Lecture I

Preferences, Utility, Demand

Federico Trionfetti

M1 2024-2025
Micro I
Table of content

1 The research question for this 6 Utility maximization

lecture 7 Indirect utility
2 Preferences 8 Expenditure minimization
Rationality Hicksian demand functions
Existence of a utility function Properties of the expenditure
Other properties function
Three key concepts 9 Exp. Min. ô to U. Max.
3 Three famous utility functions 10 Some identities
Cobb Douglas 11 Money metric utility functions
Leontief 12 Welfare evaluation of price
C.E.S changes
4 Marginal rate and elasticity of
13 Slutsky equation
substitution.
5 Correspondence between CES, 14 Discrete Choice
CD, and Leontief

Federico Trionfetti Lecture I M1 2024-2025 Micro I 2 / 107

 The research question for this lecture

The research question for this lecture

The objective is to build a theory of demand, that is, a relationship

between tastes (or preferences), prices, and income so as to explain con-
sumers’ choices.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 3 / 107

 Preferences

Preferences (V 7.1, MCWG )1

Agents have ‘preferences’ in the sense that they are able to choose be-
tween alternatives. The alternatives consists typically in consumption
bundles (denoted x, y, z, etc.). We use the following notation to express
preference relations.
Preference relations:
x ľ y : means x is no worse than (or at least as good as) y.
x ą y : means x is preferred to y.
x „ y : means x is indifferent to y
We assume preferences to be rational. This is the minimum require-
ment. We also assume a number of additional properties that are very
convenient for analysis of agents decision. We analyze these properties
beginning with rationality.
1
Throughout the course I will use V and MCWG to refer, respectively, to the
textbook by Varian and the textbook by Mass-Collel, Whinston, and Green.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 4 / 107
 Preferences Rationality

Rationality
Definition 1
Preferences are rational if they possess the following properties:
1 Completeness. For all x and y in X, either x ľ y or x ĺ y or
x „ y.
2 Transitivity. For all x, y, and z in X, if x ľ y and y ľ z, then
x ľ z.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 5 / 107

 Preferences Rationality

Rationality has implications on ‘reflectivity’. If preferences are rational

then:
1 the relationship ‘ą’ is not reflexive (x ą x never holds) and it is
transitive (if x ą y and y ą z, then x ą z )
2 the relationship ‘„’ is reflexive (x „ x), transitive (if x „ y and
y „ z, then x „ z), and symmetric (if x „ y then y „ x).
3 if x ą y ľ z, then x ą z
The assumption of rationality and its implications are often violated in
real situations. It is indeed extremely difficult to evaluate alternatives
that are far from common experience. Certainly these assumption should
not be taken literally but rather can be taken as a rule of thumb that
guides most consumption decisions.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 6 / 107

 Preferences Rationality

Example of violations are numerous. Here are some examples.

1 Just perceptible differences. When alternatives are very close to
each other, often the transitivity is violated. So we have x „ y and
y „ z but x ą z. (example of the gradually darkening gray paint).
2 The framing problem. Identical alternatives may be perceived
differently depending on the way they are framed. Example
3 Condorcet paradox for collective decision. This violation of
transitivity occurs when a group (a family, a community, etc.) has
to take a collective decision by majority voting. Example

Federico Trionfetti Lecture I M1 2024-2025 Micro I 7 / 107

 Preferences Existence of a utility function

Existence of a utility function.

In practice, we shall consider quantitative consumption decisions (how
much to buy of an item). Then it is convenient to consider preferences
defined over real vectors located in a real set X “ Rk` . The set X takes
the name of consumption set.
Definition 2
A function u : X Ñ R is a utility function representing preference
relations if, for all x, y P X,

x ľ y ô upxq ě upyq (1)

Note that any strictly increasing transformation of upxq is itself a utility

function. That is, if f is a strictly increasing function, then f pupxqq is a
utility function. Such transformation is often called a positive monotonic
transformation.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 8 / 107

 Preferences Existence of a utility function

Ordinality and cardinality. The intensity of the relationship between

consumption and the utility is subjective. Thus, two individuals may
have the same preferences relations and yet derive different utilities from
the same bundle. Properties of the utility function that are invariant to
strictly increasing transformation are called ordinal. Properties that are
not preserved by a strictly increasing transformation are said cardinal.
The preference relationship represented by a utility function is an ordinal
property, the value of utility associated with any consumption vector is
a cardinal property.
Preferences are ordinal whereas utility is cardinal.
Proposition 1
A preference relationship can be represented by a utility function if and
only if it is rational.

Proof.
Skipped

Federico Trionfetti Lecture I M1 2024-2025 Micro I 9 / 107

 Preferences Other properties

Other properties.2
A number of very convenient assumptions are adopted in applications of
microeconomics. Here is a list:
Continuity. For all y in X the vectors tx : x ľ yu and
tx : x ĺ yu are closed sets. Therefore the sets tx : x ą yu and
tx : x ă yu are open sets. It says that the bundle no worse than y
and those no better than y are separated by only one bundle, y
itself. Hence, continuity of preferences.
Weak Monotonicity. x ě y implies x ľ y. At least as much of
everything is at least as good.
Monotonicity. x " y implies x ą y. More of everything is better.
Strong Monotonicity. x ě y ^ x ‰ y implies x ą y. At least as
much of everything and strictly more of at least one thing is better.

2
Since notation varies across textbooks it is useful to define our own. Click here .
Federico Trionfetti Lecture I M1 2024-2025 Micro I 10 / 107
 Preferences Other properties

... Other properties

Local non-satiation. Given any x in X and any  ą 0 then there
is a vector y with }x ´ y} ă  such that y ą x. Note that strong
monotonicity implies local non-satiation.
Convexity. Given x, y, and z in X such that x ľ z and y ľ z,
then it follows that tx ` p1 ´ tqy ľ z for all t P r0, 1s.
Strict convexity. Given x, y, and z in X such that x ľ z and
y ľ z, then it follows that tx ` p1 ´ tqy ą z for all t P p0, 1q.
Convexity (strict or not) is a very widely used assumption in economic
theory and it is necessary in many circumstances to obtain meaningful
results. Fortunately, this assumption has descriptive power since it cap-
tures a taste for diversification. It says that a mix is preferred (strongly
or not) to extremes.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 11 / 107

 Preferences Three key concepts

Three key concepts

Definition 3
Indifference set Ipxq. The i.s. of x is the set of all vectors x
that are indifferent to x. Formally Ipuq “ tx P X : x „ xu.
Upper contour set V puq. The u.c.s of a vector x is the set of all
vectors x that are no worse than x. Formally
V puq “ tx P X : x Á xu.
Marginal Utility. Is the change in utility obtained from an
infinitesimal increase in the consumption of a good:

Bupxq
(2)
Bxi
...

Federico Trionfetti Lecture I M1 2024-2025 Micro I 12 / 107

 Three famous utility functions

Three famous utility functions

Cobb-Douglas. Th Cobb-Douglas utility function is:
ź
upxq “ pxi qαi (3)
řn
where αi ą 0 for all i and i“1 αi “ 1.
Leontief. The Leontief utility function is:

upxq “ minpai xi q (4)

where ai ą 0 for all i.

