Mathematics for Economics and Finance (Fall 2023)

Assignment 1 Solutions
Professor: Norman Schürhoff
Due date: 5 November 2023

Section I (25 points)

1. Look at the following pairs of statements and determine if they are logically equivalent.

(a) ˜(p∧q), (˜p) ∨ (˜q),

(b) (p ⇒ q)∧(q ⇒ r), p ⇒ r,
(c) p ∨ (q ∧ r); (p ∨ q) ∧ (p ∨ r),
(d) A ⇒ B ⇒ ˜C, C ⇒ ˜(A ⇒ B).

2. Prove that function h : R → R, defined by h(x) = x3 − x, is non-injective.

3. Discuss the properties of injectivity, surjectivity, and bijectivity for the function

f : R → R, f (x) = ex .

4. Use one of the proof methods seen in class to prove the following propositions:
(a) Let A, B ⊆ RN be convex sets. Then A ∩ B is convex. Prove that A ∪ B doesn’t have to be
convex.
(b) Prove the following statement for n ≥ 1 using mathematical induction:
n
X n(2n − 1)(2n + 1)
(2k − 1)2 =
3
k=1

Solution

1. (a) Yes, the truth table is:

p q p∧q ˜(p∧q) ˜p ˜q (˜p) ∨ (˜q)

T T T F F F F
T F F T F T T
F T F T T F T
F F F T T T T

(b) No, (p ⇒ q)∧(q ⇒ r) ⇒ (p ⇒ r), (p ⇒ q)∧(q ⇒ r) ⇍ (p ⇒ r).

(c) Yes, the truth table is:

p q r q∧r p ∨ (q∧r) (p ∨ q) (p ∨ r) (p ∨ q) ∧ (p ∨ r)
F F F F F F F F
F F T F F F T F
F T F F F T F F
F T T T T T T T
T F F T T T T T
T F T T T T T T
T T F T T T T T
T T T T T T T T

1
 (d) Yes, the truth table is:

A B C A⇒B ˜C A ⇒ B ⇒ ˜C ˜(A ⇒ B) C ⇒ ˜(A ⇒ B)

T T T T F F F F
T T F T T T F T
T F T F F T T T
T F F F T T T T
F T T T F F F F
F T F T T T F T
F F T T F F F F
F F F T T T F T

2. Prove by contradiction:
Assume that h is injective, i.e. ∀x1 , x2 s.t. x1 ̸= x2 ⇒ f (x1 ) ̸= f (x2 )
h(x) = 0 ⇒ x3 − x = 0 ⇒ x = 0, x = 1, x = −1
x1 = 0, x2 = 1, x1 ̸= x2 ⇒ f (x1 ) = f (x2 ) = 0 ⇒contradiction⇒ h(x) is non-injective.
3. • Surjectivity: ∀x ∈ R ex > 0, i.e. Rf = R+ \{0}
Rf ̸= Cf = R ⇒ f : R → R, f (x) = ex is not surjective.
• Injectivity: x1 ̸= x2 , let x1 > x2 , then f (x1 ) > f (x2 ), because f (x) = ex is monotonic. Hence
f : R → R, f (x) = ex is injective.
• Bijectivity: f : R → R, f (x) = ex is not surjective and thus is not bijective.

(a) Use proof by contradiction:

There exist two convex A, B ⊆ RN s.t. A ∪ B is not convex. Thus, the union of two convex sets
is not always convex.
(b) Proof by mathematical induction:
P1
1. n = 1 ⇒ k=1 (2 − 1)2 = 12 = 1∗1∗3
3
Pn n(2n−1)(2n+1)
2. Assume that it is true for k=1 (2k − 1)2 = 3
3. if n = k + 1 :

2
 n+1 n
!
X X
2 2
(2k − 1) = (2k − 1) + (2(n + 1) − 1)2
k=1 k=1
n(2n − 1)(2n + 1)
= + (2n + 1)2
3
(2n + 1)(n(2n − 1) + 3(2n + 1))
=
3
(2n + 1)(2n2 − n + 6n + 3)
=
3
(2n + 1)(2n2 + 5n + 3)
=
3
(2n + 1)(2n2 + 3n + 2n + 3)
=
3
(2n + 1)(2n + 3)(n + 1)
=
3
(n + 1)(2(n + 1) − 1)(2(n + 1) + 1)
=
3

