yes, it is possible that the market is complete, e.g. consider assets in column 1
and 2 in the following market X:
 
1 0 0
X = 0 1 0 .
1 0 1

(e) One example of such a market is:

 
1 0 0 0
X = 0 1 0 1 .
1 0 1 1

Note that rank(X) = 3.

√ √
2. The eigenvalues are λ1 = 1, λ2 = (1 + 41)/2, λ3 = −(1 − 41)/2. The market is
complete, because det(U ) = λ1 λ2 λ3 6= 0 ⇒ rank(U ) = 3. The definitess of the matrix
cannot be determined without further computations because it is not symmetric. 4
points

3. 3 points
1 1

− 15  θ1 = 12 b1 + 15 b2 − 15 b3
 
2 5
U θ = b ⇒ θ = (U )−1 b =  1
2
− 25 −5 3 
b⇒ θ2 = 12 b1 − 25 b2 + 35 b3
− 12 1 4
θ3 = − 12 b1 + 15 b2 + 45 b3

5 5

Thus, the set of allocations contains a single vector.

4. The market Z is not complete, because there are less assets than there are the states
of the world. There are no redundant assets, because neither can be expressed as a
linear combination of the other. 3 points

5. We have to solve the following system in terms of θ1 and θ2 :

5
 
 b1 = θ1 + θ2 1
b2 = 3θ1 + θ2 ⇐⇒ b3 = − b1 + b2 .
2
b3 = 5θ1 + 3θ2


This expression is just a hyperplane in R3 (a plane), so the vector being the solution
of the system of equations (i.e. describing all possible payoffs attainable) is:
       
b1 b1 1 0
b = b2  =  b2  =  0  b1 + 1 b2 , with b1 , b2 ∈ R.
b3 − 21 b1 + b2 − 21 1
This is just the linear span of the two vectors:
    
 1 0 
 0  , 1 .
− 12 1
 

4 points

6. Expected payoffs: µ1 = 1/2 + 3/4 + 5/4 = 5/2, µ2 = 1/2 + 1/4 + 3/4 = 3/2.
Variances of payoffs: σ12 = 1/2(1 − 5/2)2 + 1/4(3 − 5/2)2 + 1/4(5 − 5/2)2 = 11/4
σ22 = 3/4.
Correlation coefficient: ρ = µ12σ−µ 1 µ2
1 σ2
= (4 − 3.75)/(1.66 × 0.87) ≈ 0.17, where µ12 is
the expectated payoff of the product of the two assets. 3 points

7. It is rational to say that you would choose asset 1, because it pays more or the same
in every state of nature. 3 points

6
Question III (25 points)
You are a HEC student and you have to allocate your precious free time between studying
mathematics and econometrics. The number of hours spent on studying each subject is
denoted by x and y, respectively. Your utility function is

U (x, y) = exp (min{x, ay}) ,

where a > 0, and your resource constraint is

Rx + M y ≤ W.

Let D be the constraint set, that is, the set of all feasible time allocation bundles.

1. Give an interpretation to the functional form of the utility function and the parameters
a, R, M , and W .

2. Formulate the optimization problem. What does the constraint set D look like?

3. Is constraint qualification satisfied? What does it imply?

4. Is the maximum of U (x, y) achieved on D? Explain why or why not.

5. Is the solution unique? Provide an argument without solving the problem.

6. Can the Kuhn-Tucker method be applied to this optimization problem? If yes, formu-
late Kuhn-Tucker conditions. If not, explain why.

7. Solve the optimization problem.

Hint: before solving the problem, think intuitively which composition results in the
highest utility.

8. How do the solution and the value function depend on (R, M, W )? What are the
properties of the value function and the solution in terms of W ?