Constant Elasticity of Substitution. The C.E.S. utility
function is: « ff 1
ÿn ρ

upxq “ ai pxi qρ , ρă1 (5)

i“1

where ai ą 0 for all i.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 13 / 107
 Three famous utility functions Cobb Douglas

Figure 1: Cobb Douglas (pavilion)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 14 / 107

 Three famous utility functions Cobb Douglas

Figure 2: Graphical construction of the upper contour set. (Cobb-Douglas)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 15 / 107

 Three famous utility functions Cobb Douglas

Figure 3: Upper contour set V puq (Cobb-Douglas)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 16 / 107

 Three famous utility functions Cobb Douglas

Figure 4: Indifference curve Ipuq (Cobb-Douglas)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 17 / 107

 Three famous utility functions Leontief

Figure 5: Utility function Leontief (half pyramid)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 18 / 107

 Three famous utility functions Leontief

Figure 6: Graphical construction of V puq and Ipuq (Leontief)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 19 / 107

 Three famous utility functions Leontief

Figure 7: Upper contour set V puq (Leontief)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 20 / 107

 Three famous utility functions Leontief

Figure 8: Indifference curve Ipuq (Leontief)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 21 / 107

 Three famous utility functions C.E.S

Figure 9: C.E.S (Kite)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 22 / 107

 Three famous utility functions C.E.S

Figure 10: Graphical construction of V puq and Ipuq (C.E.S)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 23 / 107

 Three famous utility functions C.E.S

Figure 11: Upper contour set V puq (C.E.S)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 24 / 107

 Three famous utility functions C.E.S

Figure 12: Indifference curve Ipuq (C.E.S)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 25 / 107

 Marginal rate and elasticity of substitution.

Marginal rate and elasticity of substitution.

Marginal rate of substitution (MRS). Is the rate at which one good
must substitute for the other in order to keep utility constant.
dxj
MRS ” , at constant u. (6)
dxi
The MRS obtains from the implicit function theorem. We skip theorem
but we use the operational aspect. For simplicity consider a function of
only two variables. Given the function upxq the total differential is
Bu Bu
du “ dx1 ` dx2 (7)
Bx1 Bx2
When we compute the MRS we want u to remain constant, that means
we want du “ 0. This implies the following relationship:
Bupxq
dx2 Bx1
“ ´ Bupxq (8)
dx1
Bx2
which is the MRS we defined (6).
Federico Trionfetti Lecture I M1 2024-2025 Micro I 26 / 107
 Marginal rate and elasticity of substitution.

Elasticity of substitution. It measures the percentage change in the

consumption ratio px2 {x1 q over the percentage change in the MRS.

dpx2 {x1 q M RS
σ“ (9)
x2 {x1 dM RS
Intuitively, the elasticity of substitution measures the easiness with which
a consumer is willing to substitute one good with another and be equally
happy.
Graphically, the elasticity of substitution measures the curvature of the
indifferent curve.
To compute it, it is convenient to rewrite it as follows:
M RS
x2 {x1
σ“ dM RS
(10)
dpx2 {x1 q

graphical representation ... Ñ

Federico Trionfetti Lecture I M1 2024-2025 Micro I 27 / 107
 Marginal rate and elasticity of substitution.

Graphical representation of the elasticity of substitution:

tanpα1 q ´ tanpαq tanpβq

σ“ (11)
tanpαq tanpβ 1 q ´ tanpβq

Figure 13: Elasticity of Substitution

Federico Trionfetti Lecture I M1 2024-2025 Micro I 28 / 107

 Marginal rate and elasticity of substitution.

Calculations of MRS and Elasticity of Substitution.

1) Consider a Cobb-Doublas utility f pxq “ px1 qα px2 q1´α . Taking the
ratio of marginal utilities we obtain the MRS:
α x2
MRSCD “ ´ (12)
1 ´ α x1
To compute the elasticity of substitution we first compute
dMRSCD α
“´ (13)
dpx2 {x1 q 1´α

Using equations (12) and (13) in (10) we obtain σ “ 1 as follows:

MRS α
x2 {x1 ´ 1´α
σ“ dMRS
“ α “ 1.
dpx2 {x1 q
´ 1´α

Federico Trionfetti Lecture I M1 2024-2025 Micro I 29 / 107

 Marginal rate and elasticity of substitution.

2) Consider a Constant Elasticity of Substitution utility:

1
u “ ra1 xρ1 ` a2 xρ2 s ρ (14)

The MRS is ˆ ˙1´ρ

a1 x2
MRSCES “´ (15)
a2 x1
To compute the elasticity of substitution we first do

a1 x2 ´ρ
ˆ ˙
dMRSCES
“ ´p1 ´ ρq (16)
dpx2 {x1 q a2 x1

and then
´ ¯1´ρ ´ ¯´1
MRS
x1 {x2
´ aa12 x2
x1
x2
x1 1
σ“ “ ¯´ρ “ (17)
dMRS 1´ρ
´
dpx2 {x1 q ´p1 ´ ρq aa12 x2
x1

Federico Trionfetti Lecture I M1 2024-2025 Micro I 30 / 107

 Marginal rate and elasticity of substitution.

3) Consider the Leontief utility function u “ min ta1 x1 , a2 x2 u and as-

sume a1 x1 ą a2 x2 . Then Bu{Bx1 “ 0 and Bu{Bx2 “ a2 . Inserting these
derivatives in equation (7) with du “ 0 gives

0 “ 0 ˆ dx1 ` a2 dx2 (18)

Therefore, the only way that u can remain constant is that dx2 “ 0 for
any dx1 .

Assume now that a1 x1 ă a2 x2 . Then

0 “ a1 dx1 ` 0 ˆ dx2 (19)

which requires dx1 “ 0 for any dx2 .

In either case the MRS and σ are both zero. Try a graphical represen-
tation.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 31 / 107

 Marginal rate and elasticity of substitution.

Indifference curves and elasticity of substitution.

Figure 14: Three CES and one Cobb-Douglas

Federico Trionfetti Lecture I M1 2024-2025 Micro I 32 / 107

 Marginal rate and elasticity of substitution.

Indifference curve of the Leontief.

if u “ a1 x1 ñ x2 ě u{a2 et x1 “ u{a1
u “ minpai xi q ñ (20)
if u “ a2 x2 ñ x1 ě u{a1 et x2 “ u{a2

The indifference curve is given by the set of x2 et x1 described in (20).

It is convenient to rewrite here in a more compact way.
" * " *
u y u u
IpuqLeo “ x2 ě ^ x1 “ Y x2 “ ^ x1 ě (21)
a2 a1 a2 a1

Federico Trionfetti Lecture I M1 2024-2025 Micro I 33 / 107

 Correspondence between CES, CD, and Leontief

Correspondence between CES, CD, and Leontief

Consider a weighted CES
« ff 1
n
ÿ ρ

f pxq “ ai pxi q ρ
, ρă1 (22)
i“1

where ai ą 0. Let’s
ř now impose that the coefficients ai are weights in the
sense that and ni“1 ai “ 1. Then the function (22) is a weighted power
mean. The elementary ‘unweighted’ power mean occurs for ai “ 1{n.

From CES to Cobb-Doublas. We will show that the limit for ρ Ñ 0

of a CES is a Cobb-Douglas. Begin by writing (22) as
« ff
n
1 ÿ
f pxq “ exp ln ai pxi qρ (23)
ρ i“1

Federico Trionfetti Lecture I M1 2024-2025 Micro I 34 / 107

 Correspondence between CES, CD, and Leontief

Now take the limit.