Section II (15 points)

Consider the following matrix:  
1 2 3 4
 2 3 1 2 
 
 1 1 1 −1 
1 0 −2 −6
1. Compute the determinant, the cofactors, the trace, the kernel, and the adjoint and inverse matrices.
Do the exercise “by hand.”
2. Use a computer software like Matlab/Octave/Excel/etc. to do the computations. Also, perform
a spectral and Cholesky decomposition of the matrix using computer software. Please enclose the
code/spreadsheet/etc.

Solution
1. (a) Determinant

2 3 4 1 3 4 1 2 4 1 2 3
|A| = 1(−1)1+4 3 1 2 +0(−1)2+4 2 1 2 +(−2)(−1)3+4 2 3 2 +(−6)(−1)4+4 2 3 1
1 1 −1 1 1 −1 1 1 −1 1 1 1

= (−1)[2 ∗ 1 ∗ (−1) + 3 ∗ 2 ∗ 1 + 4 ∗ 3 ∗ 1 − 1 ∗ 1 ∗ 4 − 1 ∗ 2 ∗ 2 − (−1) ∗ 3 ∗ 3]

+ 2[1 ∗ 3 ∗ (−1) + 2 ∗ 2 ∗ 1 + 4 ∗ 2 ∗ 1 − 1 ∗ 3 ∗ 4 − 1 ∗ 2 ∗ 1 − (−1) ∗ 2 ∗ 2]
− 6[1 ∗ 3 ∗ 1 + 2 ∗ 1 ∗ 1 + 3 ∗ 2 ∗ 1 − 1 ∗ 3 ∗ 3 − 1 ∗ 1 ∗ 1 − 1 ∗ 2 ∗ 2] = −1

3
(b) Cofactors

3 1 2
c11 = (−1)1+1 1 1 −1
0 −2 −6
= [3 ∗ 1 ∗ (−6) + 1 ∗ (−1) ∗ 0 + 2 ∗ 1 ∗ (−2) − 0 ∗ 1 ∗ 2 − (−2) ∗ (−1) ∗ 3 − (−6) ∗ 1 ∗ 1]
= −22
2 1 2
c12 = (−1)1+2 1 1 −1
1 −2 −6
= (−1) ∗ [2 ∗ 1 ∗ (−6) + 1 ∗ (−1) ∗ 1 + 2 ∗ 1 ∗ (−2) − 1 ∗ 1 ∗ 2 − (−2) ∗ (−1) ∗ 2 − (−6) ∗ 1 ∗ 1]
= 17
2 3 2
c13 = (−1)1+3 1 1 −1
1 0 −6
= [2 ∗ 1 ∗ (−6) + 3 ∗ (−1) ∗ 1 + 2 ∗ 1 ∗ 0 − 1 ∗ 1 ∗ 2 − 0 ∗ (−1) ∗ 2 − (−6) ∗ 1 ∗ 3]
=1
2 3 1
1+4
c14 = (−1) 1 1 1
1 0 −2
= (−1)[2 ∗ 1 ∗ (−2) + 3 ∗ 1 ∗ 1 + 1 ∗ 1 ∗ 0 − 1 ∗ 1 ∗ 1 − 0 ∗ 1 ∗ 2 − (−2) ∗ 1 ∗ 3]
= −4
2 3 4
c21 = (−1)2+1 1 1 −1
0 −2 −6
= (−1)[2 ∗ 1 ∗ (−6) + 3 ∗ (−1) ∗ 0 + 4 ∗ 1 ∗ (−2) − 0 ∗ 1 ∗ 4 − (−2) ∗ (−1) ∗ 2 − (−6) ∗ 1 ∗ 3]
=6
1 3 4
c22 = (−1)2+2 1 1 −1
1 −2 −6
= [1 ∗ 1 ∗ (−6) + 3 ∗ (−1) ∗ 1 + 4 ∗ 1 ∗ (−2) − 1 ∗ 1 ∗ 4 − (−2) ∗ (−1) ∗ 1 − (−6) ∗ 1 ∗ 3]
= −5
1 2 4
c23 = (−1)2+3 1 1 −1
1 0 −6
= (−1)[1 ∗ 1 ∗ (−6) + 2 ∗ (−1) ∗ 1 + 4 ∗ 1 ∗ 0 − 1 ∗ 1 ∗ 4 − 0 ∗ (−1) ∗ 1 − (−6) ∗ 1 ∗ 2]
=0
1 2 3
c24 = (−1)2+4 1 1 1
1 0 −2
= [1 ∗ 1 ∗ (−2) + 2 ∗ 1 ∗ 1 + 3 ∗ 1 ∗ 0 − 1 ∗ 1 ∗ 3 − 0 ∗ 1 ∗ 1 − (−2) ∗ 1 ∗ 2]
=1

