You are on page 1of 48

# Topic 2: Constrained optimisation

ECON30010 Microeconomics

9 - 16 March

ECON30010

## Topic 2: Constrained optimisation

9 - 16 March

1 / 48

Topic 1 refresher
In most economic problems, the following assumptions seem reasonable:
I

compared.

## Transitivity: if bundle x is at least as good as y and y is at least as

good as bundle z, then x is at least as good as z.

## Continuity (not an exact definition): if bundle x is better than bundle

y and bundle z is sufficiently close to y, then x is better than z.

Theorem
If preferences over bundles of goods x satisfy completeness, transitivity,
and continuity, then there exists a continuous utility function u(x) that
represents these preferences.
Why is this theorem important? Because mathematicians are very good at
working with functions, and we can use a lot of their tools.
ECON30010

9 - 16 March

2 / 48

I
I
I

I
I

## Having a continuous utility function is very useful, but not enough.

Budget set
Utility maximisation problem
Strict monotonicity We cover these two axioms here because they
are related to the maximisation problem.
Convexity
The Lagrange method
Marshallian demand

The name of this topic sounds like math. And our main focus will be
math: it is going to be applied to an economics problem, but my main
goal is to make sure you are comfortable with constrained optimisation.
Why?
We ourselves will need it later, but, more importantly, other subjects will
need it, and much more desperately than we would.
So, any effort to become fully comfortable with constrained optimisation
will pay off handsomely in the future (when your macro lecturer will zip past
the calculations and straight to the optimal solution).
ECON30010

9 - 16 March

3 / 48

Constraint

Constraint
I

I
I
I

## A typical agent wants stuff. If fifth TV set in a room does not

make an agent any happier, something else (more time? extra hike in
mountains?) would make the agent more satisfied.
If an agent to face a meaningful problem where the solution is not
giving an agent an infinite amount of something we need to
introduce some constraint.
Earlier, we have dealt with a simple constraint: an agent had \$5 that
she needed to allocate.
This is not the most typical constraint the agent face, although we
will see plenty of similar simple constraints in this subject.
We now turn to a more typical one: budget constraint.
I

## Budget constraint is often thought of as income and prices, but you

will see on a tutorial that we can interpret the same constraint as
something completely different.

## Even though I focus on budget constraint, everything we study

applies to other types of constraints, including \$5 one.
ECON30010

## Topic 2: Constrained optimisation

9 - 16 March

4 / 48

Constraint

Budget Constraint
Budget constraint: good 1 and good 2 cost money and an agent (even
Warren Buffet) has a limited amount of it.
A consumer is restricted to choose a consumption bundle q = (q1 , q2 )
such that p1 q1 + p2 q2 Y . Right now, we are not interested where prices
p1 , p2 and income Y come from.

Y
p2

q2

If q1 = 0, p2 q2 = Y q2 = Y /p2

Budget line p1 q1 + p2 q2 = Y
If q2 = 0, p1 q1 = Y q1 = Y /p1

Budget set p1 q1 + p2 q2 Y
q1
Y /p1
ECON30010

9 - 16 March

5 / 48

## Consumer maximisation problem

Formally, we can write this problem as:

max u(q1 , q2 )
q1 ,q2

subject to p1 q1 + p2 q2 Y

## Note that I can write my \$5 problem as:

max u(x, y )
x,y

subject to x + y = 5

ECON30010

9 - 16 March

6 / 48

## Indifference curves: ECON20002 recap

In ECON20002 you have learned1 that the indifference curve is the
solution to u(x, y ) = u0 , for different levels of u0 .

q2

q1

1

ECON30010

9 - 16 March

7 / 48

## Solving consumer maximisation problem graphically:

ECON20002 recap

Y
p2

q2

q1
Y /p1
We are looking for the North-East-most indifference curve2 , such that it
touches budget constraint (to guarantee the feasibility of the solution).
2

Utility increases in that direction see the picture on the previous slide
ECON30010

9 - 16 March

8 / 48

## What can go wrong? Satiation

Consider an agent, Ann, with preferences:
uA (q1 , q2 ) = (q1 1)2 (q2 1)2
A similar function appeared in Tutorial 1. We know that this function is
maximised at q = (1, 1).
Suppose now that Ann has plenty of money
(\$4) and prices are p1 = p2 = 1. Her budget
constraint is q1 + q2 4.
From Tutorial 1,3 we know that Anns
indifference curves are circles. We can also
find that at point q = (2, 2) indifference
curve touches the budget line.4

Y
p2

q2

q1
Y /p1

.