řn
ln ρ
„ 
i“1 ai pxi q
lim f pxq “ exp lim (24)
ρÑ0 ρÑ0 ρ
Note that the argument of the exponential is the limit of a fraction of
two functions of ρ. This allows us to apply the L’Hôpital’s rule that
limzÑz0 pgpxq{hpxqq “ limzÑz0 pg 1 pxq{h1 pxqq. Therefore (recalling that
dz t {dt “ z t ln z) we have:
» řn ρ fi
řn ai pxi q lnρ xi
i“1
i“1 ai pxi q
lim f pxq “ exp – lim fl (25)
ρÑ0 ρÑ0 1
« ff « ff
n
ÿ n
ź
“ exp ai ln xi “ exp ln xai i . (26)
i“1 i“1
n
ź
lim f pxq “ xai i (27)
ρÑ0
i“1

which is a Cobb-Douglas.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 35 / 107
 Correspondence between CES, CD, and Leontief

From CES to Leontief. We will show that the limit for ρ Ñ ´8 of a

CES is a Leontief. Let xk “ mintxi u.
«
n ˆ ˙ρ ff ρ1
ÿ xi
f pxq “ xk ai (28)
i“1
xk

Therefore
«
n ˆ ˙ρ ff ρ1
ÿ xi
lim f pxq “ lim xk ai (29)
ρÑ´8 ρÑ´8
i“1
xk
“ xk r0s0 “ xk (30)

Therefore
lim f pxq “ mintxi u. (31)
ρÑ´8

which is a Leontief.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 36 / 107
 Correspondence between CES, CD, and Leontief

Check Point
End of Week 1.

The student should check she or he is able to easily do following:

Obtain the formula for the MRS from the total differential.
Compute the marginal rate of substitution using the formula of
the ratio of marginal utilities.
Compute the elasticity of substitution
Obtain the indifference curve by inversion of the utility function
Obtain the MRS as the derivative of the indifference curve and
check that it is the same as the MRS obtained in the second bullet
point above.
Of course, all items except the first should be done for the for three
famous utility functions.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 37 / 107

 Utility maximization

Utility maximization (Varian, Sect. 7.2)

Consumers are utility maximizers.

max upxq s.t. px “ m. (32)

x

The solution to this problem is a function vpp, mq known as the indirect

utility function.

Incidentally, notice that the “=” sign implies that we assumed strict
monotonicity.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 38 / 107

 Utility maximization

The Lagrangian of the problem above is:

Lpλ, xq “ upxq ´ λppx ´ mq, (33)

The F.O.C. characterizing an interior solutions (for ‘smooth’ utility func-

tion) are:

BLpλ, xq Bupxq
“ 0 ùñ “ λpi @i (34)
Bxi Bxi
BLpλ, xq
“ 0 ùñ px “ m. (35)
Bλ
The solution of this system is a vector x˚ and a scalar λ˚ . The vector x˚
is known as the vector of demands because each element is the quantity
of a good demanded.
The functions x˚i pp, mq are the demand functions. Sometimes they are
called the ‘Marshallian’ or ‘market’ demand functions.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 39 / 107

 Utility maximization

The F.O.C. for utility maximization have a straightforward interpreta-

tion. Taking the ratio of any two BLp.q{Bxi we obtain

pi Bupxq{Bxi
“ (36)
loop
moj on Bupxq{Bxj
looooomooooon
Economic R.S. ´M RS

Note that the l.h.s. is equal to the ratio Bpx{Bx

Bpx{Bxi
j
and, as such, it may be
called the Economic Rate of Substitution.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 40 / 107

 Utility maximization

A graphical representation of utility maximization.

The optimal consumption vector
is represented by the blue dot. At
the optimal consumption vector,
the slope of the indifference curve
is the same as the slope of the
budget line. This is the graphical
counterpart of the mathematical
condition shown in equation (36)
and pasted here.

pi Bupxq{Bxi
“
loop
moj on Bupxq{Bxj
looooomooooon
Figure 15: Utility maximization. Economic R.S. ´M RS

Federico Trionfetti Lecture I M1 2024-2025 Micro I 41 / 107

 Indirect utility

Indirect utility (Varian, Sect. 7.3)

Plugging x˚ pp, mq into upxq gives a function known as the indirect utility
function, denoted vpp, mq. An example is in Fig. 16 which represents
the indirect utility associated with a Cobb-Douglas utility.

Figure 16: Indirect utility function (Cobb-Douglas example)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 42 / 107

 Indirect utility

Utility and indirect utility

Figure 17: Utility and Indirect Utility (Cobb-Douglas example)

Comment on terminology: upper-contour and lower-contour.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 43 / 107

 Indirect utility

The indirect utility function has the following four properties:

Non-increasing in p and non-decreasing in m. That is:
vpp, mq ě vpp1 , mq for any p1 ě p ^ p1 ‰ p. Analogously for m.

Proof. Let B “ tx : px ď mu and B 1 “ tx : p1 x ď mu. Then

B 1 Ă B.3 But then the maximum utility obtained by choosing in
B must be at least as high as the maximum utility obtained by
choosing in B 1 .
Analogously for m.
Homogeneous of degree 0 in (p, mq.

Proof. The budget set is unaltered by the multiplication of p and

m by any t ą 0.

3
Ă means proper subset.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 44 / 107
 Indirect utility

Quasi-convex in p. Recall that a function is quasi-convex if the

lower contour set is convex. The lower contour set is

tp : vpp, mq ď ku . (37)

Therefore, to prove quasi-convexity we have to show that the lower

contour set is a convex set for any k ą 0.

Proof. skipped.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 45 / 107

 Expenditure minimization

Expenditure minimization (Varian Sect. 7.3)

We can pose the consumer problem as that of minimizing the expenditure
needed to achieve a given level of utility.

min px, s.t. upxq “ u. (38)

x

The Lagrangian of the problem above is:

Lpλ, xq “ px ´ λpupxq ´ uq, (39)

The F.O.C. characterizing an interior solutions are:

BLpλ, xq Bupxq
“ 0 ùñ pi ´ λ “ 0 @i (40)
Bxi Bxi
BLpλ, xq
“ 0 ùñ upxq ´ u “ 0. (41)
Bλ
The solution to this problem is a vector x˚ .
Federico Trionfetti Lecture I M1 2024-2025 Micro I 46 / 107
 Expenditure minimization Hicksian demand functions

Hicksian demand functions. The expenditure-minimizing quantities

x˚ depend on the ‘parameters’ of the problem: p and u. Thus we write
hi pp, uq to represent the demand for good i. These demands take the
name of Hicksian or Compensated demand functions.
Replacing the vector x˚ in place of x in the expression for expenditure px
we obtain the Expenditure Function epp, uq. The expenditure function
represents the least expenditure needed to achieve utility u.

If the expenditure function is already known, then the hicksian demand

function may be derived from the expenditure function as follows:

Bepp, uq
hi pp, uq “ @i. (42)
Bpi

Federico Trionfetti Lecture I M1 2024-2025 Micro I 47 / 107

 Expenditure minimization Properties of the expenditure function

Properties of the expenditure function

Homogeneous of degree 1 in p.
Non-decreasing in p.
Concave in p.
Continuous in p.
We prove the first three properties.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 48 / 107

 Expenditure minimization Properties of the expenditure function

Proposition 2
The expenditure function is homogeneous of degree 1 in p. That is, for
any scalar t we have eptp, uq “ tepp, uq.

Proof.
The constraint of the expenditure minimization problem, upxq “ u, is
insensitive to price changes. Therefore, the solution vector x˚ remains
unchanged when prices change proportionally. But the sum we spent
before the price change is no longer sufficient to afford x. We now need
a sum eptp, uq “ tpx˚ . But the latter is exactly tepp, uq.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 49 / 107

 Expenditure minimization Properties of the expenditure function

Proposition 3
The expenditure function is non-decreasing in any pi . That is, if
p1 ě p ^ p1k “ pk @k ‰ i, then epp1 , uq ě epp, uq.

