University of Warwick

Department of Statistics

ST 339 Introduction to Mathematical Finance

Solutions for Problem Sheet 3

Solution 3-1

(a) We have to show that R⪰ is complete and transitive.

R First, we argue completeness. So let
ν1 , ν2 ∈ M. Set y1 := U (x) ν1 (dx) and y2 := U (x) ν2 (dx) ∈ R. Then either y1 ≤ y2 or
y1 ≥ y2 or y1 = y2 . Hence ⪰ is complete. Next we argue transitivity. So let ν1 , ν2 , ν3 ∈ M
such that ν1 ⪰ ν2 and ν2 ⪰ ν3 . Set
Z
yi := U (x) νi (dx), i ∈ {1, 2, 3}. (1)

Then y1 ≥ y2 and y2 ≥ y3 . This implies that y1 ≥ y3 , and so ν1 ⪰ ν3 .

(b) Let ν1 , ν2 ∈ M be such that ν1 ≻ ν2 . Moreover, let ν3 ∈ M and α ∈ (0, 1). Set

Z
yi := U (x) νi (dx), i ∈ {1, 2, 3}. (2)

Then by linearity of the integral in the integrator,

Z Z Z
U (x) (αν1 + (1 − α)ν3 )(dx) = α U (x) ν1 (dx) + (1 − α) U (x) ν3 (dx)

= αy1 + (1 − α)y3 , (3)

Z Z Z
U (x) (αν2 + (1 − α)ν3 )(dx) = α U (x) ν2 (dx) + (1 − α) U (x) ν3 (dx)

= αy2 + (1 − α)y3 , (4)

By hypothesis, y1 > y2 . Since α > 0, this implies that

αy1 + (1 − α)y3 > αy2 + (1 − α)y3 , (5)

and so we may conclude that αν1 + (1 − α)ν3 ≻ αν2 + (1 − α)ν3 .

(c) Let ν1 , ν2 , ν3 ∈ M be such that ν1 ≻ ν2 ≻ ν3 . Set
Z
yi := U (x) νi (dx), i ∈ {1, 2, 3}. (6)

Then
y1 > y2 > y3 . (7)

Set
y2 − y3
α := ∈ (0, 1). (8)
y1 − y3
Then
y1 y2 − y1 y3 y3 y1 − y3 y2
αy1 + (1 − α)y3 = + = y2 . (9)
y3 − y1 y3 − y1
By (3), it follows that
ν2 ∼ αν1 + (1 − α)ν3 .
Solution 3-2

(a) Let ν1 , ν2 ∈ M and α ∈ [0, 1]. By linearity of the integral in the integrator, we obtain
Z Z Z
µαν1 +(1−α)ν2 = x (αν1 + (1 − α)ν2 )(dx) = α x ν1 (dx) + (1 − α) x ν2 (dx)

= αµν1 + (1 − α)µν2 . (10)

Moreover, for ν∈M by linearity of the integral in the integrand and the fact that ν is a
probability measure,
Z Z Z
σν2 = 2
(x − µν ) ν(dx) = 2
x ν(dx) − 2µν x ν(dx) + (µν )2
Z
= x2 ν(dx) − (µν )2 . (11)

Combining (11) with (10) and using linearity of the integral in the integrator, we obtain

Z
2
σαν 1 +(1−α)ν2
= x2 (αν1 + (1 − α)ν2 )(dx) − (µαν1 +(1−α)ν2 )2 (12)
Z Z
= α x2 ν1 (dx) + (1 − α) x2 ν1 (dx)

− (α2 µ2ν1 + 2α(1 − α)µν1 µν2 + (1 − α)2 µ2ν2 ) (13)

= ασν21 + αµ2ν1 + (1 − α)σν22 + (1 − α)µ2ν2

− (α2 µ2ν1 + 2α(1 − α)µν1 µν2 + (1 − α)2 µ2ν2 ) (14)

= ασν21 + (1 − α)σν22
+ (α − α2 )µ2ν1 − 2α(1 − α)µν1 µν2 + ((1 − α) − (1 − α)2 )µ2ν2 (15)

