Classical Demand Theory

Stergios Athanasoglou

Microeconomics Module 1
PhD DEFAP-ECOSTAT
University of Milan - Bicocca

1 / 33
The consumer’s decision problem

Assume throughout, unless otherwise specified, that the

consumer has rational, locally nonsatiated, and continuous
preferences. This implies that there exists a continuous utility
function u representing these preferences. Also, X = RL+ ,
p  0 and w > 0.

Def. Given p and w , the utility maximization problem (UMP) is defined

as:

max u(x)
x≥0
s.t. p · x ≤ w.

The consumer chooses the consumption bundle that maximizes his

utility subject to the budget constraint.

2 / 33
The utility maximization problem

Does the UMP have a solution?

Theorem
The UMP admits an optimal solution.

Proof. Under the assumptions that p  0, w > 0, the budget set Bp,w is
closed and bounded, and thus compact. A continuous function
always has a maximum value on any compact set.

3 / 33
Walrasian demand

Recall the Walrasian demand correspondence. In the context of the

UMP it can be expressed as:

x(p, w ) = argmax u(x).

x∈Bp,w

Walrasian demand is defined as the set of optimal solutions of the

UMP.

Walrasian demand is often a function, but can be multivalued (i.e.,

a correspondence) as well.

4 / 33
Properties of Walrasian demand

Proposition

Suppose u is a continuous utility function representing locally nonsa-

tiated preferences  on X = RL+ . The Walrasian demand correspon-
dence satisfies the following properties:

(i) Homogeneity of degree 0: x(αp, αw ) = x(p, w ) for all

p, w , α > 0.

(ii) Walras’ law: p · x = w for all x ∈ x(p, w ).

(iii) Convexity/Uniqueness: If  is convex then x(p, w ) is a convex

set. Moreover, if  is strictly convex then x(p, w ) consists of a
single element.

5 / 33
Properties of Walrasian demand: Proof of convexity
Proof. Parts (i) and (ii) are trivial. Let us focus on part (iii).
1. Since  is convex, the utility function u representing it is
quasiconcave. Suppose x, x 0 ∈ x(p, w ) and x 6= x 0 .

2. Note that, by definition,

x, x 0 ∈ x(p, w ) ⇒ u(x) = u(x 0 ) ≡ u ∗ .

3. Consider x 00 = αx + (1 − α)x 0 for α ∈ [0, 1]. By quasiconcavity

u(x 00 ) ≥ min{u(x), u(x 0 )} = u ∗

4. In addition
p · x 00 = αp · x + (1 − α)p · x 0 ≤ w ⇒ x 00 ∈ Bp,w .

⇒ We conclude x 00 ∈ x(p, w ), establishing the convexity of

x(p, w ).
6 / 33
Properties of Walrasian demand: Proof of uniqueness

Now we prove that x(p, w ) is a singleton when preferences are

strictly convex.

1. Since  is strictly convex, the utility function u representing it is

strictly quasiconcave. Suppose, in contradiction of uniqueness, that
x, x 0 ∈ x(p, w ) and x 6= x 0 .

2. Letting x 00 = αx + (1 − α)x 0 for some α ∈ (0, 1), strict

quasiconcavity implies

u(x 00 ) > min{u(x), u(x 0 )} = u ∗ .

Furthermore, as before, x 00 ∈ Bp,w .

⇒ Thus x 00 is feasible and yields higher utility than either x, x 0 . We

conclude x, x 0 6∈ x(p, w ), which is a contradiction.

7 / 33
Karush-Kuhn-Tucker necessary conditions
Def. Suppose u is continuously differentiable. If x ∗ is a solution to the
UMP, then x ∗ ≥ 0, p · x ∗ ≤ w (satisfied with equality if preferences
are locally nonsatiated), and there exists a constant λ ≥ 0 such that

∂u(x ∗ )
= λpi , for all i = 1, 2, ..., L with xi∗ > 0
∂xi
∂u(x ∗ )
≤ λpi , for all i = 1, 2, ..., L with xi∗ = 0.
∂xi

The above conditions are the Karush-Kuhn-Tucker (KKT) or

first-order (optimality) conditions applied to the UMP. The scalar λ
is known as the Lagrange multiplier. (Refresh your constrained
optimization!)

A truly excellent reference on optimization, freely available online:

http://web.stanford.edu/ boyd/cvxbook/
8 / 33
Marginal rate of substitution

Def. The marginal rate of substitution of good i for good j at x ∗ is given

by
∂u(x ∗ )
∂xi
MRSij = ∂u(x ∗ )
∂xj .