7
Question III – Solution
1. a is a desirable proportion of effort utilized on y and x. R is the relative price/ effort
associated with studying mathematics. M is the relative price/ effort associated with
studying data science. W is your maximum effort available. The utility function
implies that x and y are complementary. 3 points

2. (
U (x, y) = emin{x,ay} −→ maxx,y ,
s.t. Rx + M y 6 W, x > 0, y > 0.
For simplicity, let’s discuss V (x, y) = ln U (x, y). As f (z) = ln z is an increasing
monotone differentiable function, it will not affect the solution.
(
V (x, y) = min{x, ay} −→ maxx,y ,
s.t. Rx + M y 6 W, x > 0, y > 0.
n o
The constraint set is D = (x, y) ∈ R : Rx + M y 6 W, x > 0, y > 0 .
3 points

3. Yes, it is, since the constraints are linear. It implies that the KT conditions, if KT
method can be applied, are neccessary for an optimal solution. 3 points

4. It is, by Weierstrass theorem. The objective function V (U ) is continuous and the

constraint set D is compact (closed and bounded becaue they are inequalities). 3
points

5. The problem admits a unique solution, because the transformed objective function is
concave (a minimum of two concave functions is concave), thus the problem has a
unique global maximum. 3 points

6. It cannot, because the objective function is not differentiable at optimal points. 3

points

7. Given the form of the unitility function and the corresponding indifference curves, we
know that the optimal points form a ray x = ay, y > 0. Substituting x in the budget
constraint, we get Ray + M y = (Ra + M )y = W . Thus,
(
y ∗ = Ra+M
W
,
∗ aW
x = Ra+M .

4 points

8. The value function is

n aW W o aW
v(R, M, W ) = V (x∗ , y ∗ ) = min{x∗ , y ∗ } = min ,a = .
Ra + M Ra + M Ra + M
aW
u(R, M, W ) = exp(v(R, M, W )) = e Ra+M .

8
To find the properties of the solution and the value function, we need to compute their
derivatives w.r.t. the corresponding parameters:
∂x∗ a W 2 ∂x∗ aW ∂x∗ a
∂R
= − (Ra+M )2
< 0, ∂M
= − (Ra+M )2
< 0, ∂W
= Ra+M
> 0,
∂y ∗ aW ∂y ∗ W ∂y ∗ 1
∂R
= − (Ra+M )2
< 0, ∂M
= − (Ra+M )2
< 0, ∂W
= Ra+M
>0
∂u 2
a W ∂u aW ∂u a
∂R
= − (Ra+M )2
u < 0, ∂M
= − (Ra+M )2
u < 0, ∂W
= Ra+M
u > 0.

All three functions increase with the income and decrease with both goods’ prices. All
three functions are linear in W . 3 points

9
Question IV (25 points)
As a trader in a large bank, you were tasked with investigating the statistical properties
of the payoffs of some financial asset. At your disposal are historical realizations of the
payoffs of the asset in a form of an i.i.d. sample (X1 , . . . , Xn ) of a random variable X that
you believe best describes the payoffs. You deduce that X has the following cumulative
distribution function (cdf):
 2
x
FX (x; θ) = 1 − exp − , x > 0.
2θ

You also know that the variance of the payoffs is equal to

 
4−π
V (X) = θ
2

and that
E(X 4 ) = 8θ2 .
√
1. Compute the probability that the payoff of the asset is greater than 2. How does it
change if you increase θ?

2. Prove that the expected value of X is equal to θ π2 .

p

R∞ 1
p
Hint: use the fact that 0
exp(−at2 )dt = 2
π/a.

Suppose that you want to estimate the parameter θ that governs the distribution of payoffs.

3. Derive the method of moments estimator of θ based on the first moment.

4. Compute the bias of the method of moments estimator. Is the estimator unbiased? Is
it asymptotically unbiased?

5. Describe two ways in which you would compute the standard error of the method of
moments estimator (do not carry out any computations).

6. Derive the maximum likelihood estimator of θ.

7. Compute the bias of the maximum likelihood estimator. Is the estimator unbiased?
Is it asymptotically unbiased? What is the asymptotic distribution of the maximum
likelihood estimator?

8. Compute the Cramer-Rao lower bound. Can you make a statement about the opti-
mality of the maximum likelihood estimator? If so, in what sense is the estimator
optimal?

