Econ 201A

Fall 2013
Problem Set 3 Suggested Solutions
1/ρ
1. MWG Exercise 3.D.5 We are considering a CES utility function u(x) = [xρ1 + xρ2 ] ,
ρ ∈ (−∞, 1).
(a) Let us first consider the case where ρ 6= 0. Since any strictly increasing transformation
of the utility function represents the same preference relation, we may as well solve the
utility maximization problem for
ũ(x) = u(x)ρ = xρ1 + xρ2 .
The utility maximization problem is given by
max xρ1 + xρ2 subject to x · p ≤ w.
x∈R+ 2

Moreover, we know that the constraint is binding as Walras’s Law holds because the
utility function is continuous and locally nonsatiated. This gives us the first order
conditions
ρxρ−1
1 = λp1
ρxρ−1
2 = λp2
x1 p1 + x2 p2 = w.
Solving for x1 , x2 yields the Walrasian Demand
 
w δ−1 δ−1

x(p, w) = p 1 , p2
pδ1 + pδ2
with δ = ρ/(ρ − 1) ∈ (−∞, 1).

Hence, the indirect utility is given by plugging this into u:

w
v(p, w) =
1/δ .
pδ1 + pδ2
1 1
If ρ = 0 the CES utility function is the Cobb-Douglas utility function u(x1 , x2 ) = x12 x22 ,
as
  
1/ρ 1/ρ
lim [xρ1 + xρ2 ] = lim exp ln [xρ1 + xρ2 ]
ρ→0 ρ→0
ln (xρ1 + xρ2 )
 
= lim exp
ρ→0 ρ
 xρ ln(x )+xρ ln(x ) 
1 1 2 2
xρ1 +xρ2
= lim exp  
ρ→0 1
1 1
= x12 x22 ,
 
where the third equality follows from l’Hôpital’s Rule. In that case, x(p, w) = 2pw1 , 2pw2
 1  1
2 2
and v(p, w) = 2pw1 w
2p2 . In the following we will restrict attention to the case
ρ 6= 0.

Page 1 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

(b) Homogeneity of degree zero of the demand function: For all λ > 0 it holds
 
λw δ−1 δ−1

x(λp, λw) = (λp 1 ) , (λp 2 )
(λp1 )δ + (λp2 )δ
 
w δ−1 δ−1

= p 1 , p2 = x(p, w)
pδ1 + pδ2
 
w
pδ1 + pδ2 = w.


Walras’s law: p · x(p, w) = pδ +p δ
1 2
Uniqueness: obvious.

Homogeneity of degree zero of the indirect utility function: For all λ > 0 it holds

λw
v(λp, λw) = 1/δ
((λp1 )δ + (λp2 )δ )
w
=
1/δ = v(p, w).
pδ1 + pδ2

Strict monotonicity of the indirect utility function in w and monotonicity in pi :

∂v(p, w) 1
=
1/δ > 0
∂w pδ1 + pδ2
∂v(p, w) wpδ−1
i
=−
1/δ+1 < 0.
∂pi p1 + pδ2
δ

Quasiconvexity of the indirect utility function: We need to show that {(p, w) : v(p, w) ≤
v} is convex for any v. Let us therefore assume that v(p0 , w0 ), v(p1 , w1 ) ≤ v, i.e.
 1
δ
wi ≤ v (pi1 )δ + (pi2 )δ

for i ∈ {0, 1}. Hence, for all λ ∈ [0, 1] it holds that

  1  1 
0 1 0 δ 0 δ δ 0 δ 0 δ δ
λw + (1 − λ)w ≤ v λ (p1 ) + (p2 ) + (1 − λ) (p1 ) + (p2 ) .
| {z }
≡wλ

Case 1: δ ∈ (0, 1)
1 1 1
In this case, x δ + y δ ≤ (x + y) δ holds, so
 1  1 
λ 0 δ 0 δ δ 1 δ 1 δ δ
w ≤ v λ(p1 ) + λ(p2 ) + (1 − λ)(p1 ) + (1 − λ)(p2 )
h i1
δ
≤ v λ(p01 )δ + λ(p02 )δ + (1 − λ)(p11 )δ + (1 − λ)(p12 )δ .

Moreover, x 7→ xδ is concave, so
h i1
δ
wλ ≤ v (λp01 + (1 − λ)p11 )δ + (λp02 + (1 − λ)p12 )δ .

Page 2 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

Case 2: δ < 0
sdf

Continuity of the indirect utility function: Obvious.

(c) For linear utility u(x1 , x2 ) = x1 + x2 , we get the Walrasian demand
  
w

 p
  1 , 0 if p1 < p2
x(p, w) = 0, pw2 if p1 > p2


{(x1 , x2 ) : p1 x1 + p2 x2 = w} if p1 = p2

n o
and the indirect utility function v(p, w) = max pw1 , pw2 .