Amount of good j the consumer must be given to compensate for

the loss of one unit of good i.

If xi∗ , xj∗ > 0, then the KKT conditions imply

∂u(x ∗ )
∂xi pi
∂u(x ∗ )
= .
pj
∂xj

The MRS is equal to the price ratio of the two goods.

9 / 33
Interpreting the Lagrange multiplier

The Lagrange multiplier captures the marginal change in utility

from a marginal change in wealth.

Here’s why.

Suppose x(p, w )  0 and differentiable. Then

∂u(x(p, w )) Chain Rule

= ∇u(x(p, w )) · Dw (x(p, w ))
∂w
KKT
= λp · Dw (x(p, w ))
Walras’ Law/Engel Aggr.
= λ · 1 = λ.

The marginal utility of wealth is precisely λ.

10 / 33
Example of Walrasian demand: Cobb-Douglas utility
⇒ Let’s calculate the Walrasian demand of a Cobb-Douglas utility function
for L = 2, defined as
u(x1 , x2 ) = kx1α x21−α
for some α ∈ (0, 1). We perform the calculation in a sequence of steps.

1. Easier to substitute u(x1 , x2 ) in the objective function of the UMP with

the following increasing transformation (why can we do this?)
 
1
log u(x1 , x2 ) = α log(x1 ) + (1 − α) log(x2 ).
k

2. Since the objective function is increasing, Walras’ law holds and the
budget constraint will bind at optimality. Thus, the UMP can be
rewritten as:
max α log(x1 ) + (1 − α) log(x2 )
x1 ≥0,x2 ≥0
s.t. p1 x1 + p2 x2 = w .

11 / 33
Cobb-Douglas utility
3. Since log 0 = −∞, all optimal solutions must be positive, i.e.
x ∈ x(p, w ) ⇒ x  0.

4. The KKT conditions imply that there exists λ ≥ 0 such that:

∂u(x) α
= λp1 ⇒ = λp1
∂x1 x1
∂u(x) (1 − α)
= λp2 ⇒ = λp2 .
∂x2 x2

αx2 p1
5. Dividing the first equation by the second obtains (1−α)x 1
= p2
.
Combining this equality with the binding budget constraint
p1 x1 + p2 x2 = w , yields the unique optimal solution
αw (1 − α)w
x1∗ = , x2∗ = .
p1 p2
Thus,  
αw (1 − α)w
x(p, w ) = , .
p1 p2

12 / 33
Indirect Utility function

Def. Recall the formulation of the UMP. The indirect utility function
v (p, w ) is defined by

v (p, w ) = max u(x).

x∈Bp,w

Measures the maximum achievable utility given the consumer’s

budget constraint.

13 / 33
Properties of the Indirect Utility function

Proposition
Suppose u is a continuous utility function representing locally nonsa-
tiated preferences  on X = RL+ . The indirect utility function v (p, w )
is:
(i) Homogeneous of degree 0;

(ii) Increasing in w and nonincreasing in pl for all l = 1, ..., L;

(iii) Quasiconvex;

(iv) Continuous.

14 / 33
Properties of the indirect utility function: Quasiconvexity

Proof. Let’s focus on proving (iii). We need to show that the set
{(p, w ) : v (p, w ) ≤ v̄ } is convex for all v̄ .

1. Suppose v (p, w ) ≤ v̄ and v (p 0 , w 0 ) ≤ v̄ and consider

(p 00 , w 00 ) = α(p, w ) + (1 − α)(p 0 , w 0 ) for some α ∈ [0, 1].

2. We must show that v (p 00 , w 00 ) ≤ v̄ . This is analogous to showing

that, for all x ≥ 0 satisfying p 00 · x ≤ w 00 we have u(x) ≤ v̄ .

15 / 33
Properties of the indirect utility function: Quasiconvexity

3. To this end, note that

p 00 · x ≤ w 00 ⇒ αp · x + (1 − α)p 0 · x ≤ αw + (1 − α)w 0

⇒ p · x ≤ w or p 0 · x ≤ w 0 or both.

But, if

(3a) p · x ≤ w , then u(x) ≤ v̄ . [If not, v (p, w ) > v̄ , a

contradiction.]
(3b) p 0 · x ≤ w 0 , then u(x) ≤ v̄ . [If not, v (p 0 , w 0 ) > v̄ , a
contradiction].

⇒ We conclude that u(x) ≤ v̄ , implying v (p 00 , w 00 ) ≤ v̄ .

16 / 33
The expenditure minimization problem

Def. For given p  0 and u ∈ R, the expenditure minimization problem

(EMP) is given by

min p·x
x≥0
s.t. u(x) ≥ u.

