一、3预算集与效用最大化

3.1 预算集的定义
已经在一、6.1中介绍过，参见 MWG 2.D。

3.2 效用最大化问题的求解∗ （MWG: 3.D, M.G, M.J, M.K, M.L;

JR: 1.3, 1.4.1, A2.3; V: chapters 7, 8 and 27）

3.3 马歇尔需求对应（函数）的性质
（MWG: 3.D, Appendix A to Chapter 3; JR: 1.5.1; V: chapters 7, 8 and 9

3.4 间接效用函数的定义与性质 （MWG: 3.D; JR: 1.4.1; V: chapters 7, 8 and 9

3.5 间接效用函数与马歇尔需求的关系（将在后面的一、5中讲）

3.D The Utility Maximization Problem

1. The UMP: max u(x) s.t. p · x ≤ w
x≥0

I Figure 3.D.1: single solution versus multiple solutions

2. Proposition 3.D.2

Assume u(·) is continuous, representing locally non-satiated % . Then, the Walrasian demand
correspondence x(p, w) has the following properties:

(1) HD0 in (p, w)

(2) Walras’ Law: p · x = w, ∀x ∈ x(p, w)

(3) Convexity/Uniqueness: If % is convex, so that u(·) is quasiconcave, then x(p, w) is a convex

set. If % is strictly convex, so that u(·) is strictly quasiconcave, then x(p, w) consists of a single
element.

I Proof (follows from examination of the UMP):

Idea of Proof for (1): (αp, αw), α > 0, implies the same set of feasible consumption bundles as
(p, w).

Idea of Proof for (2): Use local non-satiation. Suppose p · x < w for some x ∈ x(p, w). Then,
can find another bundle y near x that is feasible (p · y < w) and strictly preferred to x (y ≻ x),
contradicting the optimality of x.

Proof of (3):

(i) convex %

Let x and x′ be two elements of x(p, w).

1
 To show x(p, w) is convex, we need to show that any convex combination x′′ = αx + (1 − α)x′ ,
α ∈ [0, 1], also belongs to x(p, w).

Since p · x ≤ w and p · x′ ≤ w, we have p · x′′ = p · [αx + (1 − α)x′ ] ≤ w. Thus, x′′ is feasible under
(p, w).

Next, note that u(x) = u(x′ ). Denote the utility level by u∗ . Quasiconcavity of u(·) implies that
u(x′′ ) ≥ u∗ .

Therefore, x′′ is feasible under (p, w) and one of the best choices. So x′′ ∈ x(p, w).

(ii) strictly convex %

In this case, we prove by contradiction. Suppose that x(p, w) is not a singleton. Then, it contains
at least two distinct elements. Pick two distinct elements x and x′ from the set x(p, w).

Going through the same argument as in case (i) above but using strict quansiconcavity of u(·),
we can get u(x′′ ) > u∗ , implying that neither x nor x′ belongs to x(p, w). This contradicts the
optimality of x and x′ . Therefore, x(p, w) must be a singleton.

3. First-order condition characterization for continuously differentiable u(·)

3.1 Kuhn-Tucker conditions

I fundamental mathematical result: Theorem M.K.2 (Kuhn-Tucker conditions)

• See also the references listed in the section title above, as well as Appendix H, Gravelle, H. and
Rees, R. (2004). Microeconomics, 上海财经大学出版社..

I a simpler, yet rigorous, way to motivate the Kuhn-Tucker conditions: Chapter 13 (especially
section 13.1), Chiang, A. C. and Wainwright, K. (刘学、顾佳峰 译) (2006). 数理经济学的基本
方法（第四版）, 北京大学出版社. (We will take this route, due to time constraint.)