For Leontief utility u(x1 , x2 ) = min(x1 , x2 ), we get the Walrasian demand

 
w w
x(p, w) = ,
p1 + p2 p1 + p2
w
and the indirect utitlity v(p, w) = p1 +p2 .

ρ → 1: Then, δ → −∞.

Case 1: p1 < p2
In this case,
wpδ−1
1
w
p w
lim = limδ→−∞  1 δ = and
δ→−∞ pδ1 + pδ2 p
1+ p2 p1
1

wpδ−1 w
lim δ 2 δ = limδ→−∞ p
 2δ
p1
= 0,
δ→−∞ p1 + p2 +1
p2

so the Walrasian demand of the CES utility converges to the Walrasian demand of the
linear utility function. The indirect utitlity for CES utitlity also converges to the indirect
utility for linear utitlity:
w
w p1 w
lim = lim   δ  = p .
pδ1 + p2δ 1/δ
δ→−∞

δ→−∞ p2 1
1+ p1

Case 2: p1 > p2
Analogously to Case 1.

wpδ−1
w
Case 3: p1 = p2 = p In this case, pδ +p
1
δ = 2p . This is an element of the Walrasian
1 2
demand of the CES Walrasian demand. The indirect utility functions also coincide:
w w
1/δ = p .
(pδ +pδ )
ρ → −∞: Then, δ → 1. By plugging in δ = 1 into the CES Walrasian demand and
indirect utility function, we get the corresponding functions for the Leontief utility.

Page 3 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

(d) Note that for the CES utility function the following holds:
 δ−1
x1 (p, w) p1
=
x2 (p, w) p2
 δ−2
∂ x1 (p, w) p1
p1 = (δ − 1) .
∂ p2 x2 (p, w) p2

Hence, the elasticity of substitution between goods 1 and 2 is constant and given by
1
ξ1,2 (p, w) = −(δ − 1) = .
1−ρ
Hence, for linear utility function ξ12 (p, w) = ∞, for Leontief utility function ξ12 (p, w) =
0, and for Cobb-Douglas utility function ξ12 (p, w) = 1.

2. Suppose the consumer can consume n goods, and the first good is electricity. The government
can impose a usage tax t on every unit of electricity consumed beyond a hurdle of H kilowatt-
hours, so the marginal price of each of the first H kWh is p and the marginal price of each
additional kWh above H is p + t.

(a) Describe the consumer’s budget set Bp,w,t,H for fixed p ∈ Rn++ , wage w ∈ R+ , tax t, and
hurdle H.

Solution: The nonlinear price scheme causes a “kink” in the consumer’s budget sets as
follows (
p·x≤w if x1 ≤ H
Bp,w,t,H =
p · x + t(x1 − H) ≤ w if x1 > H.
Assuming that taxes are nonnegative (which is what we do in subsequent questions),
we can equivalently write this as the
 intersection of two standard linear budget sets as
Bp,w,t,H = x ∈ Rn++ | p · x ≤ w ∩ x ∈ Rn++ | p · x + tx1 ≤ w + tH .



(b) Suppose the consumer’s utility function u : Rn+ → R is continuous. Prove the following:
The Walrasian demand correspondence, which now depends on the price vector, wage,
tax and hurdle, x∗ : Rn++ × R+ × R+ × R+ ⇒ Rn+ , defined by

x∗ (p, w, t, H) = arg max u(x)

x∈Bp,w,t,H

is upper hemicontinuous, and the indirect utility function v : Rn++ ×R+ ×R+ ×R+ ×R+ ⇒
R, defined by
v(p, w, t, H) = max u(x)
x∈Bp,w,t,H

is continuous.

Solution: This is the perfect setup to apply Berge’s theorem. Since we are given a
continuous utility function, all that remains to be shown is that the budget set corre-
spondence is continuous, and the two desired results will follow immediately. However,
this is faster set than done.