## Note: tutorials are very useful!

How: by symmetry, we can guess that the point is on the intersection of budget line
and a 45 line, then find this point.
4

ECON30010

9 - 16 March

9 / 48

## Consumer maximisation problem

Strict monotonicity
I

This would not happen if utility function u were increasing: then for
any point in the interior of the budget set you could increase utility by
giving more of some good.

## However, we should not impose this assumption on utility function:

utility function is a fake, and we need to know what it means in terms
of something that is not a fake (something that we can observe, such
as choice/preferences).

## This leads us to a new condition on preferences, strict monotonicity:5

Definition
If bundle x contains
I

## then a consumer with strictly monotone preferences strictly prefers x to y.

5

This condition was introduced in ECON20002 under the name more is better.
ECON30010

9 - 16 March

10 / 48

## Implication of strict monotonicity.

Theorem: With strictly monotonic preferences, we can replace inequality
with equality = in the consumer maximisation problem, so that
p1 q1 + p2 q2 = Y .
Proof: Suppose not.6 That is, suppose that we cannot replace by =
because the optimal consumption of a consumer, (q1 , q2 ), is such that
p1 q1 + p2 q2 = Y 0 < Y . Then consumer has Y Y 0 of income left over,
which can be spent on good 1 and good 2 (say, equally). Yet, if
preferences are strictly monotonic, then the consumer is better off. So, if
(q1 , q2 ) is an optimal consumption bundle, preferences could not be
strictly monotonic.
6

The proof technique where you assume that the statement of the theorem is wrong
and then prove that the condition of the theorem is not satisfied is called proof by
contradiction. Here the statement is that we can replace with = and the condition
is preferences are strictly monotonic.
ECON30010

9 - 16 March

11 / 48

## Strict monotonicity, example 1, washing dishes

Definition
If bundle x contains
I

## Suppose that the (only) good is washing

dishes. It does not satisfy monotonicity.
However, this is not a problem: we only need to
define the good as not washing dishes.

ECON30010

9 - 16 March

12 / 48

## Strict monotonicity, example 2, exercise & chocolate

Definition
If bundle x contains
I

## Exercise is good, but too much of exercise is bad.

Eating one chocolate bar is fantastic, but fifth kilo may
cause death.
In most cases, these examples are not a cause of concern
because: (1) goods would be more generally defined (e.g.
consumed over longer periods of time) and (2) the range of
consumption in the problem is where the satiation does not
happen (you wont eat chocolate only).
You still need to examine your problem to see whether the assumption fits.
ECON30010

9 - 16 March

13 / 48

## What if no strict monotonicity?

Strict monotonicity is a technical assumption: it is much easier to solve
a problem if this assumption is imposed, but we are not doomed if this
assumption does not make sense in our problem. There are more
sophisticated methods to find a solution.
With equality, we will use the Lagrange method. With inequalities, we
would need to use the Kuhn-Tucker method.

You can check when they lived and guess which method is easier.
ECON30010

9 - 16 March

14 / 48

## What can go wrong? Non-convexity

Suppose that indifference curves look like this:
Y
p2

q2

## Two optimal consumption bundles

q1
Y /p1

We then have two optimal consumption bundles. This is not a big problem
if finding optimal bundle is our ultimate goal. However, if we want to
know how consumption changes in response to change in prices, having
two bundles is not convenient: we do not know which one the agent would
be consuming.
Hence, we want to rule out this situation too.
ECON30010

9 - 16 March

15 / 48

## Convexity in chocolate and candy

(recap of ECON20002)

The usual assumption is that our marginal utility is decreasing: you are
ecstatic about the first bar of chocolate, but not so much about the
second bar.
This naturally leads to convexity assumption: if you are indifferent
between
x: 3 bars of chocolate & 1 candy and
y : 3 candies & 1 bar of chocolate,
then you must prefer (z) 2 bars of chocolate and 2 candies to x and y .
Does this assumption always hold? Not necessarily: both aspirin and
paracetamol are good against a headache (so, you are, conceivably,
indifferent), but having 1/2 of aspirin and 1/2 of paracetamol is a recipe
for a disaster.
Similarly to strict monotonicity: this is unlikely to be a problem with more
generally defined goods. At the same time, as always, when you think about
your specific problem, you do need to think whether this assumption is satisfied.
ECON30010