The EMP computes the minimal level of wealth that ensures utility
level u.

Existence of a solution to the EMP holds under very general

conditions.

17 / 33
UMP-EMP Duality

UMP-EMP Duality

Suppose u is a continuous utility function representing locally non-

satiated preferences  on X = RL+ and that prices are given by p.
Then,

(i) If x ∗ is optimal in the UMP for wealth w , then x ∗ is optimal in

the EMP for required utility u = u(x ∗ ). The minimum
expenditure is exactly w .

(ii) If x ∗ is optimal in the EMP when the required utility is

u > u(0), then x ∗ is optimal in the UMP for wealth w = p · x ∗ .
The maximum utility is exactly u.

18 / 33
Expenditure function

Def. Consider the EMP given p, u. The expenditure function e(p, u) is

defined by
e(p, u) = min p · x.
x≥0, u(x)≥u

⇒ UMP-EMP duality connects the indirect utility function to the

expenditure function:

e(p, v (p, w )) = w , v (p, e(p, u)) = u.

For a fixed price vector p, e(p, ·) and v (p, ·) are inverses to one
another.

19 / 33
Properties of the expenditure function

Proposition

Suppose u is a continuous utility function representing locally non-

satiated preferences on X = RL+ . The expenditure function e(p, u)
satisfies the following properties

(i) Homogeneity of degree 1 in p .

(ii) Strictly increasing in u and nondecreasing in pl for all l.

(iii) Concavity in p.

(iv) Continuity in both p and u.

20 / 33
Hicksian demand
Def. The set of optimal solutions of the EMP is denoted by h(p, u) and
is known as the Hicksian demand correspondence (function if single
valued).

Proposition
Suppose u is a continuous utility function representing locally non-
satiated preferences  on X = RL+ . Then for any p  0 the Hicksian
demand correspondence h(p, u) satisfies the following properties

(i) Homogeneity of degree 0 in p;

(ii) No excess utility: For all x ∈ h(p, u), we have u(x) = u.

(iii) Convexity/Uniqueness: If  is convex, then h(p, u) is a convex

set. Moreover, if  is strictly convex, then h(p, u) consists of a
single element.

21 / 33
KKT conditions for the EMP

Def. Suppose u is continuously differentiable. If x ∗ ∈ h(p, u) is a

solution to the EMP, then x ∗ ≥ 0, u(x ∗ ) ≥ u (satisfied with
equality if preferences are locally nonsatiated), and there exists a
constant λ ≥ 0 such that

∂u(x ∗ )
λ = pl , for all l = 1, ..., L with xl∗ > 0
∂xl
∂u(x ∗ )
λ ≤ pl , for all l = 1, ..., L with xl∗ = 0
∂xl

Once again, the variable λ is the Lagrange multiplier.

22 / 33
Relation between Hicksian and Walrasian demand

⇒ Once again, UMP-EMP duality implies that the Hicksian and

Walrasian demands are intimately related:

h(p, u) = x(p, e(p, u)), x(p, w ) = h(p, v (p, w )).

The first relation explains why the Hicksian is also known as the
compensated demand correspondence

As prices vary, h(p, u) is the demand that would arise if the

consumer’s wealth were adjusted to keep utility at u. This is
known as Hicksian wealth compensation.

Note that, as prices vary, h(p, u) keeps utility fixed and lets wealth
vary vs. x(p, w ) that keeps wealth fixed but allows utility to vary.

23 / 33
Compensated law of demand

Compensated Law of Demand

Suppose u is a continuous utility function representing locally non-
satiated preferences  on X = RL+ and that h(p, u) consists of a
single element for all p  0. Then h(p, u) is such that for all p 0 , p 00 ,

(p 00 − p 0 ) · [h(p 00 , u) − h(p 0 , u)] ≤ 0

Proof. By the definition of the EMP we have

p 00 · h(p 00 , u) ≤ p 00 · h(p 0 , u)
p 0 · h(p 00 , u) ≥ p 0 · h(p 0 , u).
Subtracting these two inequalities yields the result.

Roughly speaking, keeping the desired utility level fixed, Hicksian

demand and prices move in opposite directions.
24 / 33
Example of Hicksian demand: Cobb-Douglas utility
Recall the Cobb-Douglas utility function u(x1 , x2 ) = x1α x21−α for some
α ∈ (0, 1).

