Lecture 2

0.1 Classical Consumer Theory

1. There are n types of goods in the world. We use x = (x1 ; x2; :::::; xn ) to
denote a consumption bundle, where xi represents the quantity of good i
in bundle x. We assume that the goods are perfectly divisible. So xi can
be any non-negative real number. (Note that in economics “good” may
have a temporal dimension. For example, an orange today and an orange
tomorrow are two di¤erent goods.)
2. The price of each good i is constant and denoted by pi . A consumer has
a …xed income I. Let p = (p1 ; :::; pn ) 0 be a price vector. The budget
set is
Xn
p i xi I
i=1
The equation means that the total expenditure must be no greater than
the income.
3. Typically, we will assume that u is twice di¤erentiable. Recall that u is
di¤erentiable at x if it can approximated by a linear function at x. That
is, if there is a linear functional ru (x) such that
u (x + h) u (x) ru (x) h
lim = 0:
khk!0 khk
We call
@u @u
ru (x) ; ::::;
@x1 @xn
the gradient of u at x. The function u is twice di¤erentiable if ru is
di¤erentiable (or equivalently, if we can approximate u by a quadractic
function).

0.2 Consumer Optimization

1. The general consumer optimization problem:
max u (x)
s:t: p:x y;
x 0;
where x 2 <n+ represents a consumption bundle,and y the income of the
consumer.
2. Since the budget set is compact (closed and bounded) and u is continuous,
there exists a solution to this problem. We call the solution to this problem
x (p; y) (x1 (p; y) ; :::; xn (p; y))

1
 the Marshallian demand function. Also, read Rubinstein pp. 53.
3. The Marshallian demand of good i, xi (p; y), is the quantity of good i the
consumer would consume given income and prices. Note that the demand
of a good depends on all prices. If we plot x1 (p; y) against pi , holding y
and all prices other than pi constant, we get the demand curve of good i.
A change in y or some pj , j 6= i, would be represented by a shift of the
demand curve.
4. Assume that the utility function is continuously di¤erentiable. Let
L (x; ) u (x) + (y px)
be the Lagrangian function. A consumption bundle x 0 satis…es the
…rst-order condition of optimization if there exists such that
@u(x )
@xi
= 8xi > 0
pi
and
@u(x )
@xi
8xi = 0:
pi
The parameter measure the marginal utility of money. It is the extra
utility an extra unit of income can buy.
5. Assuming di¤erentiability brings two advantages: 1. It allows to make
precise statements about small changes. and 2. It allows us to apply stan-
dard comparative techniques to examine how the optimal consumption
bundle and welfare will change with prices and income.
6. The …rst-order conditions are necessary conditions for optimality. They
are also su¢ cient if the preferences are convex.

0.2.1 Convex Preferences

1. Recall from intermediate micro that the marginal rate of substitution be-
tween good 1 and good 2 is the amount of good 2 a consumer is willing to
give up for one extra unit of good 1. Mathematically:
dx2
M RS jU =U :
dx1
If MRS equals 2, then you are willing to pay 2 units of y for 1 units of x.
Mathematically, the indi¤erence curve is represented by a function y (x)
such that U (x; y(x)) = C, where C is the level of utility. Thus
@U @U dx2
+ =0
@x1 @x2 dx1
@U
dx2 @x1
, = @U
dx1 @x2

2
 A standard assumption is that dM RS=dx1 is decreasing— as the consumer
consumes more of good 1, he is willing to pay for good 1 (in terms of good
2) decreases. Another way to put it is that the consumer prefers balanced
consumption bundles (ie. some of everything).
2. Convex preferences is the extension of DMRS to the n goods case.
3. De…nition. A preference relation is convex if for any x and x0 , x x0
0 0 0
implies tx + (1 t ) x ( ) x all t 2 [0; 1].
4. Convexity means that a consumer would weakly prefer any consumption
bundle lying on the line segment connecting any two bundles on the same
indi¤erence curve to either of two bundles.
5. Another way to de…ne convex preferences is to say that the better-than
set is convex. (De…nition: A set is convex if the line segment connecting
any two points in the set is contained by the set.)
6. We can also express the convexity of in terms of the utility function that
represents : u (x) is (strictly) quasiconcave if and only if is (strictly)
convex. A function is quasiconcave if for any x, x0 , and 2 (0; 1)

u ( x + (1 ) x0 ) min [u (x) ; u (x0 )] :