9 - 16 March

16 / 48

## Consumer maximisation problem

Convexity
I have told you what convexity means when we talk about 3 chocolates
and 1 candy. We need to be a bit more general than that.
Formal definition of convexity:7

Definition
Suppose that a consumer is indifferent between bundle x and y. Consider
a bundle z := x + (1 )y for any such that 0 < < 1.8
I

## The preferences are strictly convex if z  x, z  y .

Do not confuse convex preferences and convex functions. These are two different
notions. (Only if you are very curious: The set of better allocations the inside of the indifference curve is a convex set;
the set above the convex function is a convex set. Convex set is the set that contains a straight line that connects any two
points of the set.)
8

## This is called a convex combination of x and y.

I do not need the second relation, z  y , because I have already assumed
transitivity. Check that you understand why. The same applies to strict convexity.
9

ECON30010

9 - 16 March

17 / 48

## Strict convexity on a graph

Definition
Suppose that a consumer is indifferent between bundle x and y. Consider a
bundle z := x + (1 )y for any such that 0 < < 1.
I

## Straight line connects x and y.

Any point on this line is z from
the definition.

q2
y

z
x
q1

If utility increases in the northeast direction (as it usually does), then the
whole straight line should be above the indifference curve (recall that
points above indifference curve are more desirable bundles).
ECON30010

9 - 16 March

18 / 48

Y
p2

## Straight line connects x and y.

Any point on this line is z from
the definition.

q2
y
z

x
q1
Y /p1
With strict convexity, we will have a unique solution to our maximisation
problem.
ECON30010

9 - 16 March

19 / 48

## Consumer maximisation problem

Convexity on a graph
Definition
Suppose that a consumer is indifferent between bundle x and y. Consider a
bundle z := x + (1 )y for any such that 0 < < 1.
I

q2

z
y
q1

ECON30010

9 - 16 March

20 / 48

## No unique solution with convexity

Y
p2

q2

s
S
q1
Y /p1
Convexity holds, but solution is not unique: there is an interval of
solutions from s to S.
ECON30010

9 - 16 March

21 / 48

## Consumer maximisation problem

Non-convexity
Definition
Suppose that a consumer is indifferent between bundle x and y. Consider a
bundle z := x + (1 )y for any such that 0 < < 1.
I

## The preferences are convex if z  x, z  y .

Y
p2

q2
x z

q1
Y /p1
These preferences are not convex.
ECON30010

9 - 16 March

22 / 48

More assumptions

## I will also assume that utility function is differentiable, or twice

differentiable. These assumptions are also traceable to assumptions on
preferences. However, it becomes too technical and I will skip that.

ECON30010

9 - 16 March

23 / 48

## Solving consumer maximisation problem

max u(q1 , q2 )
q1 ,q2

subject to p1 q1 + p2 q2 = Y
This is called constrained maximisation problem because you need to (i)
find max u(q1 , q2 ) and (ii) satisfy your constraint (here p1 q1 + p2 q2 = Y ).
Problem (i) alone is called unconstrained optimisation problem. You can
guess that constrained maximisation problem is usually more difficult than
an unconstrained one.
How do you solve constrained maximisation problem?
I Sometimes you can see what the solution is (for example, when
goods are perfect complements or perfect substitutes);
I You can substitute in budget constraint (q1 = 1 (Y p2 q2 )) (this is
p1
what you have done in ECON20002)
I Or you can use the Lagrange method
I

You can also use a shortcut which comes from the Lagrange method;
this is what youve also done in ECON20002.