Proof.
Let x1 be the solution of the expenditure minimization problem for
prices p1 . Then

epp1 , uq “ p1 x1 ě px1 ě epp, uq “ px (43)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 50 / 107

 Expenditure minimization Properties of the expenditure function

Proposition 4

The expenditure function is concave in p. That is, if t P r0, 1s and

p2 “ tp ` p1 ´ tqp1 then epp2 , uq ě tepp, uq ` p1 ´ tqepp1 , uq.

Proof.
Let tx, x1 , x2 u be the expenditure minimizing consumption vectors for,
respectively, tp, p1 , p2 u. By definition we know that

tpx2 ě tpx “ tepp, uq (44)

p1 ´ tqp x 1 2 1 1
ě p1 ´ tqp x “ p1 ´ tqepp , uq. 1
(45)

Adding these two equations and noticing that

epp2 uq “ tpx2 ` p1 ´ tqp1 x2 proves the property.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 51 / 107

 Exp. Min. ô to U. Max.

Exp. Min ô U. Max (Varian, App. to Ch. 7)

You may have noticed that the ratio of any two equations (40) satisfy
condition (36).

Bupxq
eq. (40) for i pi ´ λ “0
Bxi
Bupxq
eq. (40) for j pj ´ λ “0
Bxj

pi Bupxq{Bxi
eq. (36) “
loopj on
mo Bupxq{Bxj
looooomooooon
Economic R.S. ´M RS

Yet condition (36) was obtained from utility maximization while (40)
were obtained from expenditure minimization. This fact raises the sus-
picion that the solution of the U-max and E-min are related. We discuss
these matters in the following slides.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 52 / 107
 Exp. Min. ô to U. Max.

Utility maximization implies expenditure minimization.

Proposition 5

Assume that upxq is continuous and that preferences satisfy local

non-satiation. Let x˚ be the solution of

max upxq s.t. px “ m, (46)

x

and let u˚ “ upx˚ q. Then x˚ is also the solution of the corresponding

expenditure minimization problem, that is:

x˚ “ min px s.t. upxq “ u˚ . (47)

x

Proof in next slide.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 53 / 107

 Exp. Min. ô to U. Max.

Proof.
U-max ñ E-min. We want to show that x˚ solves E-min. Suppose
not and let x1 be the solution to E-min; therefore by the statement of
the E-min problem px1 ă px˚ and upx1 q “ u˚ “ upx˚ q. But continuity
and non-satiation make that there exists a bundle x2 close enough
(slightly preferred) to x1 such that px2 ă px˚ “ m and that
upx2 q ą upx˚ q. But this contradicts the assumption that x˚ is the
solution of U-max.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 54 / 107

 Exp. Min. ô to U. Max.

Expenditure minimization implies utility maximization.

Proposition 6
Assume that upxq is continuous and that preferences satisfy local
non-satiation. Let x˚ be the solution of

min px, s.t. upxq “ u. (48)

x

and let m˚ “ px˚ . Then x˚ is also the solution of the corresponding

utility maximization, that is:

x˚ “ max upxq s.t. px “ m˚ , (49)

x

Proof in next slide.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 55 / 107

 Exp. Min. ô to U. Max.

Proof.
U-max ð E-min. We want to show that x˚ solves U-max. Suppose
not and let x1 be the solution to U-max; therefore by the statement of
the U-max problem upx1 q ą upx˚ q and px1 “ px˚ “ m˚ . (To
understand these two equalities remember that the U-max problem
corresponding to the E-min is one in which the income is equal to the
least expenditure. Therefore, the equality px1 “ m˚ comes from the
hypothesis that x1 is utility maximizing and px˚ “ m˚ comes from the
fact that x˚ is cost minimizing). Since, px˚ ą 0 and utility is
continuous, there exist a t P p0, 1q such that ptx1 ă px˚ “ m˚ and that
uptx1 q ą upx˚ q. But this contradicts the assumption that x˚ is the
solution of E-min.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 56 / 107

 Some identities

Some identities (Varian, Sect. 7.4)

The expenditure function and the indirect utility function are linked.
This can be seen by writing the two problems in a compatible way.

vpp, mq “ max upxq s.t. px “ m. (50)

x

Let x˚ be the solution to this problem and let u˚ “ upx˚ q be the asso-
ciated utility. Now let’s write the E-min problem as

epp, u˚ q “ min px, s.t. upxq “ u˚ . (51)

x

These two problems have the same solution x˚ . A set of identities follows
from this result.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 57 / 107

 Some identities

A first set of identities:

1 e pp, vpp, mqq ” m. The minimum expenditure necessary to reach
utility vpp, mq is m.
2 v pp, epp, uqq ” u. The maximum utility from income epp, uq is u.
3 xi pp, mq ” hi pp, vpp, mqq. The Marshallian demand at income m
is the same as the Hicksian demand at utility vpp, mq.
4 hi pp, uq ” xi pp, epp, uqq. The Hicksian demand at utility u is the
same as the Marshallian demand at income epp, uq.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 58 / 107

 Some identities

Check Point
End of Week 2.

The student should check she or he is able to easily do following:

Understand the graphical correspondence between the first order
conditions and the graphical representation of utility maximization
and expenditure minimization.
Understand the correspondence between U-max and E-min.
Maximize utility and minimize expenditure.
Be able to pass from the expenditure function to the indirect
utility by using the identities studied above.
Be able to pass from the market demand function to the
compensated demand function using the identities studied above.
Be able to compute the perfect price index.
Of course, all items except the first two should be done for the Cobb-
Douglas, CES, and Leontief.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 59 / 107
 Money metric utility functions

Money metric utility functions (Varian, Sect. 7.5)

The problem with the indirect utility is that we are unable to measure
the level of utility, the problem with the expenditure function is that it
depends on utility, which is not observable. Thus, interpersonal com-
parisons are not meaningful because they are based on different units of
measure.

To circumvent this problem we ask consumers how much money is nec-

essary at prices p to be as well off as by consuming the bundle of goods
x. 4 Note that the consumer does not need to consume x at prices p to
be as happy as consuming x.

4
Which corresponds to the popular expression “to put one’s wallet where one’s
mouth is.”
Federico Trionfetti Lecture I M1 2024-2025 Micro I 60 / 107
 Money metric utility functions

So, what is the least sum needed for a happiness equal to upxq at prices
p? Formally, the problem is

min pz, s.t. upzq “ upxq. (52)

z

The solution to this problem is the expenditure function epp, upxqq. This
particular expenditure function is known as the money metric utility
function and it is usually denoted by mpp, xq; i.e.,

mpp, xq ” epp, upxqq (53)

The money metric utility function is defined over observable variables

pp, xq. Note also that mpp, xq is a utility function because it is a mono-
tonic transformation - through p - of the utility function upxq. So,
mpp, xq is at the same time a utility function with respect to x and an
expenditure function with respect to p.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 61 / 107

 Money metric utility functions

Graphical representation mpp, xq :

1 choose an arbitrary x, this choice
identifies an indifference curve (blue).
2 chose an arbitrary price vector p,
this choice identifies the slope of a
budget line (green line).
3 to find mpp, xq shift the green line
(keeping the slope unchanged) un-
til you touch the indifference curve
identified by x. This will give you
z. The sum pz is the expendi-
ture needed to achieve utility upxq
at prices p, that is, mpp, xq. Figure 18: Money metric utility

Federico Trionfetti Lecture I M1 2024-2025 Micro I 62 / 107

 Money metric utility functions

Quick test:
1 Two individuals are asked how much utility they derive from a cer-
tain vector x. One says 15 the other says 9. The same individuals
are asked to tell how much money they need at prices p to be as
happy as by consuming x. Both individuals say they need 3 €. Is
this answer consistent with the previous ones? Is it possible that
they have the same preferences?
2 Two individuals are asked how much utility they derive from a cer-
tain vector x. Both say 10. Do they have the same preferences?
3 Two individuals are asked how much utility they derive from a cer-
tain vector x. Both say 10. Then they are asked how much money
they need at prices p to be as happy as by consuming x. One
says 2 €, the other says 5 €. Are these answers consistent with the
previous ones? Do the two individuals have identical preferences?