A function is strictly quasiconcave if for any x, x0 , and 2 (0; 1)

u ( x + (1 ) x0 ) > min [u (x) ; u (x0 )] :

The claim follows directly from the de…nition. Roughly speaking, quasi-
concavity means that a utility function cannot go down and back up along
a line segment joining two points. Draw some examples. (Note that
is convex does not imply that u is also convex. The term “convexity”
have di¤erent meanings depending whether it is referred to a preference
relation, a function, or a set.)
7. A real function u is di¤erentiable at x if it can approximated by a linear
function at x; that is, if there is a linear functional

@u @u
ru (x) ; ::::; ;
@x1 @xn
which we call the gradient of u, such that
u (x + h) u (x) ru (x) h
lim = 0:
khk!0 khk

The above eqn means that the di¤erence between u (x + h) u (x) and
the linear function ru (x) h decreases of an order greater than khk. The
gradient gives the marginal utility of each good. The gradient therefore
tells us the marginal rate of substitution between goods. (When we say the

3
 marginal rate of substitution is 1 banana to 1 orange, its means that we
are always willing to exchange x banana for tx orange for any t > 1 when
x is su¢ ciently small.) In physical sciences, we often interpret ru (x) is
the direction of greatest accent, but this interpretation is not very useful
in economics (because Euclidean distance often doesn’t have an economic
meaning).
8. A function u is twice di¤erentiable if ru is di¤erentiable (or equivalently
if u can be approximated by a quadractic function).
9. Suppose u is di¤erentiable. Then u is quasiconcave if and only if u (y)
u (x) implies that
ru (x) (y x) 0; (1)
This condition means that the the better-than side lies on upper-right
side of the tangent to u at x. (Economically, this means that any trade
can only be favorable if it is favorable according to the marginal rate of
substitution.)
10. Proof: Suppose u (y) u (x). Then, for all t 2 [0; 1]

u (x + t (y x)) u (x)
u (x + t (y x)) u (x)
) lim 0
tky xk!0 t ky xk
) ru (x) (y x) 0:

The last inequality follows from the di¤erentiability of u at x. (More

exactly, since u (x + t (y x)) u (x) 0, if limkhk!0 u(x+h) u(x)
khk
ru(x)h

will go to positive in…nity if ru (x) (y x) < 0.

11. Now, suppose for any x

ru (x) (y x) 0; (2)

whenever
u (y) u (x) :
If the proposition were wrong, there would exist x, y;and 2 [0; 1] such
that
u (y) u (x) > u ( y + (1 ) x)
and
ru ( y + (1 ) x) 6= 0:
The second condition follows as u (x) is strictly greater than u ( y + (1 ) x).
(Recall that f 0 (x) = 0 for all x 2 [x0 ; x00 ], then by the fundamental the-
orem of calculus, f (x0 ) = f (x00 ) :) By continuity, the second (strict) in-
equality implies that

u (x + h) > u ( y + (1 ) x)

4
 for any h whose norm khk is su¢ ciently small. This implies that

ru ( y + (1 ) x) (x ( y + (1 ) x))
= ru ( y + (1 ) x) (x y) > 0:

But u (y) u (x) requires that

ru ( y + (1 ) x) (y ( y + (1 ) x))
= (1 ) ru ( y + (1 ) x) (y x) 0:

12. If u is strictly increasing and with non-vanishing gradients (i.e. ru (x) 6=

0), we can replace the weak inequality sign in (1) with the strict inequality
sign for all y with u (y) > u (x). That is, x is a solution to the problem

max u (y)
y

s:t: ru (x) y ru (x) x:

Proof: if there exists some y such that u (y) > u (x) and that ru (x) y
ru (x) x. By monotonicity, for some small "

u (y ":e) > u (x) ;

ru (x) (y ":e) < ru (x) x;

which contradicts the fact that u is quasi-concave.