= ασν21 + (1 − α)σν22 + α(1 − α)(µν1 − µν2 )2 . (16)

(b) Let ν1 , ν2 , ν3 ∈ M be such that ν1 ≻ ν2 ≻ ν3 . Dene the function f : [0, 1] → R by

2
f (α) = µαν1 +(1−α)ν3 − σαν 1 +(1−α)ν3

= αµν1 + (1 − α)µν3 − ασν21 − (1 − α)σν23 − α(1 − α)(µν1 − µν3 )2 . (17)

Then f is continuous and satises

f (0) = µν3 − σν23 and f (1) = µν1 − σν21 . (18)

By hypothesis,
f (1) > µν2 − σν22 > f (0). (19)

As f is continuous, by the intermediate value theorem, there is α ∈ (0, 1) such that

f (α) = µν2 − σν22 . (20)

This implies that

ν2 ∼ αν1 + (1 − α)ν3 .
As M0 is a convex subset of M, the continuity axiom is a fortiori satised on M0 .
 (c) Let ν1 = δ4 and ν2 = δ0 . Then

µν1 − σν21 = 4 − 0 = 4 and µν2 − σν22 = 0 − 0 = 0. (21)

Thus, ν1 ≻ ν2 . Now let ν3 = ν2 . If the independence axiom holds, then

1 1 1 1
ν1 + ν2 ≻ ν2 + ν2 = ν2 (22)
2 2 2 2
However by part (a),
 
1 1 1 1 1
µ 1 ν1 + 1 ν2 − σ 21 ν +1ν
= ×4+ ×0− × 0 + × 0 + (4 − 0)2 =2−4
2 2 2 1 2 2 2 2 2 2 4
= −2 < 0 = µν2 − σν22 . (23)

1
Thus,
2 ν1 + 21 ν2 ≺ ν2 , and the independence axiom fails on M.
(d) Let ν1 , ν2 ∈ M0 be such that ν1 ≻ ν2 . As µν1 = µν2 = 0, this means that

−σν21 > −σν22 (24)

Let ν3 ∈ M0 and α ∈ (0, 1). It follows from part (a) and (24) that

2
−σαν 1 +(1−α)ν3
= −ασν21 − (1 − α)σν23 > −ασν22 − (1 − α)σν23 = −σαν
2
2 +(1−α)ν3
(25)

As µνi = 0 for i ∈ {1, 2, 3}, it follows that

αν1 + (1 − α)ν3 ≻ αν2 + (1 − α)ν3 ,

and thus ⪰ satises the independence axiom on M0 .
As µν = 0 for ν ∈ M0 , it follows that
Z
µν − σν2 = −σν2 = −x2 ν(dx), ν ∈ M0 .

Thus ν1 ⪰ ν2 for ν1 , ν2 in M0 if and only if

Z Z
2
−x ν1 (dx) ≥ −x2 ν2 (dx).

So the desired function is U (x) = −x2 . Note that U is continuous and strictly concave
but not (strictly) increasing. For this reason it fails to be a utility function in the sense of
Denition 3.16 in the lecture notes.

Solution 3-3

(a) Since U2 is strictly increasing, it has an inverse function U2−1 : U2 (D) → D. As U2 is twice
continuously dierentiable, so is U2−1 , and U2 (D) is open because D is open. Dene the
function G : U2 (D) → R by
G(x) = U1 (U2−1 (x)).
Then G is also twice continuously dierentiable, and we have with y = U2−1 (x)
d U ′ (y)
G′ (x) = U1 (y) = 1′ > 0, (26)
dx U2 (y)
d U1′ (y)
 ′′
U1 (y) U2′′ (y)U1′ (y)

′′ 1
G (x) = ′ = ′ − ′ 2 ′ (27)
dx U2 (y) U (y) (U2 (y)) U2 (y)
′
 ′′ 2 ′′

U1 (y) U1 (y) U2 (y)
= − ′
(U2′ (y))2 U1′ (y) U2 (y)
′
U1 (y)
= (−A1 (y) + A2 (y)) ≤ 0. (28)
(U2′ (y))2
Thus, G is strictly increasing and concave.
 (b) The denition of the certainty inequality, part (a), Lemma 3.1 in the lecture notes, and
Jensen's inequality give
Z Z Z
U1 (cU1 (ν)) = U1 (x) ν(dx) = G(U2 (x)) ν(dx) = G(y) ν U2 (dy) (29)
Z  Z 
U2
≤G y ν (dy) = G U2 (x) ν(dx) (30)