Federico Trionfetti Lecture I M1 2024-2025 Micro I 63 / 107

 Money metric utility functions

In association with the money metric utility function we may construct

the money metric indirect utility function by replacing in it the indirect
utility obtained from a price vector q and an income m:

µpp; q, mq ” epp, vpq, mqq (54)

That is, µpp; q, mq measures how much money one would need at prices
p to be as happy as one would be facing prices q and income m.

The money metric utility function is very useful for the welfare evalua-
tions of changing economic conditions (such as a tax, a subsidy, change
in income, change in price, etc.).

Federico Trionfetti Lecture I M1 2024-2025 Micro I 64 / 107

 Money metric utility functions

Graphical representation of µpp; q, mq.

1 choose an arbitrary m and q, this
choice identifies a budget line (red).
2 utility maximization under m and q
will give you x and the correspond-
ing indirect utility.
3 chose another arbitrary price vector
p, this choice identifies the slope of
another budget line (green line).
4 to find µpp; q, mq shift the green line
(keeping the slope unchanged) un-
til you touch the indifference curve
identified by max upxq s.t. m and Figure 19: Money metric
q (blue). This will give you z. The indirect utility
sum pz is the expenditure needed to
achieve the indirect utility vpq, mq at
the new prices p, that is, µpp; q, mq.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 65 / 107
 Welfare evaluation of price changes

Welfare evaluation of price changes5

We use the money metric utility and indirect utility to evaluate the
consequences of price changes on welfare. If we could observe the indirect
utilities than we would just measure the change in indirect utilities. But
we don’t. So, we resort to the money metric functions.

For our purposes we use two tools:

Compensating and equivalent variation
Consumer surplus

5
Varian, Ch. 10; Mass-Colell et al. Sect.3.I (p.80).
Federico Trionfetti Lecture I M1 2024-2025 Micro I 66 / 107
 Welfare evaluation of price changes

Compensating and equivalent variation (Ch. 10, sect.1).

Consider a consumer facing a change in price and income from pq0 , m0 q
to pq1 , m1 q. What is the effect of such change on welfare? The direct
measure of the welfare change is
vpq1 , m1 q ´ vpq0 , m0 q (55)
The problem is that this welfare difference depends on the particular
utility function chosen to represent preferences. As such, it is not good
for interpersonal comparison.
One way to put a unit of measure on it is to use the money metric indirect
utility functions. Recall that µpp; q, mq is defined as epp, vpq, mqq, that
is, the minimal expenditure needed to reach utility level vpq, mqq at
prices p. Then the welfare effects of the change from pq0 , m0 q to pq1 , m1 q
can be measured as
µpp; q1 , m1 q ´ µpp; q0 , m0 q (56)
It remains to choose the reference price-vector q with respect to which
we compute the money metric indirect welfare change.
... Federico Trionfetti Lecture I M1 2024-2025 Micro I 67 / 107
 Welfare evaluation of price changes

... There are two obvious choices, p “ q0 or p “ q1 , and each has a

name: The equivalent variation and the compensating variation.

EV “ µpq0 ; q1 , m1 q ´ µpq0 ; q0 , m0 q “ µpq0 ; q1 , m1 q ´ m0 (57)

CV “ µpq ; q , m q ´ µpq ; q , m q “ m ´ µpq ; q , m q
1 1 1 1 0 0 1 1 0 0
(58)

The EV responds to the question of what income change at current prices

is equivalent to the change from pq0 , m0 q to pq1 , m1 q. For instance, a
price reduction (q1 ď q0 ) or an income increase (m1 ą m0 ) or both
would make (59)>0. In this case the EV measure the (positive) income
increase that is equivalent to the price reduction in the eyes of the
consumer. The EV is an ex-ante situation projected into the future.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 68 / 107

 Welfare evaluation of price changes

EV “ µpq0 ; q1 , m1 q ´ µpq0 ; q0 , m0 q “ µpq0 ; q1 , m1 q ´ m0 (59)

CV “ µpq ; q , m q ´ µpq ; q , m q “ m ´ µpq ; q , m q
1 1 1 1 0 0 1 1 0 0
(60)

The CV responds to the question of what income changes are nec-

essary to compensate the consumer for the change from pq0 , m0 q to
pq1 , m1 q. For instance, a price reduction (q1 ď q0 ) or an income in-
crease (m1 ą m0 ) or both would make (60)>0. In this case the EV
measure the (positive) income decrease that would ‘compensate’ for the
price reduction in the eyes of the consumer. The CV is an ex-post situ-
ation projected into the past.

Note that signpEV q “ signpCV q but the magnitude is not the same.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 69 / 107

 Welfare evaluation of price changes

Figure 20: Equivalent variation of a price decrease

“ m0 (61)
¯¯ “ µpq ; q , m q0 1 1
(62)
Dashed-Green minus Solid-Red = income increase equivalent to the price
reduction.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 70 / 107
 Welfare evaluation of price changes

Figure 21: Compensating variation of a price decrease

¯¯ “ m1 (63)
“ µpq ; q , m q
1 0 0
(64)
Dashed-Red minus Dashed-Green = income to be taken away to com-
pensate for the price decrease.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 71 / 107
 Slutsky equation

Slutsky equation (V. 8.2; MCWG 3.G.3)

We study the effect of a price change on demand.
Proposition 7
The effect of a price change on demand can be decomposed in a
substitution effect and an income effect as follows:

Bxi pp, mq Bhi pp, vpp, mqq Bxi pp, mq

“ ´ xi pp, mq (65)
Bpi Bpi
loooooooomoooooooon Bm
loooooooooomoooooooooon
Substitution ef f ect Income ef f ect

Proof
Let x˚ be the u-max’ing vector at pp, mq and let u˚ “ upx˚ q. Then, by
identity 4 in page 58 we have

hi pp, u˚ q ” xi pp, epp, u˚ qq. (66)

Federico Trionfetti Lecture I M1 2024-2025 Micro I 72 / 107

 Slutsky equation

Proof cont’d
By differentiation of identity (66) we obtain

Bhi pp, u˚ q Bxi pp, mq Bxi pp, mq Bepp, u˚ q

“ ` . (67)
Bpi Bpi Bm Bpi

The l-h-s term is the change in Hicksian demand in response to a

price change.
The first term on the r-h-s is the change in demand at constant m.
The second term on the r-h-s is the change in demand when m
changes, multiplied by the change in expenditure needed to keep
utility at u˚ when the price changes. Note that I am using the fact
that that m “ epp, u˚ q.
The last multiplicand is precisely x˚ . Substituting and rearranging
proofs the proposition.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 73 / 107

 Slutsky equation

Incidentally,

Bxi pp, mq Bhi pp, vpp, mqq Bxi pp, mq

“ ´ xi pp, mq ž 0? (68)
Bpi Bpi
loooooooomoooooooon Bm
loooooooooomoooooooooon
Substitution effect Income effect

To answer the question remember that the expenditure function is con-

cave in prices (see Proposition 4). The concavity implies that Bhpp,uq
Bpi “
B 2 epp,uq
Bp2i
ă 0. So, the substitution effect is certainly negative. The income
effect may be positive or negative depending on the utility function. The
total effect is then positive if the income effect is positive and larger in
magnitude than the substitution effect.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 74 / 107

 Slutsky equation

Graphical representation of the Slutski decomposition

Figure 22: Effects of a price increase: Total, Subst., Income

Federico Trionfetti Lecture I M1 2024-2025 Micro I 75 / 107

 Slutsky equation

Example 7
Study the Cobb-Douglas example in Varian Sect. 8.2

Link to the appendix on Slutsky equation for Cobb-Douglas.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 76 / 107

 Slutsky equation

Check Point
End of Week 3.