ECON30010

9 - 16 March

24 / 48

## Constrained Optimisation: Lagranges Method

By Year 3, we know perfectly well how to solve unconstrained optimisation
problem (take a derivative, equate it to zero (that is obtain first-order condition,
FOC), argue/check/think whether FOC gives you a correct solution). We can do
this mindlessly, except for the last step (argue/check/think).
The Lagrange method turns something that you dont know how to solve
into something that you know how to solve. It introduces an auxiliary
problem.
Define Lagrangian as
L(q1 , q2 , ) = u(q1 , q2 ) + [Y p1 q1 p2 q2 ],
where , called a Lagrange multiplier, is an additional decision variable
(along with q1 and q2 ).

max u(q1 , q2 )
q1 ,q2

max L(q1 , q2 , )

q1 ,q2 ,

subject to p1 q1 + p2 q2 = Y
ECON30010

9 - 16 March

25 / 48

## Constrained Optimisation: Lagranges Method, Example 1

max L(q1 , q2 , ) = u(q1 , q2 ) + [Y p1 q1 p2 q2 ]

q1 ,q2 ,

L(q1 ,q2 ,)

q1
L(q1 ,q2 ,)
q2

L(q1 ,q2 ,)

ECON30010

=0
=0
= 0.

9 - 16 March

26 / 48

## Constrained Optimisation: Lagranges Method, Example 1

max L(q1 , q2 , ) = u(q1 , q2 ) + [Y p1 q1 p2 q2 ]

q1 ,q2 ,

ECON30010

9 - 16 March

27 / 48

## Constrained Optimisation: Lagranges Method, Example 1

max L(q1 , q2 , ) = u(q1 , q2 ) + [Y p1 q1 p2 q2 ]

q1 ,q2 ,

ECON30010

9 - 16 March

28 / 48

## Lagrangian: A more careful look

L(q ,q ,)
1 ,q2 )
1 2

= u(q
p1
q1
q1
u(q1 ,q2 )
L(q1 ,q2 ,)
= q2 p2
q

L(q1 ,q2 2 ,)
= Y p1 q1 p2 q2

=0
=0
= 0.

MU1 (q1 , q2 ) = p1
1 (q1 ,q2 )
MRS = MU
MU2 (q1 ,q2 )
MU2 (q1 , q2 ) = p2

p1 q1 + p2 q2
p q + p q = Y
1 1
2 2
=

= pp12
= Y.

MU1
MU2
=
p1
p2

## is equal to marginal utility of good i divided by the price of that

good; that is, an increase in utility if income Y is increased by 1 cent.
ECON30010

9 - 16 March

29 / 48

## Shadow value of a constraint: an example

Revisit our utility function u(q1 , q2 ) = q1 q2 .
Recall that MU1 = q2 , MU2 = q1 , q1 =
previous slide = q2 /p1 = q1 /p2

Y
2p1 ,

and q2 =

Y
2p2 ;

from the

## Let Y = 1, p1 = 1/2, p2 = 2. Then q1 = 1, q2 = 1/4, u(q1 , q2 ) = 0.25 and

the shadow value of a constraint is = 1/2.
Let Y = 1.01 now; p1 = 1/2, p2 = 2 as before. Then q1 = 1.01,
q2 = 1.01/4, u(q1 , q2 ) = 1.012 /4 = 0.255025.
Note that
u = 0.255025 0.25 = 0.005025 0.005 = 1/2 (1.01 1) = Y
Change in u is almost equal to times the change in how binding the
constraint is. It would have been exactly 0.005 if the increase in income Y
has been even smaller (you can try that at home).
ECON30010

9 - 16 March

30 / 48

## Constrained Optimisation: Lagranges Method, Example 2

Consider another utility function: u(q1 , q2 ) = q1 q2 + 3q1 . Let p1 = p2 = 1
and Y = 2
The Lagrangian is
L(q1 , q2 , ) = q1 q2 + 3q1 + [2 q1 q2 ],
L(q1 ,q2 ,)

q1
L(q1 ,q2 ,)
q

L(q1 ,q2 2 ,)

= q2 + 3 = 0
= q1 = 0
= 2 q1 q2 = 0.