= G(U2 (cU2 (ν))) = U1 (cU2 (ν)). (31)

As U1 is increasing, this implies cU2 (ν) ≥ cU1 (ν), whence

ρ1 (ν1 ) = µν − cU1 (ν) ≥ µν − cU2 (ν) = ρ2 (ν). (32)

Solution 3-4

(a) Lottery ν1 is rst order stochastic dominant over lottery ν2 if and only if any agent who
has a preference order that admits a von-Neumann Morgenstern representation and is
1
monotone , weakly prefers ν1 over ν2 , independent of their particular preferences.

Lottery ν1 is second order stochastic dominant over lottery ν2 if and only if any agent who
has a preference order that admits an expected utility representation , weakly prefers
2 ν1
over ν2 , independent of their particular preferences.

Neither rst nor second order stochastic dominance is a preference order of M in the
sense on Denition 3.3 in the lecture notes because they fail to be complete. Indeed, let
ν1 = 51 δ−1 + 54 δ0 and ν2 = 15 δ−2 + 45 δ2 , and U1 (x) = −2−x and U2 (x) = −4−x . Then both
U1 and U2 are utility functions (and in particular strictly increasing). On the one hand,
Z
1 4 6
U1 (x) ν1 ( dx) = × (−2) + × (−1) = − , (33)
5 5 5
Z
1 4 1
U1 (x) ν2 ( dx) = × (−4) + × (− ) = −1, (34)
5 5 4
and thus Z Z
U1 (x) ν1 ( dx) < U1 (x) ν2 ( dx). (35)

On the other hand

Z
1 4 8
U2 (x) ν1 ( dx) =× (−4) + × (−1) = − , (36)
5 5 5
Z
1 4 1 65
U2 (x) ν2 ( dx) = × (−16) + × (− ) = − , (37)
5 5 16 20
and thus, Z Z
U2 (x) ν1 ( dx) > U2 (x) ν2 ( dx). (38)

So ν1 is neither rst nor second order stochastic dominant over ν2 , and also ν2 is neither
rst nor second order stochastic dominant over ν1 .
(b) Since each utility function is in particular strictly increasing, it follows that rst order
implies second order stochastic dominance.

1
andhas both ν1 and ν2 in the set of lotteries that can be compared by their preference order
2
and has both ν1 and ν2 in the set of lotteries that can be compared by their preference order
 To see that the converse is false, let ν1 be a standard normal distribution and ν2 = δµν1 = δ0 .
It follows from Lemma 3.13(b) in the lecture notes that

Z Z Z
U (x) ν2 (dx) = U (x) δµν1 (dx) > U (x) ν1 (dx) (39)

for all utility functions U : R → R for which U is integrable with respect to ν1 (and
ν2 ). So ν2 is second order stochastic dominant over ν1 . However, ν1 is not rst order
stochastic dominant over ν1 . Indeed, consider the function U (x) = exp(x), which is s
strictly increasing. Using the properties of the moment generating function of a normal
distribution, we obtain

Z Z ∞  2  
1 x 1
U (x) ν1 ( dx) = exp(x) √ exp − dx = exp (40)
−∞ 2π 2 2
Z Z
≥ 1 = exp(x) δ0 (dx) = U (x) ν2 (dx). (41)

Thus ν2 is not rst order stochastic dominant over ν1 .