4
 2 3 4
c31 = (−1)3+1 3 1 2
0 −2 −6
= [2 ∗ 1 ∗ (−6) + 3 ∗ 2 ∗ 0 + 4 ∗ 3 ∗ (−2) − 0 ∗ 1 ∗ 4 − (−2) ∗ 2 ∗ 2 − (−6) ∗ 3 ∗ 3]
= 26
1 3 4
c32 = (−1)3+2 2 1 2
1 −2 −6
= (−1)[1 ∗ 1 ∗ (−6) + 3 ∗ 2 ∗ 1 + 4 ∗ 2 ∗ (−2) − 1 ∗ 1 ∗ 4 − (−2) ∗ 2 ∗ 1 − (−6) ∗ 2 ∗ 3]
= −20
1 2 4
c33 = (−1)3+3 2 3 2
1 0 −6
= [1 ∗ 3 ∗ (−6) + 2 ∗ 2 ∗ 1 + 4 ∗ 2 ∗ 0 − 1 ∗ 3 ∗ 4 − 0 ∗ 2 ∗ 1 − (−6) ∗ 2 ∗ 2]
= −2
1 2 3
c34 = (−1)3+4 2 3 1
1 0 −2
= (−1)[1 ∗ 3 ∗ (−2) + 2 ∗ 1 ∗ 1 + 3 ∗ 2 ∗ 0 − 1 ∗ 3 ∗ 3 − 0 ∗ 1 ∗ 1 − (−2) ∗ 2 ∗ 2]
=5
2 3 4
c41 = (−1)4+1 3 1 2
1 1 −1
= (−1)[2 ∗ 1 ∗ (−1) + 3 ∗ 2 ∗ 1 + 4 ∗ 3 ∗ 1 − 1 ∗ 1 ∗ 4 − 1 ∗ 2 ∗ 2 − (−1) ∗ 3 ∗ 3]
= −17
1 3 4
c42 = (−1)4+2 2 1 2
1 1 −1
= [1 ∗ 1 ∗ (−1) + 3 ∗ 2 ∗ 1 + 4 ∗ 2 ∗ 1 − 1 ∗ 1 ∗ 4 − 1 ∗ 2 ∗ 1 − (−1) ∗ 2 ∗ 3]
= 13
1 2 4
c43 = (−1)4+3 2 3 2
1 1 −1
= (−1)[1 ∗ 3 ∗ (−1) + 2 ∗ 2 ∗ 1 + 4 ∗ 2 ∗ 1 − 1 ∗ 3 ∗ 4 − 1 ∗ 2 ∗ 1 − (−1) ∗ 2 ∗ 2]
=1
1 2 3
c44 = (−1)4+4 2 3 1
1 1 1
= [1 ∗ 3 ∗ 1 + 2 ∗ 1 ∗ 1 + 3 ∗ 2 ∗ 1 − 1 ∗ 3 ∗ 3 − 1 ∗ 1 ∗ 1 − 1 ∗ 2 ∗ 2]
= −3