I summary of procedure to solve maximization problems with inequality constraints (including

non-negativity constraints):

• First, rewrite the optimization problem into the following form:

max f (x1 , . . . , xn )
x1 ,...,xn

s.t. g i (x1 , . . . , xn ) ≤ ri , i ∈ {1, 2, ..., m}

xj ≥ 0, j ∈ {1, 2, ..., n}

• Second, write out the Lagrangian function:

∑
m
L = f (x1 , . . . , xn ) + λi (ri − g i (x1 , x2 , ..., xn ))
i=1

2
 • Third, write out the Kuhn-Tucker first-order necessary conditions:

∂L ∂L
≤ 0, xj ≥ 0, and xj = 0, j ∈ {1, 2, ..., n}
∂xj ∂xj
∂L ∂L
≥ 0, λi ≥ 0, and λi = 0, i ∈ {1, 2, ..., m}
∂λi ∂λi

• Solve the KT conditions by trial and error. In this step, the complementary slackness conditions
play an important role.

I Example 1: min x21 + x22 , s.t. x1 + x2 ≥ 2, x21 − 2 ≤ − 12 , x1 ≥ 0, x2 ≥ 0.

x1 ,x2

Solution:

Step 1: Transform the problem into a maximization problem:

max − x21 − x22

x1 ,x2

s.t. − x1 − x2 ≤ −2,
3
x21 ≤ , x1 ≥ 0, x2 ≥ 0.
2

Step 2: Write out the Lagrangian function:

3
L = −x21 − x22 + λ1 (−2 + x1 + x2 ) + λ2 ( − x21 )
2

3
 Step 3: Write out the Kuhn-Tucker conditions
∂L ∂L
= −2x1 + λ1 − 2x1 λ2 ≤ 0, x1 ≥ 0, and x1 = 0 [1]
∂x1 ∂x1
∂L ∂L
= −2x2 + λ1 ≤ 0, x2 ≥ 0, and x2
= 0 [2]
∂x2 ∂x2
∂L ∂L
= −2 + x1 + x2 ≥ 0, λ1 ≥ 0, and λ1 = 0, [3]
∂λ1 ∂λ1
∂L 3 ∂L
= − x21 ≥ 0, λ2 ≥ 0, and λ2 = 0, [4]
∂λ2 2 ∂λ2

Step 4: Solve by trial and error: Make assumptions with respect to multipliers or choice variables,
drop some of the constraints, and find out whether there are solutions satisfying all the conditions.

• Case 1: Suppose x1 = 0, x2 = 0.

[3]: −2 ≥ 0 (discard Case 1)

• Case 2: Suppose x1 = 0, x2 > 0.

[1]: λ1 ≤ 0 ⇒ λ1 = 0 ⇒ [2] : −2x2 = 0 ⇒ x2 = 0 (contradicts x2 > 0).

• Case 3: Suppose x1 > 0, x2 = 0.

√
[3]: x1 ≥ 2 contradicts [4]: x1 ≤ 3/2 ≈ 1.22.

• Case 4: Suppose x1 > 0, x2 > 0.

[2]: x2 = λ1
2 ⇒ λ1 > 0 ⇒ [3] : x1 + x2 = 2

Suppose λ2 > 0.
√ √ √
Then [4]: x1 = 3/2 ⇒ [3] : x2 = 2 − 3/2 ⇒ λ1 = 4 − 2 3/2.
√ √ √
Then [1]: (−2 3/2) + (4 − 2 3/2) − 2( 3/2)λ2 ⇒ λ2 < 0 (contradicts λ2 > 0).

Therefore, λ2 = 0 ⇒ [1] : x1 = λ1
2 . Recall that x2 = λ1
2 , so we have x1 = x2 .

⇒ x1 = 1, x2 = 1.

• (Actually, the first three constraints imply that x2 > 0, so the fourth constraint x2 ≥ 0 is
satisfied with strict inequality. Therefore, we could have directly written the second line of the
∂L
Kuhn-Tucker conditions as ∂x2 = 0. Then our calculations may be simplified.)