(c) Let ν1 be normally distributed with mean µ1 and variance σ12 and ν2 be normally distributed
2
with mean µ2 and variance σ2 and assume that ν1 is second-order stochastic dominant over
ν2 . For γ > 0, set Uγ (x) = − exp(−γx). Then using the formula for the moment generating
function of normal random variable, we obtain

∞
(x − µi )2
Z Z  
1
Uγ (x)νi (dx) = − exp(−γx) q exp − dx (42)
−∞ 2πσi2 2σi2
2
 
2 σi
= − exp −γµi + γ , i ∈ {1, 2}. (43)
2

As ν1 is second-order stochastic dominant over ν2 , it follows that

2 2
   
2 σ1 2 σ2
− exp −γµ1 + γ ≥ − exp −γµ2 + γ for all γ ∈ (0, ∞) (44)
2 2

which by monotonicity of the exponential function is equivalent to

σ12 σ2
−γµ1 + γ 2 ≤ −γµ2 + γ 2 2 for all γ ∈ (0, ∞). (45)
2 2
Dividing (45) by −γ and letting γ go to 0, it follows that

µ1 ≥ µ2 . (46)

γ2
Dividing (45) by
2 and letting γ go to ∞, it follows that

σ12 ≤ σ22 . (47)

(d) First, we calculate

1
x2 1
Z Z
x 1 1
µν1 = x ν1 (dx) = dx = = − = 0, (48)
−1 2 4 −1 4 4
x3 1
Z Z 1 2  
x 1 1 1
σν21 2
= (x − µν1 ) ν1 (dx) dx = = − − = , (49)
−1 2 6 −1 6 6 3
 and
Z  
8 3 1 1
µν2 = x ν2 (dx) = × − + ×3=− (50)
9 4 9 3
2
1 2 25
Z   
8 3 1 1
σν22 2
= (x − µν2 ) ν2 (dx) = × − + + × 3+ = . (51)
9 4 3 9 3 18

Thus, µν1 > µν2 and σν21 < σν22 .

Next, let Uγ be as in (c). Then

Z Z 1 1
1 1
− exp(−γx) dx =
Uγ (x) ν1 (dx) = exp(−γx)
−1 2 2γ −1
1
= (exp(−γ) − exp(γ)), (52)
2γ
Z  
8 3 1
Uγ (x) ν2 (dx) = − exp γ − exp(−3γ) (53)
9 4 9
R R
R proceed to show that Uγ (x) ν1 (dx) < Uγ (x) ν2 (dx) for γ suciently
We large. As
Uγ (x) ν2 (dx) < 0, this is equivalent to showing that
R
U (x) ν1 (dx)
R γ >1 (54)
Uγ (x) ν2 (dx)
for γ suciently large. Letting γ → ∞, we obtain
1
2γ (exp(γ) − exp(−γ))
R
Uγ (x) ν1 (dx)
lim R = lim
Uγ (x) ν2 (dx) γ→∞ 89 exp 43 γ + 19 exp(−3γ)


γ→∞
1 1 7




2γ exp 4 γ − exp − 4 γ
= lim 8 1 15

= ∞, (55)
9 + 9 exp − 4 γ
γ→∞

and so (54) is satised for all γ suciently large. Thus, ν1 is not second order stochastic
dominant over ν2 .

Solution 3-5

(a) By Lemma 2.3. in the lecture notes, there is a one-to-one correspondence correspondence
between x0 -feasible portfolios ϑ ∈ R1+d parametrised in numbers of shares and portfolios
π∈ H 1+d−1 parametrised in fractions of wealth. Moreover, if ϑ ∈ R
1+d is an x -feasible
0
portfolio and π is the corresponding fraction of wealth, then

ϑ · S1 − ϑ · S0
= π · R. (56)
ϑ · S0
Rearranging and using that ϑ · S 0 = x0 gives

ϑ · S 1 = x0 π · R + x0 . (57)

Thus,



U ϑ · S 1 = − exp − γϑ · S 1 = − exp(−γx0 ) exp(−γx0 π · R
Hence, 
 

E U ϑ · S 1 = exp(−γx0 )E − exp(−γx0 π · R .
As exp(−γx0 ) is independent of ϑ or π, it follows that solving (∗) is equivalent to solving
solving (∗∗).
 (b) Let π ∈ H 1+d−1 . Since R is multivariate normal distributed with mean vector µ and
covariance matrix Σ, it follows from π·R is normal distributed with mean µπ and σπ2 . It
follows from Proposition 2.9 and Lemma 2.3 in the lecture notes that