The student should check she or he is able to easily do following:

Obtain the money metric utility and money metric utility indirect
utility for each of the the three famous utility functions
Be able to compute the Slutsky decomposition for the three utility
functions.
Find the mathematical expression for the Equivalent Variation
and the Compensating variation for the Cobb-Douglas. Consider
for instance a decline in p1 and compute the CV and EV. You may
do it also for the CES and Leontief if you like.
Make sure you are able to answer the questions on CV and EV
that you will find in past exams (past exams are in Amétice).

Federico Trionfetti Lecture I M1 2024-2025 Micro I 77 / 107

 Discrete Choice

Discrete Choice

We have studied agents choice under the assumption of continuity. A

consumer chose the quantity of a good s/he wishes to consume, this
is a choice among an uncountable number of alternatives (continuity).
Choice, however, is often discrete; think of vacation destination, the city
we want to live in, the number of cars we want to have, the type of car
we want to have, the university where we want to study, the subject we
want to study, etc. Discrete choice models represent agent’s choice from
among a finite number of alternatives. Discrete choice models have a
large number of applications. One interesting feature is that the modeler
(the researcher) does not need to know all the reasons why a particular
choice is made.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 78 / 107

 Discrete Choice

Since this is a theory of choice among alternatives we begin by defining

the choice set.

Choice set (A). The choice set must have three features:
It must contain a finite number of alternatives.
It must be exhaustive.
The alternatives must be mutually exclusive.

In what follows N denotes the number of elements in A (numerosity of

A).

Federico Trionfetti Lecture I M1 2024-2025 Micro I 79 / 107

 Discrete Choice

Utility. The utility that an agent derives from alternative i is

Ui “ ui ` i (69)

where ui and i are, respectively, the deterministic and the stochastic

component of utility. Note the subscript i refers to the alternative, not
to the agent. To keep things simple we assume that agents are identi-
cal. Yet, choices will be heterogeneous across agents because of different
realizations of the stochastic component i . An agent will choose to con-
sume item i over any item j if ui ` i ą uj ` j for all j ‰ i. Thus, the
probability that an agent chooses i is

Pi “ P r rui ` i ą uj ` j @j ‰ is . (70)

We observe that such probability has two interesting properties:

Only difference in utility matters: adding c to all Ui gives the
same Pi .
The scale of utility is irrelevant: multiplying by c all Ui gives the
same Pi .
Federico Trionfetti Lecture I M1 2024-2025 Micro I 80 / 107
 Discrete Choice

We also see that the probability Pi is a cumulative distribution. This is

seen by rewriting it in this way

Pi “ P r rj ´ i ă ui ´ uj @j ‰ is (71)

In words, Pi is equal to the probability that the random variable j ´i is

below the deterministic value ui ´ uj . Then, it is clear that the probabil-
ity depends on the distribution of the stochastic components. Depending
on the assumptions we make on the distribution of the stochastic com-
ponent we have different variants of discrete choice models. We choose
the variant that gives rise to the LOGIT representation of Pi .

Federico Trionfetti Lecture I M1 2024-2025 Micro I 81 / 107

 Discrete Choice

LOGIT. In the LOGIT class of discrete choice model the typical as-
sumption is that the stochastic components are i.i.d. and distributed
as double exponential (Gumbel) or as a Fréchet. We use Gumbel: the
cumulative distribution and density function are, respectively
´ p ´η
µ q
F pq “ e ´e
(72)
1 ´ ´η ´η
´ ¯
´p µ q
f pq “ e µ
e´e (73)
µ
The scale parameter of the distribution is µ while η is the location pa-
rameter (the mode). The mean of the distribution is η ` µγ where γ
is the Euler constant (« 0.58). We set the mean to zero by choosing
η “ ´µγ. We could scale the distribution by setting µ “ 1 but we do
not do it because below we’ll discuss the role of the scale parameter in
the choice probability.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 82 / 107

 Discrete Choice

We can now apply the double exponential to eq. (71). This will give the
LOGIT formula.6 Remember that an agent will choose item i over any
item j if ui ` i ą uj ` j for all j ‰ i. Thus,

Pi “ P r rj ă i ` ui ´ uj @j ‰ is (74)
ż8 N
ź
“ f pi q F pi ` ui ´ uj q di (75)
´8 j‰i
i `ui ´uj
ˆ ˙
N
1
ż8 ´
i
¯ i ´ `γ
´ µ `γ ´e´p µ `γ q ź µ
“ e e ´e
e di (76)
´8 µ j‰i

I rewrite (76) in the next page rearranging terms.

6
Logit is an abbreviation of “logistic probability unit.”
Federico Trionfetti Lecture I M1 2024-2025 Micro I 83 / 107
 Discrete Choice

i `ui ´uj
ˆ ˙
N
p µi `γ q ź 1 ´
ż8 ´ `γ
´
i
¯
´ µ `γ
Pi “ e ´e
e ´e
e µ
di (77)
´8 j‰i
µ
Now do the following: ” ´ ¯ ı
Operate the change of variable δ “ exp ´ µi ` γ
´ ¯
i
´ `γ
Note that dδ “ di “ ´ µ1 δdi ă 0
´ µ1 e µ

Operate the change of variable yi “ exppui {µq and observe that

ep´ui `uj q{µ “ yj {yi
Then we obtain
ż8 N δy
´ j
ź
Pi “ e´δ e yi dδ (78)
0 j‰i

The limits of integration have been inverted once because δ is decreasing

in i (thus inverting the direction of integration) then inverted again
because dδ ă 0 and then cut to 0 because δ P r0, 8s. Indeed lim δ “ 0
i Ñ8
and lim δ “ 8.
i Ñ ´8
Federico Trionfetti Lecture I M1 2024-2025 Micro I 84 / 107
 Discrete Choice

I rewrite (78) here. Now after simple algebraic manipulation we obtain

ż8 N
ź δyj
´
Pi “ e´δ e yi
dδ
0 j‰i

ż8 ˜ řN ¸
´δ j“1 yj
Pi “ exp dδ (79)
0 yi

and finally we obtain the LOGIT formula.

yi
P i “ řN (80)
j“1 yj

Federico Trionfetti Lecture I M1 2024-2025 Micro I 85 / 107

 Discrete Choice

Undoing the change of variables yi we obtain

ui
eµ
Pi “ ř uj (81)
N
j“1 e
µ

We observe that
1
lim Pi “ (82)
µÑ8 N
1 if ui ą max uj , @j ‰ i
"
lim Pi “ (83)
µÑ0 0 otherwise

The first limit above tells us that when the variance is infinite (µ Ñ
8) each alternative gets the same chance. The second limit tells us
that when the distribution is degenerate (µ Ñ 0) only the deterministic
component matters and only one and the same alternative will be chosen
by all agents.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 86 / 107

 Discrete Choice

Lastly, we observe that the relative probability between two alternatives

does not depend on the utility of any other alternative
ˆ ˙
Pi ui ´ uj
“ exp (84)
Pj µ

This property is called Independence from Irrelevant Alternatives (IIA).