3.2 Kuhn-Tucker necessary conditions for the UMP

• if x∗ ∈ x(p, w) is a solution to the UMP, then there exists a Lagrange multiplier λ ≥ 0 such
that for all ℓ = 1, · · · , L :

∂u(x∗ )
≤ λpℓ , with equality if x∗ℓ > 0.
∂xℓ

4
 • equivalently, if we let ∇u(x) = [∂u(x)/∂x1 , · · · , ∂u(x)/∂xL ] denote the gradient vector of u(·)
at x, we can write the conditions above as

∇u(x∗ ) ≤ λp and x∗ · [▽u(x∗ ) − λp] = 0

• The interpretations discussed below are based on MWG Section 3.D and Nicholson and Snyder
(2007) Chapter 4.

I interior solution (Figure 3.D.4.a)

∂L ∂u(x∗ )
=0 ⇒ = λpℓ , i = 1, . . . , L or ∇u(x∗ ) = λp
∂xℓ ∂xℓ

• interpretation using just one FOC: Rewrite the FOC as

M Ui
pi = , ∀i = 1, . . . , L.
λ

Consider a situation where a person’s marginal utility of income (λ) is constant over some range.
Then variations in the price he or she must pay for good i, pi , are directly proportional to the
extra utility derived from that good.

◃ At the margin, therefore, the price of a good reflects an individual’s willingness to pay for
one more unit.

◃ This is a result of considerable importance in applied welfare economics because willingness

to pay can be inferred from market reactions to prices. This insight can be used to evaluate
the welfare effects of price changes and to discuss a variety of questions about the efficiency
of resource allocation.

• interpretation 1 using any two FOC’s: From the FOC’s for any two goods ℓ and k, we can get

M Uℓ pℓ
= .
M Uk pk

◃ The consumer’s internal, psychic rate of trade-off between the two goods, M RSℓk , is equated
to the external, market-offered rate at which the goods can be traded, pℓ /pk .

◃ If M RSℓk ̸= pℓ /pk , there will be a way to increase utility by adjusting consumption. For
example, if M RSℓk > pℓ /pk , increasing the consumption of good ℓ by dxℓ > 0, combined
with a decrease in good k’s consumption equal to dxk = −(pℓ /pk )dxℓ < 0, would be feasible
and would yield a utility change of M Uℓ dxℓ − M Uk (pℓ /pk )dxℓ > 0. See Figure 3.D.4.a, where
ℓ = 1 and k = 2.

5
 • interpretation 2 using any two FOC’s: Rewrite the condition slightly to get
M Uℓ M Uk
= = λ.
p p
| {zℓ } | {zk }
M U per yuan M U per yuan
spent on good ℓ spent on good k

◃ Each good purchased should yield the same marginal utility per yuan spent on that good.
Each good therefore should have an identical (marginal) benefit-to-(marginal)-cost ratio. If
this were not true, one good would promise more ‘marginal enjoyment per dollar’ than some
other good, and funds would not be optimally allocated.

An extra yuan should yield the same ‘additional utility’ no matter which good it is spent on.
The common value for this extra utility is given by the Lagrange multiplier for the consumer’s
budget constraint (that is, by λ). Consequently, λ can be regarded as the marginal utility of
an extra dollar of consumption expenditure (the marginal utility of ‘income’).

I boundary solution (Figure 3.D.4.b)

∂L ∂u(x∗ )
= − λpℓ ≤ 0, < 0 for some ℓ or ∇u(x∗ ) ≤ λp;
xℓ ∂xℓ
∂L ∂u(x∗ )
if = − λpℓ < 0, then xℓ = 0.
xℓ ∂xℓ
• interpretation using only the FOC for good ℓ for which x∗ℓ = 0 : Rewrite its FOC as
M Uℓ
pℓ > .
λ

◃ If the price of good ℓ exceeds its marginal value to the consumer (M Uℓ /λ), it will not be
purchased (xℓ = 0).

◃ Thus, the mathematical results conform to the commonsense idea that individuals will not
purchase goods that they believe are not worth the money.