10
Question IV – Solution
1.
√ √ √
   
2 1
Pr(X > 2) = 1 − FX ( 2) = exp − = exp − ; lim Pr(X > 2) = 1.
2θ θ θ→∞

3 points

2.  
u=x
Z ∞ Z ∞
x
  2
  du =1
 − x2θ 2
 
E(X) = xf (x; θ)dx = x e
dx =  x
 dv = x e − 2θ 

0 0 θ θ 
2
  
− x2θ
v = −e
s   √
 2  ∞
2√
 Z ∞  2 r
− x2θ − x2θ 1 1 θπ
= −xe + e dx = π/ = θπ = .
0 0 2 2θ 2 2
| {z }
l’Hopital⇒0−0=0

3 points

3. We obtain this estimator by equating the first moment of the distribution with the
empirical first moment (i.e., the sample mean). Therefore:
r
π 2 2
θ = X ⇒ θ̂M M = X .
2 π
3 points

4. We know that:  
2 2 2
E(θ̂M M ) = E X = (E(X)2 + V(X)).
π π
Now, we know that V(X) = V(X1 )/n = n1 (E(X12 ) − E(X1 )2 ) and E(X) = E(X1 ).
Therefore,
2 2θ − π2 θ π
   
4−π
E(θ̂M M ) = + θ = + 1 θ.
π n 2 πn
So, the bias of θ̂M M is:
4−π
bias(θ̂M M ) = E(θ̂M M ) − θ = .
πn
Clearly the estimator is biased. However, the bias converges to zero as n → ∞, showing
that the estimator is asymptotically unbiased. 4 points

5. One can get the standard error directly, by e.g. directly computing V ar(θ̂M M ), as well
as by using some numerical approach, e.g. by Monte Carlo simulation. In this method,
one would simulate a many random samples and compute the simulated estimates,
which could then be used to determine the standard error of the proposed estimator.
3 points

11
6. The likelihood function is given by:
n n
! n
!
Xi2
 
Y Xi Y 1 X
L(θ, X) = exp − = θ−n Xi exp − Xi2 .
i=1
θ 2θ i=1
2θ i=1

Therefore, the log-likelihood function is:

n
! n
Y 1 X 2
`(θ, X) = −n log(θ) + log Xi − X .
i=1
2θ i=1 i
Taking the FOC gives the maximum likelihood estimator of θ:
n n
∂`(θ, X) 1 1 X 2 1 X 2
= −n + 2 X =0 ⇒ θ̂M L = X .
∂θ θ 2θ i=1 i 2n i=1 i
3 points
7. We know that: !
n
1 X 2 1
E(θ̂M L ) = E X = E(X12 ) = θ.
2n i=1 i 2
Therefore, the ML estimator is unbiased and naturally also asymptotically unbiased.
We know that θ̂M L is asymptotically normal with:
√
 
d 1
n(θ̂M L − θ) −→ N 0, .
I(θ)
3 points
8. Let’s first compute the Fisher information number for a single observation:
 2
X12
  
∂ `(θ; X1 ) 1 1 2 1
I(θ) = E − 2
= E − 2 + 3 = − 2 + 2 = 2.
∂θ θ θ θ θ θ
The Cramer-Rao lower bound tells us that, for any unbiased estimator θ̂ of θ, we have:
1 θ2
V(θ̂) ≥ = .
nI(θ) n
The ML estimator is unbiased and if its variance matches the above lower bound, then
we know that it will also be the best unbiased estimator of θ. Therefore we have to
compute V(θ̂M L ):
n
!
1 X 2
V(θ̂M L ) = X
2n i=1 i
n
!
1 X
= V Xi2
4n2 i=1
1
= 2
nV(X12 )
4n
1   2
2 


2 2
= E X1 − E X1
4n
1
θ2
= 8θ2 − (2θ)2 = .
4n n

12
Therefore we can conclude that the ML estimator is the best unbiased estimator of θ.
3 points

13