1. Applying the KKT conditions to the EMP (note that all feasible solutions
to the EMP must satisfy x  0) yields the following:
λ αx1α−1 x21−α = p1
λ (1 − α)x1α x2−α = p2
 −α
α 1−α x2
x1 x2 =u⇒ x2 = u.
x1

2. Dividing the first two equalities and raising to the −α power yields
 −α  −α  −α
p1 α x2
= .
p2 1−α x1
Now, use the remaining KKT condition to simplify:
 −α  −α  α
p1 α u (1 − α)p1
= ⇒ h2 (p, u) = u .
p2 1−α x2 αp2

25 / 33
Example: Cobb-Douglas utility

p1 α x2
3. Substituting x2 = h2 (p, u) into p2 = 1−α x1 and solving for x1 yields:
 1−α
αp2
h1 (p, u) = u .
(1 − α)p1

4. Finally, we may use the Hicksian demands to obtain the expenditure

function
(1 − α)α−1 α 1−α
 
e(p, u) = p · h(p, u) = p1 p2 u.
αα

26 / 33
Relationship between Hicksian and expenditure functions
[Note: For the next three results, strictly convex preferences are
imposed to ensure that the UMP and EMP always have unique
solutions.]

Proposition
Suppose u is a continuous utility function representing locally nonsa-
tiated and strictly convex preferences  on X = RL+ . For all (p, u)
the Hicksian demand function h(p, u) is the derivative vector of the
expenditure function with respect to prices:

h(p, u) = ∇p e(p, u).

Many ways to prove this result. E.g., we may use the Envelope
theorem.

27 / 33
Envelope Theorem
Consider the following optimization problem, parameterized by
q ∈ RS :
max f (x; q)
x∈RL
s.t. gi (x, q) = bi , i = 1, 2, ..., m. (1)

Envelope theorem
Suppose φ(q) and x(q) are the optimal value function and maximizer
of problem (1) as a function of q, respectively. Assume φ is differen-
tiable and suppose λ1 , ..., λm are the Lagrange multipliers associated
with the constraints. Then, for any q̄ ∈ RS ,
M
∂φ ∂f (x(q̄); q̄) X ∂gm (x(q̄); q̄)
(q̄) = − λm , s = 1, 2..., S.
∂qs ∂qs m=1
∂qs

28 / 33
Application of Envelope Theorem
Proof. The EMP is equivalent to
max −p · x
x≥0
s.t. u(x) = u.
[Recall that the constraint binds at optimality because of local
nonsatiation].

The optimal value function and maximizer of the above problem are
by definition −e(p, u) and h(p, u). In the envelope theorem
notation: q̄ = (p, u), φ(q̄) = −e(p, u), x(q̄) = h(p, u), f (x; q̄) =
−p · x, g (x; q̄) = u(x) − u.

⇒ The envelope theorem implies that at any p̄ we will have

∂e(p̄, u)
− = −hl (p̄, u) − λl · 0
∂pl
∂e(p̄, u)
⇒ = hl (p̄, u), l = 1, ..., L.
∂pl

29 / 33
Slutsky equation

While the Hicksian is not directly observable (u is one of its

arguments), we show that its price derivative is.

Slutsky equation
Suppose u is a continuous utility function representing locally nonsa-
tiated and strictly convex preferences  on X = RL+ . Then for all
(p, w ) and u = v (p, w ) we have

∂hl (p, u) ∂xl (p, w ) ∂xl (p, w )

= + xk (p, w ), for all l, k
∂pk ∂pk ∂w

or equivalently

Dp h(p, u) = Dp x(p, w ) + Dw x(p, w )x(p, w ).

30 / 33
Roy’s Identity

Can we obtain an analogous result to h(p, u) = ∇p e(p, u) for the

Walrasian demand?

Roy’s Identity
Suppose u is a continuous utility function representing locally nonsa-
tiated and strictly convex preferences  on X = RL+ . Suppose also
that the indirect utility function is differentiable at (p̄, w̄ )  0. Then
∂v (p̄,w̄ )
∂pl
xl (p̄, w̄ ) = − ∂v (p̄,w̄ )
, for all l = 1, .., L.
∂w

31 / 33
Proof of Roy’s Identity

Proof. Once again, we can use the Envelope Theorem, this time applied to
the UMP.

Here the value function and maximizers are v (p, w ) and x(p, w ),
respectively. The envelope theorem implies:

∂v (p̄, w̄ )
= 0 − λxl (p̄, w̄ ), l = 1, ..., L
∂pl
and
∂v (p̄, w̄ )
= λ.
∂w
Combining these two facts yields the result.

Roy’s Identity tells us that, under certain assumptions, it is easier to

compute Walrasian demand from indirect rather than from direct
utility.

32 / 33
Recap

Figure 3.G.3 in MWG

33 / 33