• interpretation 1 using two FOC’s, one for a good purchased and one for a good not purchased
(See Fig. 3.D.4.b): Rewrite the FOC’s as
M U1 p1
> .
M U2 p2

◃ Can you interpret?

• interpretation 2 using two FOC’s, one for a good purchased and one for a good not purchased
(See Fig. 3.D.4.b): Rewrite the FOC’s as
∂u(x∗ )/∂x1 ∂u(x∗ )/∂x2
=λ≥
p1 p2
| {z } | {z }
M U per yuan M U per yuan
spent on good 1 spent on good 2

6
 ◃ The last yuan of the consumer brings him a (weakly) larger increment in utility if it is spent
on good 1 than if it is spent on good 2. So the consumer wishes to consume more of good 1,
even after spending all of his income on good 1 alone.

I value of Lagrange multiplier at optimum = marginal utility of wealth at optimum

• derivation on p. 54 for interior case:

◃ ∇u(x(p, w)) · Dw x(p, w) = λp · Dw x(p, w) = λ

◃ The last equality follows from the Walras’ Law: p · x(p, w) = w ⇒ p · Dw x(p, w) = 1.

I (global and local sufficient conditions: Mathematical Appendix M.K of MWG)

I (continuity of x(p, w) : Appendix A to MWG Chapter 3)

4. The Indirect Utility Function v(p, w)

I value function for the UMP v(p, w) ≡ u(x(p, w))

• references for value function (MWG M.L)

• Cobb-Douglas example.
1−α
max u(x1 , x2 ) = xα
1 x2
{x1 ,x2 }

s.t. p1 x1 + p2 x2 ≤ w

◃ solution ⇒ Walrasian demand functions

αw (1 − α)w
x1 (p, w) = , x2 (p, w) =
p1 p2

◃ Substituting these solutions into the objective function, we get the indirect utility function
( )α ( )1−α
αw (1 − α)w −(1−α)
v(p, w) ≡ u(x(p, w)) = = αα (1 − α)1−α p−α
1 p2 w
p1 p2

5. Proposition 3.D.3

When u(·) is continuous and represents a locally nonsatiated preference %, v(p, w) has the foll-
owing properties:

(1) HD0

(2) Strictly increasing in w and nonincreasing in pℓ for any ℓ

(3) quasiconvex: i.e., the set {(p, w) : v(p, w) ≤ v} is convex for any v

(4) continuous in p and w

7
I Proof of (3):

According to the definition of quasi-convexity, suppose that v(p, w) ≤ v and v(p′ , w′ ) ≤ v, and
we want to show that, for any α ∈ [0, 1], v(p′′ , w′′ ) ≤ v, where p′′ ≡ αp + (1 − α)p′ and w′′ ≡
αw + (1 − α)w′ .

Equivalently, we want to show that, for all x satisfying p′′ · x ≤ w′′ , we have that u(x) ≤ v.

Note that p′′ · x ≤ w′′ is

αp · x + (1 − α)p′ · x ≤ αw + (1 − α)w′ .

Thus, we cannot simultaneously have p · x > w and p′ · x > w′ . In other words, either p · x ≤ w or
p′ · x ≤ w′ (or both) must hold. (Note that, when p · x ≤ w holds, we may have either p′ · x ≤ w′
or p′ · x > w′ . Similarly, p′ · x ≤ w′ can be compatible with either p · x ≤ w or p · x > w.)

If p · x ≤ w holds, we have u(x) ≤ v(p, w) ≤ v, so we have the result. Similarly, if p′ · x ≤ w′

holds, we have u(x) ≤ v(p′ , w′ ) ≤ v, and we also have the result.

• idea of proof (See Figure 3.D.5): The budget line under (p′′ , w′′ ) is a convex combination of
those under (p, w) and (p′ , w′ ). So it cannot be higher than the higher segments of the latter
two budget lines. Then, the maximal utility attainable under (p′′ , w′′ ) cannot be greater than
those under the latter two.