How to puzzle your banqueters by violating IIA. The waiter arrives and says: I only have chicken or fish, we
had beef but it is finished. Everybody makes a choice, including you, and for instance you choose chicken.
The waiter goes to the kitchen and comes back to say says: Oh I actually have beef. And you say: “In that
case I will choose fish”.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 87 / 107

 Discrete Choice

The ex-ante indirect utility of the set A is (proof omitted):

˜ ¸
N
ÿ ui
U pAq “ µ ln eµ (85)
i“1

and it has the following properties.

Property 1
˜ ¸
maxtui u N
ÿ ui ´maxtui u
lim U pAq “ lim µ ln e µ e µ (86)
µÑ0 µÑ0
i“1
« ˜ ¸ff
ˆ maxtu u ˙ N ui ´maxtui u
i ÿ
“ lim µ ln e µ ` ln e µ (87)
µÑ0
i“1
˜ ¸
ˆ maxtu u ˙ N ui ´maxtui u
i ÿ
“ lim µ ln e µ ` lim µ ln e µ (88)
µÑ0 µÑ0
i“1

continues ...
Federico Trionfetti Lecture I M1 2024-2025 Micro I 88 / 107
 Discrete Choice

... just rewriting the last line

˜ ¸
ˆ maxtu u ˙ N ui ´maxtui u
i ÿ
“ lim µ ln e µ ` lim µ ln e µ
µÑ0 µÑ0
i“1

The first term is just max tui u ,

˜ ¸
N
ÿ ui ´maxtui u
“ max tui u ` lim µ ln e µ (89)
µÑ0
i“1
maxtui u´maxtui u
One of the terms in pq of (89) is e µ “ 1. Therefore
¨ ˛
Nÿ
´1 ui ´maxtui u
“ max tui u ` lim µ ln ˝1 ` e µ ‚ (90)
µÑ0
i‰maxtui u

(91)
from which we have
lim U pAq “ max tui u ` 0 (92)
µÑ0

Federico Trionfetti Lecture I M1 2024-2025 Micro I 89 / 107

 Discrete Choice

... just rewriting the last line

lim U pAq “ max tui u

µÑ0

In words: when µ Ñ 0 the variance of the random variable goes to zero,

consumers choices converge on one item only, and that item is the one
that has the highest deterministic component of utility; it follows that
the indirect utility of A coincides with the indirect utility of the item
chosen.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 90 / 107

 Discrete Choice

Property 2. Even when all options give the same utility (ui “ u @i)
the numerosity of A matters. Remember that
˜ ¸
ÿN ui
U pAq “ µ ln eµ
i“1

Simplifying we obtain

U pApuqq “ u ` µ ln N (93)

We see that U pApuqq ą u. A consumer likes the availability of many

alternatives per se. The utility derived from the availability of many
options is measured by µ ln N . Only one good will be chosen but the
wider the choice the happier the consumer is.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 91 / 107

 Discrete Choice

Relationship between Discrete Choice and CES demand. Re-

member that the market demand function for good i obtained maxi-
mization of CES utility is
1 1
ai1´ρ piρ´1
xi “ 1 ρ m (94)
řN 1´ρ ρ´1
j“1 aj pj
xi is the quantity demanded. The expenditure is di “ pi xi
1 ρ
ai1´ρ piρ´1
di “ 1 ρ m (95)
řN 1´ρ ρ´1
j“1 aj pj

Now remember from (81) that the probability that alternative i is chosen
in the LOGIT model is:
ui
eµ
Pi “ ř uj
N
j“1 e
µ

Federico Trionfetti Lecture I M1 2024-2025 Micro I 92 / 107

 Discrete Choice

Consider a consumer whose deterministic part of utility ui is the log of

the quantity that can be purchased given the income m times a param-
eter ai . That is ui “ lnpai pm{pi qq. Then we have
1
pai {pi q µ
Pi “ ř 1 (96)
N µ
j“1 pai {pi q

1
Note that m cancels out. Now, let ai “ pai q ρ and µ “ 1´ρ
ρ . Then the
probability that item i is chosen becomes
1 ρ
ai1´ρ piρ´1
Pi “ 1 ρ (97)
řN 1´ρ ρ´1
j“1 aj pj

which is identical to the expenditure share of the CES. Furthermore,

the expected expenditure on item i for a consumer whose income is m
is Pi m, which is identical to di except that the former is the expected
expenditure while the latter is certain.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 93 / 107
 Examples and exercises

Examples and exercises.

Example 0
Expenditure minimization for a Cobb-Douglas Utility

min pp1 x1 ` p2 x2 q s.t. px1 qα px2 q1´α “ u. (98)

x1 ,x2

L “ p1 x1 ` p2 x2 ´ λ px1 qα px2 q1´α ´ u . (99)

“ ‰

The First Order Conditions are

BL
“0 ùñ p1 “ λαxα´1
1 x1´α
2 (100)
Bx1
BL
“0 ùñ p2 “ λp1 ´ αqxα1 x´α
2 (101)
Bx2
BL
“0 ùñ px1 qα px2 q1´α “ u (102)
Bλ
Federico Trionfetti Lecture I M1 2024-2025 Micro I 94 / 107
 Examples and exercises

The system (100)-(102) can be solved as follows:

˘1
1 Solve (102) for x . You will obtain x “ uxα´1 α .
`
1 1 2
2 Substitute the x1 you have obtained above into (101), then solve
for x2 . You will obtain x2 “ λp1 ´ αqu{p2 .
3 Substitute the x2 you have obtained in point 2 into the x1 you have
obtained in point 1. You will obtain x1 “ u rp1 ´ αqλ{p2 spα´1q{α .
4 Substitute the x2 obtained in point 2 and the x1 obtained in point
3 into (100) and solve for λ. You obtain
# „  α´1 +α´1 „
λp1 ´ αqu 1´α

λp1 ´ αq α
p1 “ λα u (103)
p2 p2
˙ pα´1q2 `1´α
1´α
ˆ
pα´1q2 α
p1 “ αλ 1` α `1´α
puqα´1`1´α (104)
p2
suite Ñ
Federico Trionfetti Lecture I M1 2024-2025 Micro I 95 / 107
 Examples and exercises

... suite
˙ 1´α
1´α α
ˆ
1
p1 “ αλ α (105)
p2
1 ´ α 1´α
ˆ ˙
α α
p1 “ α λ (106)
p2

and finally
´ p ¯α ˆ p ˙1´α
1 2
λ˚CD “ (107)
α 1´α

5 Now substitute λ˚CD in x1 obtained in point 3 and into the x2

obtained in point 2. You will obtain
ˆ ˙1´α ˆ ˙α
p2 α p1 p1 ´ αq
x˚1 “ u et x˚2 “ u (108)
p1 p1 ´ αq p2 α

Federico Trionfetti Lecture I M1 2024-2025 Micro I 96 / 107

 Examples and exercises

Substitute x˚1 and x˚2 in px and you will obtain the expenditure function
´ p ¯α ˆ p ˙1´α
1 2
˚
e pp, uq “ u. (109)
α 1´α