13. A su¢ cient condition for constrained maximization: When u in the max-
imization problem (??) is quasiconcave, strictly increasing, then the …rst-
order condition (??) is su¢ cient for optimality if ru (x ) 6= 0.
14. Proof: (Also see read Simon and Blume pp.533 for the same proof with
di¤erent notations.) Consider some x that satis…es the …rst order condi-
tions. From above we know that if a function u is strictly increasing and
quasi-concave, then for any x, if u (x0 ) > u (x) then

ru (x) (x0 x) > 0:

In words, this means that any x0 in the strictly-better-set lies strictly above
the tangent plane. Thus, x is a solution to the problem

max u (x)
x
s:t: ru (x ) x ru (x ) x :

Consider some x that satis…es the …rst order conditions. We know that
@u(x )
@xi :p for some positive for all i with the equality sign holds
when xi > 0. This immediately implies that

ru (x ) x = px = y

5
 and that for all x0
ru (x ) x0 px0 :
Thus, if x belongs to the budget set px y, it must also belong to the
constraint set ru (x ) x ru (x ) x . Thus, the quasiconcavity of u
implies that any x that satis…es the …rst-order condition must be a global
max to the utility maximization problem. If, in addition, u is strictly
quasiconcave, then x is unique.

0.2.2 Properties of Marshallian Demand

1. Here are some basic properties of Marshallian demand functions:

(a) x (p; y) is homogeneous to degree 0 in p and y. That is, for any t > 0;
x (p; y) = x (tp; ty) :
(b) If u is strictly increasing, then p:x (p; y) = y
(c) If u is strictly quasiconcave, then x (p; y) is unique.
(d) If x (p; y) is unique and the underlying preference relation is contin-
uous, then x (p; y) is continuous in p (and y).

2. Proof:

(a) Multiplying both prices and income by the same factor leaves the
budget set unchanged.
(b) If u is increasing, the consumer can always increase his utility be
consuming more of some goods.
(c) Suppose both x1 and x2 , x1 6= x2 , both solve the consumer optimiza-
tion problem. This necessarily means that u x1 = u x2 = u :
Since u is strictly quasi-concave, it follows that

u x1 + (1 ) x2 > u ;

a contradiction. Hence, the consumer optimization problem has a

unique solution.
(d) Read Rubinstein pp.57-58 or the appendix of Chapter 3 in Mas-
Colell for a proof. Basic Idea: Suppose not. Then there must be
some sequence pn ; y n with lim pn = p and lim y n = y such that
lim x (pn ; y n ) 6= x (p ; y ). Denote lim x (pn ; y n ) by x . Note that
px = y . By defn u (x (p ; y )) > u (x ). Hence, there must exist
some open neighborhood of u (x (p ; y )) (A) and some open neigh-
borhood of u (x ) (B) such that every member of A is strictly prefer-
able to B. But if that is the case, then there must be some member
of B that is feasible under pn ; y n and strictly preferable to x (pn ; y n )
for some n.

6
3. We often want the Marshallian demand to be di¤erentiable. A su¢ cient
condition for di¤erentiability is that the utility function be twice di¤er-
entiable and the determinant of the bordered Hessian of u (x) at x be
non-zero.
4. Defn: Indirect utility function:

v (p; y) u (x (p; y)) :

The indirect utility function is the maximum utility a consumer can achieve
given income and prices.
5. Note that the indirect utility function assigns to each budget set a number
equal to the utility of the optimal bundle in the set. A rational consumer
prefers budget sets with higher indirect utilities to ones with lower. A re-
cent literature has explored richer preferences over choice sets. See Kreps
(1989) “A representation theorem for “preference for ‡exibility,” Econo-
metrica.