I proof of (4): ...

6. Indirect utility function depends on the utility representation chosen.

If u(x) → UMP → v(p, w), then f (u(x)) → UMP → f (v(p, w)).

7. Finding the indirect utility function for CES utility

( ) 1
1− 1 1− 1 1− 1
u(x1 , x2 ) = ax1 σ + bx2 σ σ
, a, b, σ > 0, σ ̸= 1
( ) 1
1− 1 1− 1 1− 1
L= ax1 σ + bx2 σ σ
+ λ(w − p1 x1 − p2 x2 )

For visual clarity, we will use the symbol u in place of the full expression of the CES utility
function. The first-order condition with respect to x1 is:
( )
1 1
−1 1 −σ
1

1 u
1− 1
σ a 1− x1 = λp1 .
1− σ
σ

Simplifying, we have
1
1 −1 − 1
au 1− σ x1 σ = λp1 .

8
 Similarly, the first-order condition with respect to x2 is
1
1 −1 − 1
bu 1− σ x2 σ = λp2 .

Divide the equations through to get

( ) σ1
a x2 p1
=
b x1 p2

or ( )σ
x2 bp1
= (∗)
x1 ap2
From this, we get ( )σ
bp1
x2 = x1
ap2
Substituting this into the budget constraint and rearranging yield
( )σ
bp1
p 1 x1 + p 2 x1 = w
ap2

w aσ p1−σ
1
x1 =
p1 aσ p1−σ
1 + bσ p1−σ
2

Similarly, we have
w bσ p1−σ
2
x2 =
p2 aσ p1−σ
1 + bσ p1−σ
2

To get the indirect utility function, we substitute the Walrasian demand functions above into the
CES utility function:

v(p1 , p2 , w)
  1
1− σ1
σ −1 σ−1
1 (1−σ)(1− σ
1
) 1
1− σ σ −1 σ−1
1 (1−σ)(1− σ
1
) 1− σ1
=  aw p1 a p1 + bw p2 b p2 
1
(aσ p1−σ
1 + bσ p1−σ
2 )1− σ
[ ] 1
1− 1
aσ p1−σ
1 + bσ p1−σ
2
σ
= 1 w
(aσ p1−σ
1 + bσ p1−σ
2 )1− σ
w
= 1
(aσ p1−σ
1 + bσ p1−σ
2 ) 1−σ

Observe that the denominator also looks like a CES function of the prices.

I Using expression (∗) derived from the first-order conditions above, we can calculate the elasticity
of substitution for this CES utility function in the following simple way. First, take the logarithm
of both sides of (∗). We get
( ) ( )σ ( )
x2 b p1
ln = ln + σ ln .
x1 a p2

9
 Then, the elasticity of substitution can be calculated by taking derivative:
( )
x2
d ln x1
elasticity of substitution ε ≡ ( ) = σ.
p1
d ln p2

So the CES utility function indeed has a constant elasticity of substitution of σ.

8. Example: Verify the quasiconvexity of v(p, w) found above for the Cobb-Douglas utility function.
−(1−α)
I If we ignore the constant coefficients, we essentially need to verify that f (p1 , p2 , w) ≡ p−α
1 p2 w
is quasiconvex in t ≡ (p1 , p2 , w).

I We will use the second-derivative test.

I According to Theorem M.C.4 of MWG, f (t) is quasiconvex if and only if the Hessian matrix
H ≡ D2 f (t) is positive semidefinite in the subspace {z ∈ IR3+ : ∇f (t) · z = 0} for every t.

I According to Theorem M.D.3, the above is true if and only if

r ∇f (t)
π π
r Hr
(−1)S T ≥0
(r ∇f (t)π ) 0

for r = S + 1, · · · , N and every permutation π, where r Hrπ is formed by permuting both the rows
and columns of r Hr and r ∇f (t)π is formed by permuting only the rows of matrix r ∇f (t).