µπ = (µ − r1) · π + r and σπ2 = π ⊤ Σπ. (58)

By the formula for the moment generating function of a normal distribution, this implies
that
2
 
2 2 σπ


E − exp(−γx0 π · R = − exp −γx0 µπ + γ x0
2
 
1 2 2 ⊤
= − exp −γx0 ((µ − r1) · π + r) + γ x0 π Σπ
2
 
 1 ⊤
= − exp(−γx0 r) exp −γx0 (µ − r1) · π − γx0 π Σπ (59)
2
Now using that exp(−γx0 r) does not depend on π, that the function y 7→ − exp(−γx0 y)
is increasing, and that there is a one-to-one correspondence between H 1+d−1 and Rd via
the maps π 7→ π and π 7→ (1 − π · 1, π), it follows that solving (∗∗) is equivalent to solving
solving (∗∗∗).

(c) The map F : Rd → R, π 7→ (µ − r1) · π − 21 γx0 π ⊤ Σπ is twice continuously dierentiable

with

∇F (π) = (µ − r1) − γx0 Σπ (60)

HF (π) = −γx0 Σ (61)

Since the Hessian matrix HF (π) is negative denite (because Σ is positive denite) and
π= 1 −1 − r1),
the gradient has a unique zero at
γx0 Σ (µ it follows that

1 −1
π∗ := Σ (µ − r1) (62)
γx0
is the unique maximiser of (∗∗∗), and thus
 
B − rA 1 −1
π ∗ := (1 − 1 · π∗ , π∗ ) = 1− , Σ (µ − r1) (63)
γx0 γx0
is the unique maximiser of (∗∗), where A = 1⊤ Σ−1 1 and B = 1⊤ Σ−1 µ. It is ecient in
the mean-variance sense by Corollary 2.12 in the lecture notes.

(d) By (c) and (a) it follows that the unique maximiser ϑ∗ to (∗∗) satises
(
i x0 π∗i x0 − B−rA
γ for i = 0,
ϑ∗ = = 1 PD −1 (64)
j=1 Σij (µj − r), i ∈ {1, . . . , d}.
S0i for
γS i 0

The number of shares invested in the risky assets is independent of the initial wealth x0 .
Only the number of shares in the bank account depends on the initial wealth x0 . The
number of shares invested in the risky assets are inverse proportional to the parameter γ
of absolute risk aversion, i.e., the more risk averse the investor is, the less she invests into
the risky assets.

Solution 3-6

u−r d−r
Set u
e := 1+r and de := 1+r . Then
(
1+u
−1=u if ω = ω1 ,
X11 (ω) − X01 (ω) 1+r e
= 1+d (65)
1+r − 1 = de if ω = ω2 .
(a) For ϑ1 ∈ R, using (65), we have

(
x + ϑe
u if ω = ω1 ,
x + ϑ1 (X11 (ω) − X01 (ω)) = (66)
x + ϑde if ω = ω2 .

So ϑ1 ∈ A(x) if and only if

x + ϑ1 u
e≥0 and x + ϑ1 de ≥ 0. (67)

If r ≥ u, then e≤0
de < u and so by (67),

 xi
A(x) := − ∞, − . (68)
de

If d < r < u, then de < 0 < u

e and and so by (67),

h x xi
A(x) := − , − . (69)
u
e de

If r ≤ d, then 0 ≤ de < u
e and and so by (67),

h x 
A(x) := − , ∞ . (70)
u
e
(b) Dene the function F : A(x) → R by

F (ϑ1 ) = E U (x + ϑ1 (X11 − X01 )) = p1 U (x + ϑ1 u

e) + p2 U (x + ϑ1 d)
 
e
= p1 log(1 + x + ϑ1 u
e) + p2 log(1 + x + ϑ1 d)
e (71)

If r ≥ u, then e≤0
de < u and A(x) := (−∞, − xe]. So letting ϑ1 go to −∞ gives
d
 
lim F (ϑ1 ) = lim p1 log(1 + x + ϑ1 u
e) + p2 log(1 + x + ϑ1 d)
e
ϑ1 →−∞ ϑ1 →−∞

≤ lim p2 log(1 + x + ϑ1 d)
e = +∞, (72)
ϑ1 →−∞

and so u(x) = +∞.