(
q1 = q2 + 3
q2 + 3 + q2 = 2

2q2 = 1

ECON30010

9 - 16 March

31 / 48

Y
p2

q2

q1
Y /p1

## Optimal consumption bundle

(corner solution)

## Optimal consumption bundle

if can consume negative quantities

quantities.
ECON30010

9 - 16 March

32 / 48

## A diversion: a complete constrained problem

max u(q1 , q2 )
q1 ,q2

p1 q1 + p2 q2 = Y
subject to q1 0

q2 0
Yet, the solution to this problem, using Kuhn-Tucker conditions, would be
too messy. Hence, we ignore some of the constraints, with a hope that
they would not matter, but need to revisit our problem once we obtained
the solution to a simplified problem.
Surprisingly, a lot of problems in economics are solved this way: Original
problem is too hard; let us try to solve a simplified problem, hoping that its
solution will satisfy other constraints; if not, try a different simplification.
ECON30010

9 - 16 March

33 / 48

## Marshallian demand function

(
q1 =
q2 =

Y
2p1
Y
2p2

Note that we obtained the solution for any prices p1 , p2 and any income
Y . That is, we could write the solution as functions q1 (p1 , p2 , Y ) and
q2 (p1 , p2 , Y ). These functions are called Marhsallian, or uncompensated,10
demand.

10
You can look up your notes from ECON20002 if you want to know right now why
the demand is called uncompensated, or you can wait a little.
ECON30010

9 - 16 March

34 / 48

Summary

Main points:
I

## Assumptions on individual preferences allow us to represent these

preferences by a continuous utility function;
Maximisation of a utility function subject to a budget constraint
(using Lagrange method) leads to Marshallian demand functions.
I

## Lagrange method: turn constrained optimisation problem into

unconstrained maximisation problem.
Solve unconstrained maximisation problem using standard methods.

Whats next:
I

## An odd maximisation problem.

Normal, inferior and Giffen goods.

ECON30010

9 - 16 March

35 / 48

(Ii)

ECON30010

9 - 16 March

36 / 48

(II)

## max u(q1 , q2 ) = ln(q1 b1 ) 2 ln(b2 q2 )

q1 ,q2

subject to p1 q1 + p2 q2 = Y
Corresponding Lagrangian
L = ln(q1 b1 ) 2 ln(b2 q2 ) + [Y p1 q1 p2 q2 ]

ECON30010

9 - 16 March

37 / 48

## Solving problem 1: FOC

1
L
=
p1 = 0
q1
q1 b1
2
L
=
p2 = 0
q2
b2 q2
L
= Y p1 q1 p2 q2 = 0

q2 = b2 2

1
= p1
q1 b1
2
= p2
b2 q2
p1 q1 + p2 q2 = Y

b2 q2
p1
=2
q1 b1
p2
p1 q1 + p2 q2 = Y

p1
(q1 b1 )
p2

p1
(q1 b1 ))
p2
= p1 q1 + p2 b2 2p1 (q1 b1 ) = p2 b2 p1 q1 + 2p1 b1


Y p2 b2
p1
Y p2 b2
q1 = 2b1
; q2 = b2 2
b1
p1
p2
p1
1
Y b1 p1
= b2 2 (b1 p1 Y + p2 b2 ) = 2
b2
p2
p2
Y = p1 q1 + p2 q2 = p1 q1 + p2 (b2 2

ECON30010

9 - 16 March

38 / 48

## Solution to utility maximisation problem 1

Marshallian (uncompensated) demand is:
Y p2 b2
p1
Y b1 p1
q2 (p1 , p2 , Y ) = 2
b2
p2
q1 (p1 , p2 , Y ) = 2b1

## However, this is not the end of the solution:11 suppose that

Y = 3, b1 = b2 = p1 = p2 . Then q1 = 2 31
1 = 0. If we go back to the
utility function, u(q1 , q2 ) = ln(q1 b1 ) 2 ln(b2 q2 ), it is not defined if
q1 = 0 and b1 = 1 because it leads to ln(0 1), which is undefined. To
ensure that q1 b1 > 0 and b2 q2 > 0 we need to impose the following
conditions:
Y p2 b2
For q1 > b1 we need 2b1
> b1 hence Y p2 b2 < p1 b1 and
p1
for 0 < q2 < b2 we need Y p2 b2 < p1 b1 < Y p2 b2 /2
11
We always need to check that our solution makes sense, but in simple examples it is
obvious.
ECON30010

9 - 16 March

39 / 48

Types of goods

## Types of goods: normal goods

Consider good 2:
q2 (p1 , p2 , Y )

=
Y
Y



Y b1 p1
2
2
b2 =
> 0 because p2 > 0.
p2
p2

## The consumption of good 2 increases as the income of this individual

increases.
These goods are called normal.