I To begin with, let us understand the meaning of the symbols used in the two theorems:

• S : vector ∇f (t) is an N -by-S matrix where S ≤ N and the rank of ∇f (t) is S

• In our problem,


f p1
∇f (t) =  fp2 
fw

is a 3-by-1 matrix, so S = 1.

• N : this is the number of rows and columns of the Hessian H and is also the number of rows of
∇f (t)

• In our problem, N = 3.

• r Hrπ : the sub-matrix formed by taking the first r rows and the first r columns of the square
matrix H π , where H π means the permutation π of the entire matrix H

• r ∇f (t)π : is the first r rows of the permutation π of the vector ∇f (t), ∇f (t)π

• Since S = 1 and N = 3 in our case, r goes from S + 1 = 2 to 3. That is, we only need to
consider r = 2, 3.

10
 • Since P33 = 6, we have six permutations: π1 = (1, 2, 3), π2 = (1, 3, 2), π3 = (2, 1, 3), π4 = (2, 3, 1),
π5 = (3, 1, 2) and π6 = (3, 2, 1).

I Now, we can construct the determinants to be signed.

• We know the Hessian is  

f11 f12 f13
H =  f21 f22 f23 
f31 f32 f33
 
α(α + 1)p−α−2
1 pα−1
2 w −α(α − 1)p−α−1
1 p2 α−2 w −αp−α−1
1 p2α−1
 
 α−2 
=  −α(α − 1)p−α−1 pα−2 w (α − 1)(α − 2)p−α α−3
1 p2 w (α − 1)p−α p ,
 1 2 1 2

−αp−α−1
1 pα−1
2 (α − 1)p−α α−2
1 p2 0
and the gradient vector is
 
  −αp−α−1
1 pα−1
2 w
f1  
   −α α−2 
∇f (t) = f2 =  (α − 1)p1 p2 w  .
 
f3
p−α
1 p2
α−1

• For π1 = (1, 2, 3), H π1 is just H itself, and ∇f (t)π1 is also just ∇f (t).

◃ First, let r = 2. We need to sign the determinants

r ∇f (t) 2 ∇f (t)
π1 π1 π1 π1
r Hr 2 H2
(−1)S π1 T = (−1)1 π1 T
(r ∇f (t) ) 0 (2 ∇f (t) ) 0

Since r = 2, we take the first two rows and the first two columns of H π1 and take the first
two rows of ∇f (t)π1 to construct the determinants. These are
[ ]
f f12
π1
2 H2 = 11
f21 f22

and ][
f1
2 ∇f (t)
π1
= .
f2

◃ The corresponding determinant is

f11 f12 f1
2 ∇f (t)
π1 π1
2 H2
det1,2 = (−1) 1
π1 T = − f21 f22 f2 ,
(2 ∇f (t) ) 0
f1 f2 0

where the first subscript in det1,2 indicates permutation π1 and the second subscript indicates
r = 2.

◃ Next, let r = 3. Then, 3 H3π1 is H itself, and 3 ∇f (t)π1 is ∇f (t) itself.

11
 ◃ The corresponding determinant is

f11 f12 f13 f1

3 H3
π1
∇f (t)π1
f f22 f23 f2
det1,3 = (−1)1 T
3
= − 21 ,
(3 ∇f (t)π1 ) 0 f31 f32 f33 f3
f1 f2 f3 0

where the first subscript in det1,3 indicates permutation π1 and the second subscript indicates
r = 3.

• For π2 = (1, 3, 2), we have that the permuted Hessian is

 
f11 f13 f12
π2 
H = f31 f33 f32 
f21 f23 f22

and the permuted gradient vector is

 
f1
∇f (t)π2 =  f3  .
f2

◃ First, let r = 2. We take the first two rows and the first two columns of H π2 and take the
first two rows of ∇f (t)π2 to construct the determinants. These are
[ ]
π2 f11 f13
2 H2 =
f31 f33

and [
]
f1
2 ∇f (t)
π2
= .
f3

◃ The corresponding determinant is

f11 f13 f1
2 ∇f (t)
π2 π2
2 H2
det2,2 = (−1) 1
π2 T = − f31 f33 f3
(2 ∇f (t) ) 0
f1 f3 0

◃ Next, let r = 3. Then, 3 H3π2 is H π2 , and 3 ∇f (t)π2 is ∇f (t)π2 .