Page 4 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

Upper Hemicontinuity. To establish upper hemicontinuity of Bp,w,t,H , we observe

first that the correspondence is compact valued and so we can apply the sequence defini-
tion. For that, we consider an arbitrary domain-sequence {(pm , wm , tm , H m )} converg-
ing to (p, w, t, H), and an arbitrary range-sequence {xm } such that xm ∈ Bpm ,wm ,tm ,H m .
We need to show that there exists a convergent range subsequence xmk → x whose
limit x lies in the limit budget set Bp,w,t,H . In the same spirit as in the lecture,
we will use the Bolzano-Weierstrass Theorem; we just need to bound the sequence
of budget sets by the appropriate value. In part (a) we have shown however that
Bp,w,t,H ⊂ Bp,w = x ∈ Rn++ | p · x ≤ w , and so (inspired by the proof of Proposition
5.18), we define pm = mini pm i . We now have that

n
X
pm xm m m m
i ≤p ·x ≤w .
i=1

Since pm → p and wm → w, it follows that wm /pm → w/p. In other words, for any
 > 0, there exists N ∈ N such that m > N implies that
n
X wm w
xm
i ≤ m
≤ + .
p p
i=1

This imposes a bound on the sequence from N onwards, {xm }∞ m=N , and the Bolzano-
Weierstrass Theorem guarantees the existence of a convergent subsequence {xmk }∞ k=1 .
We now reached the first milestone in the proof of upper hemicontinuity: The existence
of a convergent range subsequence. It remains to show that the limit x = limk→∞ xmk
lies in Bp,w,t,H . However, since weak inequalities are preserved under limits, we conclude
from

pmk · xmk ≤ wmk and pmk · xmk + tmk xm

1 ≤w
k mk
+ tmk H mk ∀ k ∈ N,

that
p·x≤w and p · x + tx1 ≤ w + tH
or, equivalently, that x ∈ Bp,w,t,H . This establishes upper hemicontinuity.

Lower Hemicontinuity. To establish lower hemicontinuity of the budget set corre-

spondence, we again use the sequence definition. That is, we consider an arbitrary
domain-sequence {(pm , wm , tm , H m )} converging to (p, w, t, H), and an arbitrary point
in the range of the limiting point x ∈ Bp,w,t,H . We need to construct a range sequence
{xm } that converges to x and such that xm ∈ Bpm ,wm ,tm ,H m for all m.
Let us first take care of two special cases: If w = 0 then x ∈ Bp,0,t,H ⊆ Bp,0 = {0n }
implies that x is the null-vector. And whenever x = 0n , we can use the trivial sequence
defined by xm = 0n ∈ Bpm ,wm ,tm ,H m for all m.
When w > 0 and x 6= 0n , let
 m
w (p · x) (wm + tm H m )(p · x + tx1 )

m
x = min , x. (1)
w(pm · x) (w + tH)(pm · x + tm x1 )

Page 5 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

Let’s first show that each element xm of the sequence is in the budget set Bpm ,wm ,tm ,H m .
Indeed, we have
wm (p · x) wm
pm · xm ≤ (p · x) = (p · x) ≤ wm
w(pm · x) w
and
(wm + tm H m )(p · x + tx1 ) m
pm · xm + tm xm
1 ≤ (p · x + tm x1 )
(w + tH)(pm · x + tm x1 )
w m + tm H m
= (p · x + tx1 ) ≤ wm + tm H m .
w + tH
Also, since wm → w and pm → p, we have that
wm (p · x)
x → x. (2)
w(pm · x)
In addition, we know that tm → t and H m → H, so
(wm + tm H m )(p · x + tx1 )
x → x. (3)
(w + tH)(pm · x + tm x1 )
By (2) and (3), we see that the sequence defined in (1) converges to the desired x and
we have established lower hemicontinuity.

Berge’s Theorem. Now we can apply Berge’s theorem to obtain upper hemicontinuity
of x∗ and continuity of v. 

(c) Prove or provide a counterexample to the following statement: If the consumer’s utility
function is strictly quasiconcave, then |x∗ (p, w, t, H)| ≤ 1.

Solution: By Proposition 4.12, strict quasiconcavity of u implies that the underly-

ing preference % is strictly convex. Also, Bp,w,t,H is the intersection of two convex linear
budget sets and thus, is itself a convex set. We can now conclude by Proposition 4.13.

Alternatively, you can prove this claim directly by contraposition, and assume that
x, y ∈ x∗ (p, w, t, H) and x 6= y. This implies that u(x) = u(y), and we can apply the
definition of strict quasiconcavity of u to conclude that z = 12 x + 12 y satisfies
 
1 1
u(z) = u x + y > u(y).
2 2
However, this contradicts the optimality of x and y. 

3. Consider the following data set of four demand observations for two commodities.
x p w
1 (3, 9) (3, 3) 36
2 (12, 1) (1, 8) 20
3 (4, 2) (2, 3) 14
4 (1, 1) (4, 4) 8

Page 6 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

(a) Find %R , R , %I , and I for these observations. Check that the data satisfy GARP.