6. Thm 1.6. If u (x) is continuous and strictly increasing on <n+ , then the
indirect utility function v (p; y) is

(a) continuous on <n++ <+ ,

(b) homogeneous of degree zero in (p; y) ;
(c) strictly increasing in y,
(d) decreasing in p, and
(e) quasiconvex in (p; y). (A function f is quasiconvex if f is quasicon-
cave.) This property means that if a consumer is indi¤erent between
p1 ; y 1 and p2 ; y 2 , then she must strictly prefer either one to a
linear combination of the two.

7. Proof:

(a) Skip. This is obviously true when demand function is continuous.

But it is true even when it is not.
(b) Multiplying both the prices and income by the same factor leaves
the budget set unchanged. Hence, the maximum utility achievable is
unchanged.
(c) Let y 0 > y and e (1; :::; 1)

y0 y
v (p; y) u (d (p; y)) < u x (p; y) + Pn e (u is strictly increasing)
i=1 pi
< v (p; y 0 ) by de…nition of v:

7
 (d) Suppose p0 p1 . Then x0 , the optimal consumption bundle at p0 ,
1
is feasible at p . Hence,

v p1 ; y = max u (x) s:t: px y

x
0
u x
= v p0 ; y :

(e) We need to show that for all 2 [0; 1] and for all (p; y), (p0 y 0 ) 2 <n+1
++ ,

v ( p + (1 ) p0 ; y + (1 ) y0 ) max fv (p; y) ; v (p0 y 0 )g :

Let

B1 fx : px yg
0
B2 fx : p x y0 g
B3 fx : ( p + (1 ) p0 ) x y + (1 ) y0 g

Note that if px > y and p0 x > y 0 , then

p + (1 ) p0 x > y + (1 ) y0 :

Thus, if x 2 = B1 [ B2 , then x 2 = B3 . The optimal bundle under (p; y) +

(1 ) (p0 ; y 0 ) must be feasible under either (p; y) or (p0 ; y 0 ).

0.3 Expenditure Minimization

1. The expenditure minimization problem:

min p:x
x
s:t: u (x) u;
x 0:

The solution to this problem tells us the minimum expenditure that is

necessary such that the consumer’s utility is not lower than u.
2. Solving this problem tells the government the welfare payment necessary
to keep a person’s utility above a minimum level.
3. De…nitions: A solution to the expenditure minimization problem is known
as the Hicksian demand function and is denoted by

xh (p; u) :

Substituting xh (p; u) into the objective function gives us the expenditure

function
e (p; u) p:xh (p; u) :

8
4. Properties of Hicksian Demand (Mas-Colell Prop 3.E.3): Suppose u is
continuous and strictly increasing, then xh has the following properties:

(a) Homogeneity of degree zero in p. That is, xh (tp; u) = xh (p; u) for

all scalar t > 0.
(b) No excess utility: u xh (p; u) = u.
(c) Convexity/Uniqueness: If u is strictly convex, then xh is unique.

These properties are similar to those of the Marshallian demand. (Prop

3.D.2). Brie‡y, 1 is obvious from the defn. of the minimization problem;
2 follows from strict monotonicity. The argument for 3 is the similar as
the one given in Prop 3.D.2.
5. Theorem (Mas-Colell 3.E.2): Suppose u is strictly increasing and xh (p; u)
is unique for all p. Then for any p0 and p00 ;

(p00 p0 ) : xh (p00 ; u) xh (p0 ; u) 0:

6. The proof is extremely easy and elegant. By part 2 of Prop 3.E.3,

u xh (p00 ; u) = u xh (p0 ; u) :