If r ≤ d, then 0 ≤ de < u
e A(x) := [− uxe , ∞). So letting ϑ1 go to ∞ gives
and

 
lim F (ϑ1 ) = lim p1 log(1 + x + ϑ1 u e) + p2 log(1 + x + ϑ1 d)
e
ϑ1 →∞ ϑ1 →∞
≤ lim p1 log(1 + x + ϑ1 u
e) = +∞, (73)
ϑ1 →∞

and so u(x) = +∞.

(c) Dierentiating the function (71) dened in part (b) gives, for ϑ in the interior of A(x),

p1 u p2 de
F ′ (ϑ1 ) =
e
+
1 + x + ϑ1 ue 1 + x + ϑ1 de
(p1 u
e + p2 d)(1
e + x) + u e 1
edϑ
= , (74)
(1 + x + ϑ1 u
e)(1 + x + ϑ1 d)
e

where we have used that p1 + p2 = 1. As the denominator of (74) is positive for ϑ1 ∈ A(x),
the sign of the derivative is determined by the linear function in the numerator of (74).

Since u
ede < 0, this linear function is decreasing. Considered as a function on R, ϑ1 7→
(p1 u
e + p2 d)(1
e + x) + u e 1 has a zero at
edϑ

(p1 u
e + p2 d)(1
e + x) (1 + x)  
ϑ1# := − =− u − d)
p1 (e e + de , (75)
de
eu de
eu

where we have used that p1 + p2 = 1. Now, there are three cases. First if ϑ1# ≥ − xe, or
d
equivalently,
u 1 u−r 1
p1 ≥ 1 − =1−
e
, (76)
e − de (1 + x
u u−d1+x
then F′ is nonnegative on A(x) and F is increasing on A(x), whence

x
ϑ1∗ = − . (77)
u
e
Next, if ϑ1# ≤ − uxe , or equivalently

de 1 r−d 1
p1 ≤ − = , (78)
e − d (1 + x
u e u−d1+x

then F′ is nonpositive on A(x) and F is decreasing on A(x), whence

x
ϑ1∗ = − . (79)
u
e
 Finally if − uxe < ϑ1# < − xe, or equivalently
d
 
r−d 1 u−r 1
p1 ∈ ,1 − , (80)
u−d1+x u−d1+x

then F′ is nonnegative for ϑ1 ≤ ϑ1# and nonpositive for ϑ1 ≥ ϑ1# so that

(p1 u
e + p2 d)(1
e + x)
ϑ1∗ = ϑ1# = − (81)
de
eu

(d) Let Q≈P be the unique EMM for X 1. It follows from Example 1.16 in the lecture notes
that
r−d u−r
q1 := Q[{ω1 }] = and q2 := Q[{ω2 }] = .
u−d u−d
Moreover, by Example 3.22 in the lecture notes,

(
q1
dQ Q[{ω}] p1 if ω = ω1 ,
(ω) = = q2 (82)
dP P[{ω}] p2 if ω = ω2 .

Let ϑ∗ be as in (81). Then using that p1 + p2 = 1, we obtain

 
1 e − de
u p1
1+x+ ϑ1∗ u
= (1 + x) 1 −
e p1 (eu − d) + d
e e = (1 + x)p1 = (1 + x) , (83)
d −de q1
 
1   e−d
u e p2
1 + x + ϑ1∗ de = (1 + x) 1 − p2 (de − u
e) + u
e = (1 + x)p2 = (1 + x) . (84)
u u
e q2

Thus, using that U ′ (x) = 1

1+x , we obtain

1 1 q1
(
1+x+ϑ1∗ u
= 1+x p1 if ω = ω1 ,
U ′ x + ϑ1∗ (X 1 (ω) − X01 (ω)) =


(85)
e
1 1 q2
= 1+x p2 if ω = ω2 ,
1+x+ϑ1 de ∗
dQ
= (ω). (86)
dP