7 Check the properties of the expenditure function we have

demonstrated in page 48. You do not have to prove the properties
(we have proven them in page 48). You just have to check them.
Memo: The properties are
Homogeneous of degree 1 in p. (check it by multiply prices by t).
Non-decreasing in p. (check it by taking the derivatives with
respect to each pi ).
Concave in p. (check it by taking choosing two price vectors and
their convex combinations)
Continuous in p. Checked by differentiation. Given that e˚ pp, uq is
differentiable it is continuous.
8 Construct the correpsonding utility maximization problem and
check that the solution of the expenditure minimization problem is
the same as the solution of the corresponding utility maximization.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 97 / 107
 Examples and exercises

Example 1
` ˘α ´ x ¯1´α
Consider the utility function u “ xα1 2
1´α . Find the money
metric and the indirect utility function. You will find:

epp, uq “ pα1 p1´α

2 u (110)
´ ¯α ˆ x ˙1´α
α 1´α x1 2
mpp, xq “ p1 p2 (111)
α 1´α
ˆ ˙α ˆ ˙1´α
p1 p2
µpp; q, mq “ m (112)
q1 q2

A similar example is in Varian Sect. 7.5. Observe that the utility func-
tion above and the one in Varian Sect. 7.5 are not the same, and yet the
resulting mpp, xq and µpp; q, mq are the same. Why?

Federico Trionfetti Lecture I M1 2024-2025 Micro I 98 / 107

 Examples and exercises

Exercise 1
1
Consider the utility function upxq “ ra1 px1 qρ ` a2 px2 qρ s ρ .
Find the money metric and the indirect utility function.

A neat way to do this exercise is to go through the following steps:

1 Minimize the expenditure at prices p and obtain the expenditure
function.7
2 Replace the utility function in the expenditure function to obtain
the money-metric utility function.
3 Maximize utility at prices q and obtain the indirect utility.
4 Replace the indirect utility in place of utility in the expenditure
function to obtain the money-metric indirect utility function.
...
7
In Varian Sect. 7.5 you will find a similar exercise that will help you obtain the
expenditure function.
Federico Trionfetti Lecture I M1 2024-2025 Micro I 99 / 107
 Examples and exercises

... As we implement step one by writing the Lagrangian and taking

derivatives we soon realize that we get into algebraic complications. To
simplify our task we can minimize a monotonic transformation of the
utility function. A suitable monotonic transformation consists in elevat-
ing utility to the power of ρ. If we do it, write the Lagrangian, and take
derivatives we obtain the following:
BL
“0 ùñ p1 “ λρa1 px1 qρ´1 (113)
Bx1
BL
“0 ùñ p2 “ λρa2 px2 qρ´1 (114)
Bx2
BL
“0 ùñ a1 px1 qρ ` a2 px2 qρ “ uρ (115)
Bλ
Continue alone. Good luck.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 100 / 107

 Examples and exercises

Exercise 2
Additional exercises:
Varian Ch. 7: number 7.1, 7.3, 7.4.
Exercises from past exams. You will find past exams in Amétice.
Select the exercises related to this lecture and do them.
Do the exercises in the examples above without looking at the
slides.
Exercises in the course slides. They are marked by a blue color
text.
Practice these exercises again and again until you are able to do them
in without the least hesitation. This is important because in the exams
you will be under time constraint.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 101 / 107

 Examples and exercises

Exercise 3
Relationship between LOGIT and CES maximization.
1 Consider the case ρ Ñ 1. What happens to the elasticity of
substitution of the CES? What happens to di ? Now think in terms
of the LOGIT equivalent of CES maximization. What happens to
the probability that item i is chosen? What is the utility of the
choice set A ? Explain the result intuitively.
2 Consider the case ρ Ñ 0 What happens to the elasticity of
substitution of the CES? What happens to di ? Now think in terms
of the LOGIT equivalent of CES maximization. What happens to
the probability that item i is chosen? What is the utility of the
choice set A ? Explain the result intuitively.
3 Suppose ui “ u @i. What is the increase in utility of the choice set
A obtained by adding one more item to it? Suppose u “ lnpm{pq,
how much is a consumer willing to pay in order to have one
additional item in the choice set?

Federico Trionfetti Lecture I M1 2024-2025 Micro I 102 / 107

 Self evaluation

Self evaluation

Self evaluation. For the three utility functions studied above, students
must be able to easily do the following:
Utility maximization
Expenditure minimization
Verify that the indirect utility and expenditure functions have the
properties studied for general functions.
Verify the correspondence between Marchallian and Hicksian
demand functions.
Obtain the money metric utility and indirect utility functions.
If you have difficulties with this, you have a problem. If you do, let me
know.

Federico Trionfetti Lecture I M1 2024-2025 Micro I 103 / 107

 Examples of violation of rationality

Example of the framing problem. A consumer wants to buy a stereo

and a coffee machine.
Frame A. The vendor says: “In this shop the stereo costs € 345
and the coffee machine € 15 but if you go to our other shop they
will give you a € 5 discount on either item, I do not remember
which one.”
You have to decide whether to go or not to the other shop. Do you care
on which item you get the discount in order to make your decision?

Federico Trionfetti Lecture I M1 2024-2025 Micro I 104 / 107

 Examples of violation of rationality

Now consider these other two frames

Frame B. The vendor says: “In this shop the stereo costs € 345
and the coffee machine € 15 but if you go to our other shop they
will give you a € 5 discount on the coffee machine.
Frame C. The vendor says: “In this shop the stereo costs € 345
and the coffee machine € 15 but if you go to our other shop they
will give you a € 5 discount on the stereo.
Rank these three choices:
x = go to other shop if discount on coffee machine
y = go to other shop if discount on stereo
z = buy both items in first shop.
In experimental economics it is observed that in frame A most people
rank x „ y, in frame B most people rank x ą z, and in frame C most
people rank z ą y. This result violates transitivity because if x ą z and
z ą y then we should have x ą y, but we don’t. ( back )

Federico Trionfetti Lecture I M1 2024-2025 Micro I 105 / 107

 Examples of violation of rationality

Condorcet paradox. Consider three agents, three options, and the

following individual preferences:
Agent 1: A ą1 B ą1 C
Agent 2: C ą2 A ą2 B
Agent 1: B ą3 C ą3 A
Choice is made by majority voting where first a choice is made between
A and B, then between B and C, and lastly between C and A.
The voting outcome is
The first vote gives A ą B
The second vote gives B ą C
The third vote gives C ą A
These voting outcomes clearly make the collective preference non-transitive
because we have A ą B ą C ą A. ( back )

Federico Trionfetti Lecture I M1 2024-2025 Micro I 106 / 107

 Examples of violation of rationality

Notation varies from textbook to textbook. It is convenient to define

our notation here.
x ě y means that each element of vector x is greater than or equal
to each corresponding element of vector y.
Note that the condition x ě y jointly with the condition x ‰ y
means that each element of vector x is greater than or equal to
each corresponding element of vector y and that at least one
element of x is strictly greater than its corresponding element of
vector y.
x " y means that each element of vector x is greater than each
corresponding element of vector y.
ř ” ı1{2
}x ´ y} “ ki“1 pxi ´ yi q2 is the Euclidean distance between x
and y (k is the number of elements of the vectors).
The symbol ^ means ‘and’.
An open interval pa, bq, a closed interval ra, bs.
( back )
Federico Trionfetti Lecture I M1 2024-2025 Micro I 107 / 107