By de…nition, xh (p00 ; u) and xh (p0 ; u) are the cheapest way to achieve u

under p00 and p0 , respectively. Hence

p00 :xh (p00 ; u) p00 :xh (p0 ; u) ;

p0 :xh (p0 ; u) p0 :xh (p00 ; u) :

Note that this result follows entirely from the fact that xh minimizes
expenditure. It does not use any other properties of the utility functions.
7. This result is called the compensated law of demand. It states that an
expenditure-minimizing consumer should substitute low-price goods for
high-price goods. Or, more speci…cally, the change in consumption bundle
must cost less under the new prices than under the old. Note that if p00
and p0 di¤er only in the price of good i, then the result becomes

(p00i p0i ) xhi (p00 ; u) xhi (p0 ; u) 0;

which means that the Hicksian demand for good i is downward sloping.
8. Properties of the Expenditure Function (Thm 1.7.) If u (:) is continu-
ous and strictly increasing, then the expenditure function e (p; u) has the
following properties:

(a) Zero when u takes on the lowest level of utility (i.e u (0) when the
utility function is increasing).
(b) Continuous on its domain <n++ U.

9
 (c) For all p >> 0; strictly increasing and unbounded above in u.
(d) Increasing in p.
(e) Homogeneous of degree 1 in p.
(f) Concave in p:

9. Proof:

(a) The consumer can achieve u (0) by not spending anything. Since the
minimum utility is be de…nition greater than u (0), the minimum ex-
penditure that is necessary to achieve a utility greater than the lowest
level is not greater than zero. Since prices are non-negative, the total
expenditure is non-negative. Thus, the minimum expenditure is zero.
(b) Intuitively, the result follows from the continuity of u and the conti-
nuity of the budget set.
(c) This is straightforward. Suppose e is not strictly increasing in u.
Then there exist p, u0 and u1 , u1 > u0 , such that e (p; u1 ) e (p; u0 ).
Since u is continuous and strictly increasing in x, it is possible to
reduce consumption by some small " such that

u (h (p; u1 ) ":I) > u0

and
p: (h (p; u1 ) ":I) < e (p; u0 ) :
This violates the de…nition of e (p; u0 ). Suppose by way of contradic-
tion that e is bounded by some e (meaning that e (p; y) e for all
p and y). Since e is strictly increasing in u, e (p; v (p; e ) + 1) must
be greater than e , a contradiction.
(d) If p1 p0 (that is, p1i p0i for all i), then any consumption bundle x
that is feasible under p1 is also under p0 . Since e (p; u) is by de…nition
the lowest expenditure that is necessary to achieve u,

e p0 ; u p0 h p1 ; u p1 h p1 ; u e p1 ; u :

(e) This immediately follows from the fact the Hicksian demand is ho-
mogeneous in degree zero in p.
(f) Let x1 = xh p1 ; u and x2 = xh p2 ; u . By de…nition, for any x such
that u (x) u
e p1 ; u = p1 x1 p1 x
and
e p2 ; u = p2 x2 p2 x:
It follows that for any 2 [0; 1] and for any x such that u (x) u

e p1 ; u + (1 ) e p2 ; u p1 + (1 ) p2 x:

10
 The above equation holds for all x such that u (x) u, including
xh p1 + (1 ) p2 ; u . Hence,

e p1 + (1 ) p2 ; u p1 + (1 ) p 2 xh p1 + (1 ) p2 ; u
e p1 ; u + (1 ) e p2 ; u :
It is important to note that the concavity of e follows from the de…-
nition of e and does not require that u be quasiconcave.