 
−22 17 1 −4
 6 −5 0 1
Cof (A) = 
 26 −20 −2

5
−17 13 1 −3
(c) Trace Trace(A) = 1 + 3 + 1 + (−6) = −1

5
 (d) Kernel Nullspace (or kernel) : null(A)= {x ∈ Rn |Ax = 0}
Try to perform row operations
 to
 get the row echelon
 form  : 
1 2 3 4 r3 =r3 −r4 0 2 5 10 0 2 5 10
r2 =r2 −r3
2 3 1 2 1 −r4 1
 2 0 3  0 −5 −7
2 −r1 1
A=  r1 =r→  r2 =r→ 
1 1 1 −1 0 1 3 5  0 1 3 5
1 0 −2 −6  1 0 −2 −6  1 0 −2 −6 
0 0 −1 0 0 0 −1 0 0 0 −1 0
r2 =r2 −r4
r1 =r1 −2r3  0 0 −3 −1  rr23=r
=r3 +r2
2 −3r 1
0 0 0 −1 r4 =r r3 =r3 +4r2
4 −2r 1 −6r 2
0 0 0 −1
→ 
0 1 3
 →   →  
5 0 1 0 4 0 1 0 0
1 0 −2 −6 1 0 −2 −6 1 0 0 0
    
0 0 −1 0 x1 0
0 0 0 −1 x2  0
Ax = 0 :  0 1 0
   =   and this leads to :
0  x3  0
1 0 0 0 x4 0

x1 = 0
x2 = 0
−x3 = 0
−x4 = 0

Implies :

x1 = 0
x2 = 0
x3 = 0
x4 = 0

Therefore Kernel(A) = {0} and there are no non-zero vector satisfying Ax = 0.

 
−22 6 26 −17
 17 −5 −20 13 
(e) Adjoint Adj(A) = [Cof (A)]′ =  
 1 0 −2 1 
−4 1 5 −3
   
−22 6 26 −17 22 −6 −26 17
1  17
 −5 −20 13   = −17 5
 20 −13
(f) Inverse A−1 = det(A)
1
Adj(A) = −1 
 1 0 −2 1   −1 0 2 −1 
−4 1 5 −3 4 −1 −5 3
2. Refer to the python code

Section III (30 points)

The dataset dataq2.csv contains daily returns for 11 assets. Use the dataset provided with the assignment
and answer the following questions:
1. Assume that the market is made up of the 11 assets. Is the market arbitrage free? For this purpose,
a market is said to be arbitrage free is there exists no linear dependence between the returns of the
assets. Explain.
2. Now drop the 11th asset and answer the following:
(a) What is the average daily return of an equally weighted portfolio?
(b) What is the variance-covariance matrix?

6
 3. Now consider a factor analysis of the returns. Factor analysis allows for dimension reduction by
retaining only the top eigenvalues (factors) that capture the most significant variation in the data.
This reduces the dimensionality of the problem while preserving the essential information.
(a) Decompose the variance-covariance into its eigenvalues and eigenvectors.
(b) Show that each eigenvector represents a linear combination of the original variables, and each
eigenvalue represents the variance explained by its corresponding eigenvector.
(c) Perform factor analysis using 1, 2, 3, 4, 5, and 6 factors without rotation. How many factors
are actually required to perform factor analysis? Give reasons. (Hint: Check eigenvalues. The
number of factors is the number of eigenvalues exceeding 1. This is known as the Kaiser Rule.
The logic is that a factor with eigenvalue greater than 1 explains more variance than a single
observed variable.)
(d) Factor rotation helps to make the factors more interpretable. Varimax rotation is an orthogo-
nal rotation in which assumption is that there is no intercorrelations between components. In
contrast, oblique rotation is more complex, and can provide more superior results. See https://real-
statistics.com/linear-algebra-matrix-topics/varimax/. Perform factor analysis using the required
number of factors with varimax rotation. Report the factor loadings. Interpret the factor loadings.