ECON30010

9 - 16 March

40 / 48

Types of goods

## Types of goods: luxury goods

In fact, we can do even more math and look at what share of income this
individual spends on good 2, and how it changes with income:
Y q2 (p1 , p2 , Y ) q2 (p1 , p2 , Y )
p2 q2 (p1 , p2 , Y )
= p2 Y
Y
Y
Y2
2Y (2(Y b1 p1 ) b2 p2 )
b1 p1 + b2 p2
=
=
>0
2
Y
Y2
This is a luxury good. For this good, the individual spends larger share of
her income on the good as income rises. That is, if an individual spends
10% of income on good 2 at income level Y, if income increases by , the
share spent from this  would be above 10%.12 In fact, in our case it is
2 p2
possible that b1 p1Y+b
> 1, so an individual spend the whole  and more
2
on good 2.
12

ECON30010

9 - 16 March

41 / 48

Types of goods

## Types of goods: inferior goods

Consider good 1:

q1 (p1 , p2 , Y )
=
Y
Y



Y p2 b2
1
2b1
= < 0 because p1 > 0
p1
p1

## The consumption of good 1 decreases as the income of this individual

increases.
These goods are called inferior.
Note that our formula implies that if the price of good 1 is very low, the
consumption of this good decreases extremely fast as income increases. It may
sound peculiar on a first sight, but should not be surprising: if price of good 1 is
very low, then this individual consumes a lot of good 1 to begin with, so the fall
in consumption is also very large.
ECON30010

## Topic 2: Constrained optimisation

9 - 16 March

42 / 48

Types of goods

Giffen goods
Consider good 1 again:

q1 (p1 , p2 , Y )
=
p1
p1

Y p2 b2
2b1
p1


=

Y p2 b2
(p1 )2

1 ,p2 ,Y )
Suppose Y , p2 and b2 are such that Y p2 b2 > 0. Then q1 (pp
> 0.
1
It means that as price of good 1 increases, the consumption of good 1
increases. These goods are called Giffen goods. This is an unusual
property and we will explore it in a little more detail.

First, let us see what the condition Y p2 b2 > 0 implies in terms of the
2 b2
amount of good 1 consumed. The demand for good 1 is 2b1 Y p
, so
p1
if Y p2 b2 > 0, then q1 < 2b1 . That is,
consumption of good 1 is low.
ECON30010

q1 (p1 ,p2 ,Y )
p1

9 - 16 March

43 / 48

Types of goods

## Giffen property vs. Giffen goods

q1 (p1 , p2 , Y )
> 0 if Y p2 b2 > 0
p1
q1 (p1 , p2 , Y )
< 0 otherwise
p1
We see that derivative may change for different values of Y , p2 and b2 .
A more proper terminology should be a good that exhibits Giffen
property in the range b1 < q1 < 2b1 .13
All other goods can be change their properties depending on particular
prices and income; for example, a good can be normal over a range of
prices and incomes and inferior over a different range.
13

ECON30010

9 - 16 March

44 / 48

Types of goods

## Another example of constrained optimisation

We will talk more about Giffen behaviour; before we turn to that, we will
revisit our utility function that gave Giffen behaviour and will change just
one coefficient.
Old:

New:

ECON30010

9 - 16 March

45 / 48

Types of goods

## Solving problem 2: FOC

L
1
=
p1 = 0
q1
q1 b1
L
2
=
p2 = 0
q2
b2 q2
L
= Y p1 q1 p2 q2 = 0

1
= p1
q1 b1
2
= p2
b2 q2
p1 q1 + p2 q2 = Y

p1
b2 q2
=
q1 b1
p2
p1 q1 + p2 q2 = Y
.
p1
q2 = b2 (q1 b1 )
p2

p1
(q1 b1 ))
p2
= p1 q1 + p2 b2 p1 (q1 b1 ) = p2 b2 p1 b1

Y = p1 q1 + p2 q2 = p1 q1 + p2 (b2

Where is q1?
Or q2 ?
What if p2 b2 p1 b1 is not equal Y ?
ECON30010

9 - 16 March

46 / 48

Types of goods

## Indifference curves are indifference lines, like with perfect complements! It

is not surprising that we cannot find a solution (it is akin to maximising
linear function f (x) = x using first order conditions: youve tried
something like this in Tutorial 1).
ECON30010

9 - 16 March

47 / 48

Types of goods

Summary
Skills:
I

does not work.