◃ The relevant determinant is

f11 f13 f12 f1
3 ∇f (t)
π2 π2
3 H3 f f33 f32 f3
det2,3 = (−1)1 π2 T = − 31
(3 ∇f (t) ) 0 f21 f23 f22 f2
f1 f3 f2 0

• For π3 = (2, 1, 3), the determinants are, omitting some intermediate steps,

f22 f21 f2
2 ∇f (t)
π3 π3
2 H2
det3,2 = (−1)1 π3 T = − f12 f11 f1
(2 ∇f (t) ) 0
f2 f1 0

12
 f22 f21 f23 f2
3 H3
π3
∇f (t) π3
f f11 f13 f1
det3,3 = (−1)1 T
3
= − 12
(3 ∇f (t)π3 ) 0 f32 f31 f33 f3
f2 f1 f3 0
• Quiz: Write out the determinants for the remaining three permutations π4 through π6 .

I The remaining task is simply to calculate det1,2 , det1,3 , det2,2 , det2,3 , · · · , det6,2 , and det6,3 and
check their signs.

• It is found that
det1,2 > 0, det2,2 > 0, det3,2 > 0, det4,2 > 0, det5,2 > 0, det22 > 0,

det1,3 = det2,3 = det3,3 = det4,3 = det5,3 = det6,3 = 0.

That is, all the determinants when r = 2 are strictly positive, and all the determinants when
r = 3 are zero. Thus, all the determinants are non-negative.

• Therefore, according to Theorem M.D.3, H is positive semidefinite in the subspace {z ∈ IR3+ :

∇f (t) · z = 0} for every t. Then, by Theorem M.C.4, f (t) is quasiconvex. Hence, the indirect
utility function for the Cogg-Douglas utility is quasiconvex in its arguments (p1 , p2 , w).

13
I further graphical illustrations of the quasiconvexity of v(p1 , p2 , w) (figures are taken from Muñoz-
Garcia (2017), pages 97 and 98, and Varian (1992), page 104)

• Figure 2.17 of M-G: By the quasiconvexity of v(p, w), if the maximum utility associated with
a given pair of prices and wealth (point A in figure) is weakly higher than the maximum utility
associated with another pair of prices and wealth (point B), then the convex comination of points
A and B (point C) yields a maximum utility that is weakly lower than that associated with A.

1 + (1 − λ)p1 and has an intermediate

That is, if the consumer faces an intermediate price level λpA B

wealth level λwA + (1 − λ)wB , the maximum utility level that he will be able to reach is lower
than that at point A ≡ (pA
1 , wA ).

• Figure 2.18 of M-G: The pairs of prices and wealth for which the consumer reaches a lower
utility level than that under pair (p∗ , w∗ ) defines a convex set. That is, v(p, w) is quasiconvex if
the set of (p, w) pairs for which v(p, w) ≤ v(p∗ , w∗ ) is convex. In the figure, the shaded area is
the set {(p, w) : v(p, w) ≤ v(p∗ , w∗ )}, and it is convex.

• In Figure 7.2 of Varian, a typical set of ‘price indifference curves’ is depicted. These are just
the level sets of the indirect utility function in the (p1 , p2 ) dimensions. By the properties of the
v(p, w), maximum utility is non-decreasing as one moves towards the origin, and the lower contour
sets are convex. Note that the lower contour sets lie to the northeast of the price indifference
curves since indirect utility declines with higher prices.

Recommended Exercises: 3.D: 3, 4, 5, 6(b), 6(c), 7

14