Solution
1. Refer to the Python code.
2. Now drop the 11th asset and answer the following:
(a) Refer to the Python code
(b) Refer to the Python code
3. Now consider a factor analysis of the returns. Factor analysis allows for dimension reduction by
retaining only the top eigenvalues (factors) that capture the most significant variation in the data.
This reduces the dimensionality of the problem while preserving the essential information.
(a) Refer to the Python code
(b) One of the ways to show this is regress each eigenvector on the original variables.
A mathematical explanation can be given as: We start from variance-covariance matrix, S

S = E XX T − E(X)E(X)T

Let µ be a column vector of the same dimension of X and µT µ = 1, then


T 
T
µT Sµ = µT E XX T − E(X)E(X)T µ = E µT X µT X − E µT X E µT X

Note that µT X is the projection of X onto µ, such that


T 
T
E µT X µT X − E µT X E µT X = Var µT X

Since Var µT X is simply a number, so we denote it as λ, so we have

µT Sµ = λ

Since µT µ = 1, so
µµT Sµ = Sµ = λµ
which means the µ that we define in the first place is actually an eigenvector of the data covariance
matrix, the eigenvalue of which is the variance that the data has in that direction.
(c) Refer to the Python code
(d) Refer to the Python code

7
Section IV - (15 points)
1. Is function f (x, y) = ln (ex + ey ) convex? Is it strictly convex?
2. Calculate
π
+ cos π2 ex



sin 2x
lim
cos 3π


2 cosx
x→0

Hint : use Taylor’s theorem and Taylor series representation.

R x dx
3. Evaluate: √xe1+e x

Solution:
1. We will show that f (x, y) is convex, but not strictly convex. Convexity follows from that H(x, y) is
positive semidefinite on x, y ∈ R2 (see p.101, proposition 3.19).

∂f (x, y) ex ∂f (x, y) ey ∂f 2 (x, y) ex e2x ex+y

= x y
, = x y
=⇒ 2 = x y
− 2 = 2
∂x e +e ∂y e +e (∂x) e +e (ex + ey ) (ex + ey )

∂f 2 (x, y) ey e2y ex+y ∂f 2 (x, y) ∂f 2 (x, y) ex+y

2 = − 2 = 2 , = = − 2.
(∂y) ex + ey (ex + ey ) (ex + ey ) ∂x∂y ∂y∂x (ex + ey )
 
∂f 2 (x,y) ∂f 2 (x,y) ex+y x+y !
− (exe+ey )2
 
(∂x) 2 ∂x∂y (e x +ey )2 ex+y 1 −1
Thus H (x, y) =  ∂f 2 (x,y) ∂f 2 (x,y)  = x+y x+y = (ex +ey )2
.
− (exe+ey )2 e −1 1
∂y∂x (∂y)2 (ex +ey )2

ex+y
∆1 = (ex +ey )2
> 0, ∆2 = 0 ≥ 0.
Thus quadratic form H (x, y) is positive semidefinite and f (x, y) is convex. Note that if H (x, y) is
positive semidefinite, function f (x, y) can still be strictly convex.
However, one can quickly obtain a counter-example in which strict convexity doesn’t hold for f (x, y).
Take (x1 , y1 ) = (0, 0) , (x2 , y2 ) = (2, 2) , λ = 21 . If f (x, y) is strictly convex then we should have
 
λ ln (ex1 + ey1 ) + (1 − λ) ln (ex2 + ey2 ) > ln eλx1 +(1−λ)x2 + eλy1 +(1−λ)y2 .
 1 1 1 1

1 1
= 1 + ln 2 , which is equal to ln e 2 ·0+ 2 ·2 + e 2 ·0+ 2 ·2 =



However, 2 ln e0 + e0 + 2 ln e2 + e2
ln (2e) = 1 + ln 2 . Contradiction. Therefore, f (x, y) is not strictly convex.

2. We are going to approximate the given functions by polynomials around a = 0 applying Taylor’s
theorem. One may thus note that

(a) sin π2 x = π2 x + E3


 h 2
i   2

(b) cos π2 ex = cos π2 1 + x + x2! + E3 = cos π2 + π2 x + π2 x2! + E3 = −sin π2 x + π4 x2 + E3 =