Solution: The following table summarizes the expenditure if bundle xk is chosen at

prices pj :
j 1 2 3 4
k
1 36 75 33 48
2 39 20 27 52
3 18 20 14 24
4 6 9 5 8
Hence, we get
• %R = {(x1 , x3 ), (x1 , x4 ), (x2 , x3 ), (x2 , x4 ), (x3 , x4 ), (x1 , x1 ), (x2 , x2 ), (x3 , x3 ), (x4 , x4 )}
• R = {(x1 , x3 ), (x1 , x4 ), (x2 , x4 ), (x3 , x4 }
• %I = {(x1 , x3 ), (x1 , x4 ), (x2 , x3 ), (x2 , x4 ), (x3 , x4 ), (x1 , x1 ), (x2 , x2 ), (x3 , x3 ), (x4 , x4 )}
• I = {(x1 , x3 ), (x1 , x4 ), (x2 , x4 ), (x3 , x4 }.
The data obviously satisfies GARP. 

(b) Compute Afriat numbers v j , λj which satisfy condition (G) in Afriat’s Theorem.

Solution: With the initial data we have n(1) = 2, n(2) = n(3) = 1, and n(4) = 0.
The induction argument tells us that when we break the data into j = {1, 2, 3} and
k = {4}, we can find v 4 , λ4 from v 1 , v 2 , v 3 and λ1 , λ2 , λ3 .
Once we remove {x4 , p4 , w4 }, we are left with {xi , pi , wi }3i=1 and on this data set with
3 points, we have x1 R x3 . So n0 (1) = 1, n0 (2) = n0 (3) = 0. Now, with this smaller
data set, j 0 = {1}, k 0 = {2, 3}. So if we remove the two points with empty lower contour
set, we are left with only {x1 , p1 , w1 } and we can start the algorithm like in the total
induction hypothesis of the proof.
Let v 1 = 1 and λ1 = 1.
Now when we have {xi , pi , wi }3i=1 , and v 1 = 1, λ1 = 1. Since n0 (2) = n0 (3) = 0. we can
let
v2 = v3 = min v j + λj [pj · xk − wj ]
j=1
k=2,3

v 2 = v 3 = min{(1 + 1 [(3, 3) · (12, 1) − 36]), (1 + 1 [(3, 3) · (4, 2) − 36])}

= min{4, −17}
= −17
Now we are sure that
v 2 ≤ v 1 − λ1 [p1 · x2 − w1 ]
and
v 3 ≤ v 1 − λ1 [p1 · x3 − w1 ]

Page 7 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

We want to make sure now that v 1 ≤ v i − λi [pi − ·x1 − wi ] for i ∈ {2, 3}. So we set

vj − v2
λ2 > max
{j=1} p2 · xj − w 2
1 − (−17) 18
> =
(1, 8) · (3, 9) − 20 55

and
vj − v3
λ3 > max
{j=1} p3 · xj − w 3
1 − (−17) 18
> =
(2, 3) · (3, 9) − 14 19

So let λ2 = λ3 = 1, and then we have

v 1 ≤ v 2 + λ2 [p2 · x1 − w2 ]

and
v 1 ≤ v 3 + λ3 [p3 · x1 − w3 ].

Now we go back to the full data set since we already have v i , λi for i = 1, 2, 3 that
satisfy (G). And n(1) = 2, n(2) = n(3) = 1, and n(4) = 0. We break the data into
j = {1, 2, 3},and k = {4}, and the same inductive construction will allow us to define
v 4 , λ4 .
Following the same procedure we define

v4 = min v j + λj [pj · xk − wj ]
j=1,2,3
k=4
v 4 = min{(1 + [(3, 3) · (1, 1) − 36]), (−17 + [(1, 8) · (1, 1) − 20]), (−17 + [(2, 3) · (1, 1) − 14])}
= min{−29, −28, −26}
= −29

Again, this guarantees that

v 4 ≤ v i − λi [pi · x4 − wi ]

for i = {1, 2, 3}.

The last step is to find λ4 to have

v i ≤ v 4 + λ4 [p4 · xi − w4 ]

and we do so by setting

Page 8 of 9
Econ 201A
Fall 2013
Problem Set 3 Suggested Solutions

vj − v4
λ4 > max
{j=1,2,3} p4 · xj − w 4

v1 − v4 v2 − v4 v3 − v4
     
= max , ,
p 4 · x1 − w 4 p4 · x2 − w4 p4 · x3 − w4
     
1 − (−29) −17 − (−29) −17 − (−29)
= max , ,
(4, 4) · (3, 9) − 8 (4, 4) · (12, 1) − 8 (4, 4) · (4, 2) − 8
 
30 12 12 3
= max , , =
40 44 16 4

So we can let λ4 = 1 as well. And

v i ≤ v 4 + λ4 [p4 · xi − w4 ]

for all i = 1, 2, 3. Hence we found the Afriat numbers for this data set. 

Page 9 of 9