10. Di¤erentiable Utility. Solving a minimization problem involves the same

technique as solving a maximization problem. However, in this problem
the last equation comes from the utility constraint, whereas in the max-
imization problem it comes from the budget constraint. The Lagrangian
function is
L = p: x + (u u(x)):
The …rst order conditions are (assuming x >> 0):
@L @u
= pi = 0 for all i = 1; :::; n;
@xi @xi
@L
= u u(x) = 0:
@
The …rst n equations imply
@u @u
@x1 @xn
= :::: = = :
p1 pn

0.4 Relationship between the Value Function and Demand

Function.
1. The Envelope Theorem: After we solve an optimization problem, we can
substitute the solution of the problem into the objective function to ob-
tain the maximum-value function. If, in addition, the objective function,
constraints and the maximizer are all di¤erentiable with respect to the
parameters, then we can apply the envelope theorem to analyze the max-
imum value function.
2. Unconstrained Optimization Let x (a) be the solution of the maxi-
mization problem
max f (x; a) ;
x
where a is a vector of exogenous parameters. (Let’s assume that the
solution is unique.) The maximum value function is de…ned as
M (a) f (x (a) ; a) :
The Envelope Theorem: Assume that f and x (a) are di¤erentiable. Then
dM (a) @f
= :
daj @aj

11
 A change in aj a¤ect M directly as well as indirectly through x (a). The
envelope theorem tells us that we can ignore the indirect e¤ect.
3. Proof: Partially di¤erentiating M (a) with respect to aj , we have
n
dM (a) X @f @xi @f
= + :
daj i=1
@xi @aj @aj

But the …rst-order condition implies that for all i

@f (x )
= 0:
@xi
Since f is ‡at with respect to x at the optimal point x , a small change
in x has no welfare consequence. The envelope theorem is very useful
in economics. It allows us to evaluate the change in M without knowing
@xi
@aj , the sign of which are di¢ cult to ascertain in many situations. The
theorem also convey an extremely important economic intuition: A …rst-
order deviation from x results only in a is that a second-order loss in
M . For example, since the market output in equilibrium is optimal, the
distortion is proportional to the square of the tax rate.
4. Constrained Optimization The same idea applies can be applied to
the maximum-value function in constrained optimization. Let x (a) be
the solution of the maximization problem

max f (x; a)
x
s:t: g (x; a) = 0:

Here, the parameter a may appear in the constraints as well as in the

objective function.
5. Theorem: Suppose f , g, and x are di¤erentiable with respect to a, then
dM (a) @L
= :
daj @aj

It is important to note that here dM (a)

daj is equal to the partial derivative
with respect to L, the Lagrangian function, not f .
6. The proof is similar to the unconstrained case. See Reny pp. 507 for a
proof.
7. Example

max x1 x2
s:t: a 2x1 4x2 = 0:

Ans: x1 (a) = a=4, x2 (a) = a=8 and (a) = a=16.

12
 8. Roy’s identity: If v (p; y) is di¤erentiable at p0 ; y 0 and @v p0 ; y 0 =@y 6=
0, then
@v p0 ; y 0 =@pi
xi p 0 ; y 0 = ; i = 1; :::; n:
@v (p0 ; y 0 ) =@y
Using the Lagrangian function, we can write

v (p; y) = u (x (p; y)) + (p; y) (y px) :

Now applying the envelope theorem, we have

@v
= xi
@pi
and
@v
= :
@y
Hence, the result.
9. Shephard’s lemma: If e (p; u) is di¤erentiable in p at p0 ; u0 with p0 >> 0,
then
@e p0 ; u0
= xhi p0 ; u0 ; i = 1; :::; n:
@pi
The results follows directly from the envelope theorem.
@e @L
= = xhi .
@pi @pi
The Shephard’s lemma is the counterpart of Roy’s Identity. Same Intu-
ition.
10. Alternatively, the lemma follows from the fact that the function

f (p0 ) p0 xhi (p; u) E (p0 ; u) 0

with the equality sign holding when p0 = p.

11. Note that the lemma only applies when e is di¤erentiable. For example:
u (x; y) = x + y. E (px ; py ; 1) = min fpx ; py g. (0:5; 0:5) is a cost mini-
mization bundle when px = py . But an increase in px will not change
E.

13