− π2 x + π4 x2 +

+ E3
 h i   
x2 3π x2
(c) cos 3π 3π 3π
= −sin 3π

2


2 cosx = cos 2 1 − 2! + E 4 = cos 2 − 2 2! + E 4 4 x + E4 =
− 3π 2
4 x + E4

8
 where En is a remainder in Peano’s form. Thus we have

π
+ cos π2 ex π
+ E3 − π2 x + π4 x2 + E3 − π4 x2 + E3 − π4 + E1





sin 2x 2x
lim = lim = lim = lim =
cos 3π − 3π x→0 − 3π x2 + E4 x→0 − 3π + E2


2
2 cosx 4 x + E4
x→0 x→0
4 4
− π4 + 0 1
= 3π = .
− 4 +0 3
√
2t
3. Let 1 + ex = t ⇒ x = ln t2 − 1 , dx = t2 −1 dt

xex dx ln(t2 −1)(t2 −1) 2t

= 2 ln t2 − 1 dt = 2 ln (t − 1) dt+2 ln (t + 1) dt = 2 [(t − 1) ln (t − 1) − t + (t + 1
R R R R R
√
1+ex
= t t2 −1 dt
√
√
√
√

2 (t − 1) ln (t − 1)+2 (t + 1) ln (t +1)−4t+C= 2 1 + ex − 1 ln 1 + ex − 1 +2 1+ ex + 1 ln 1 + ex + 1 −
√ √ √
1+ex +1 √ √ √
1+ex +1
4 1 + ex + C = 2 1 + ex x + 2 ln √1+e x −1 − 4 1 + ex + C = 2 1 + ex [x − 2] + 2 ln √1+e x −1 +C
 
u = ln (t − 1)
 du = dt 
 = t ln (t − 1)− t−1+1 dt+C = t ln (t − 1)−
R R
ln (t − 1) dt = {Integration by parts} =  t−1
 dw = dt  t−1

v=t
t − ln (t − 1) + C = (t − 1) ln (t − 1) − t + C

 
u = ln (t = 1)
R  du = dt  R t−1+1
ln (t + 1) dt = {Integration by parts} =  t+1  = t ln (t + 1)−
 dw = dt  t+1 dt+C = t ln (t + 1)−
v=t
t + ln (t + 1) + C = (t + 1) ln (t + 1) − t + C

Section V - (15 points)

Economy can be in three states: boom, recession and normal growth. There is 20% chance that when
economy is in boom next period it will move to normal growth regime and 5% chance that the period of
boom will lead to recession. If economy is in normal growth state there is 2% chance that it will boom next
period and 8% chance that it will fall into recession. For economy which is currently in recession there is
20% chance it will become normal next year and 10% chance that the economy will boom. Today economy
is in recession. What is the probability that in 10 years an economy will be in boom or recession? What
happens with these probabilities as t −→ ∞.
Hint: Use a matrix decomposition technique to show your computations.

Solution: Let’s put exercise in matrix notation , if Bt , Rt , Nt denote probabilities that economy is in
boom, recession or in normal growth today at t .

↗ boom(100 − 20 − 5 = 75%)
If today is boom - tomorrow −→ normal(20%)
↘ recession(5%)

↗ boom(2%)
If today is normal growth - tomorrow −→ normal(100 − 2 − 8 = 90%)
↘ recession(8%)

↗ boom(10%)
If today is recession- tomorrow −→ normal(20%)

9
 ↘ recession(100 − 10 − 20 = 70%)

So, probabilities Bt+1 , Rt+1 , Nt+1 that economy tomorrow will be in boom, recession or normal growth
can be computed as follow:
      
Bt+1 0.75 0.02 0.1 Bt Bt
 Nt+1  =  0.2 0.9 0.2   Nt  = A  Nt  .
Rt+1 0.05 0.08 0.7 Rt Rt
 
Bt+10
We are asked to find  Nt+10  .
Rt+10
       
Bt+10 Bt+9 Bt+8 Bt
 Nt+10  = A  Nt+9  = A · A  Nt+8  = A10  Nt  .
Rt+10 Rt+9 Rt+8 Rt

In order to find A10 we use a matrix decomposition technique, i.e.

A = CDC −1 ,

where C is a matrix of eigenvectors and C −1 is the inverse of C. D is a diagonal matrix with eigenvalues on
principal diagonal.
A2 = CDC −1 CDC −1 = CD2 C −1 =⇒ A10 = CD10 C −1 .
So,    
Bt+10 Bt
Nt+10  = CD10 C −1 Nt  .
Rt+10 Rt

Let’s find a decomposition of A:

0.75 − λ 0.02 0.1 7.5 − 10λ 0.2 1

1
|A − λI| = 0.2 0.9 − λ 0.2 = 2 9 − 10λ 2
10
0.05 0.08 0.7 − λ 0.5 0.8 7 − 10λ
1 
(7.5 − 10λ) 63 − 160λ + 100λ2 − 1.6 − 2 (1.4 − 2λ − 0.8) + 0.5 (0.4 − 9 + 10λ)


=
10
1
= 472.5 − 630λ − 1200λ + 1600λ2 + 750λ2 − 1000λ3 − 1 + 16λ
10
−2.8 + 4λ + 1.6 + 0.2 − 4.5 + 5λ)
1
−1000λ3 + 2350λ2 − 1805λ + 455 .


=
10

|A − λI| = 0 Let’s try λ = 1 : −1000 + 2350 − 1805 + 455 = 0

⇒ λ = 1 is an eigenvalue of matrix

1 
−1000λ3 + 2350λ2 − 1805λ + 455

|A − λI| =
10
1
= · (λ − 1)(−1000λ2 + 1350λ − 455) − 1000λ2 + 1350λ − 455 = 0
10
1
−1350 ± (13502 − 4 · 455 · 1000) 2 −1350 ± 50
λ1,2 = =
−2000 −2000
=⇒ λ1 = 0.7, λ2 = 0.65.

10
So, |A − λI| = (λ − 1) (λ − 0.7) (λ − 0.65) and the eigenvalues are λ ∈ {1, 0.7, 0.65}. The corresponding
eigenvector are given by:

For λ = 1:     
−0.25 0.02 0.1 x1 0
(A − λI)x =  0.2 −0.1 0.2  x2  = 0
0.05 0.08 −0.3 x3 0
A few elementary transformations yield
   
x1 0
− 23
 
1 0 x2  = 0
0 1 − 10
3 x3 0
2 10

′
Thus, we obtain for λ = 1, the eigenvector X = 3, 3 , 1 .

For λ = 0.65:     
0.1 0.02 0.1 x1 0
(A − λI)x =  0.2 0.25 0.2 x2  = 0
0.05 0.08 0.1 x3 0
Again, after a few elementary transformations we get
   
 x 0
1 0 1  1  

x2 = 0
0 1 0
x3 0
′
Thus, we obtain for λ = 0.65, the eigenvector X = (−1, 0, 1) .

Finally, for λ = 0.7:     

0.05 0.02 0.1 x1 0
(A − λI)x =  0.2 0.2 0.2 x2  = 0
0.05 0.08 0 x3 0
Again, after a few elementary transformations we get
   
 8
 x1 0
1 0 3 x2  = 0
0 1 − 53
x3 0

′
Thus, we obtain for λ = 0.65, the eigenvector X = − 38 , 5
3, 1 .

At the end we obtain the matrices D and C with

2
− 38
   
1 0 0 3 −1
10 5
D = 0 0.65 0 , C= 
3 0 3
.
0 0 0.7 1 1 1

Note: Eigenvectors are not unique. Here, I set the arbitrary value to 1.

In order to finish the exercise, we have to find C −1 which is given by

 
1 1 1
C −1 = 5−1  1 −2 6 
−2 1 −2

11
and we end up with the yield vector
    
Bt+10 0.1608 0.1237 0.1473 Bt
Nt+10  = 0.6478 0.6761 0.6478 Nt  .
Rt+10 0.1914 0.2003 0.2049 Rt

Moreover, to find (Bt+10 , Nt+10 , Rt+10 ) as t → ∞ we have:

       
B∞ Bt 1 0 0 Bt
N∞  = CD∞ C −1 Nt  = C 0 0 0 C −1 Nt 
R∞ Rt 0 0 0 Rt
 2 2 2
 
15 15 15 Bt
=  32 2
3
2 
3 Nt

0.2 0.2 0.2 Rt

This shows that the time series is stationary. In the exercise text it is given that we are in a recession, thus
′
the starting vector becomes (0, 0, 1) . Plugging in this vector in the above formulas results in the desired
probabilities.

12