1 Mathematical and Economic Foundations

A Mathematical Preliminaries
1 Unconstrained Optimization
2 Constrained Optimization
B Consumer Optimization
1 Graphical Analysis
2 Algebraic Analysis
3 The Time Dimension
4 The Risk Dimension
C General Equilibrium
1 Optimal Allocations
2 Equilibrium Allocations
Mathematical Preliminaries

Unconstrained Optimization

max F (x)
x

Constrained Optimization

max F (x) subject to c ≥ G (x)

x
Unconstrained Optimization

To find the value of x that solves

max F (x)
x
you can:
1. Try out every possible value of x.
2. Use calculus.

Since search could take forever, let’s use calculus instead.

Unconstrained Optimization

Theorem If x ∗ solves

max F (x),
x
∗
then x is a critical point of F , that is,

F 0 (x ∗ ) = 0.
Unconstrained Optimization

F (x) maximized at x ∗ = 5
Unconstrained Optimization

F 0 (x) > 0 when x < 5. F (x) can be increased by increasing x.

Unconstrained Optimization

F 0 (x) < 0 when x > 5. F (x) can be increased by decreasing x.

Unconstrained Optimization

F 0 (x) = 0 when x = 5. F (x) is maximized.

Unconstrained Optimization

Theorem If x ∗ solves

max F (x),
x
∗
then x is a critical point of F , that is,

F 0 (x ∗ ) = 0.

Note that the same first-order necessary condition F 0 (x ∗ ) = 0

also characterizes a value of x ∗ that minimizes F (x).
Unconstrained Optimization

F 0 (x) = 0 when x = 5. F (x) is minimized.

Unconstrained Optimization

F 0 (x) = 0 and F 00 (x) < 0 when x = 5. F (x) is maximized.

Unconstrained Optimization

F 0 (x) = 0 and F 00 (x) > 0 when x = 5. F (x) is minimized.

Unconstrained Optimization

Theorem If x ∗ solves

max F (x),
x
∗
then x is a critical point of F , that is,

F 0 (x ∗ ) = 0.
Unconstrained Optimization

Theorem If

F 0 (x ∗ ) = 0 and F 00 (x ∗ ) < 0,
then x ∗ solves

max F (x)
x

(at least locally).

The first-order condition F 0 (x ∗ ) = 0 and the second-order

condition F 00 (x ∗ ) < 0 are sufficient conditions for the value of
x that (locally) maximizes F (x).
Unconstrained Optimization

F 0 (x ∗∗ ) = 0 and F 00 (x ∗∗ ) < 0 at the local maximizer x ∗∗ and

F 0 (x ∗ ) = 0 and F 00 (x ∗ ) < 0 at the global maximizer x ∗ .
Unconstrained Optimization

F 0 (x ∗∗ ) = 0 and F 00 (x ∗∗ ) < 0 at the local maximizer x ∗∗ and

F 0 (x ∗ ) = 0 and F 00 (x ∗ ) < 0 at the global maximizer x ∗ , but
F 00 (x) > 0 in between x ∗ and x ∗∗ .
Unconstrained Optimization

Theorem If

F 0 (x ∗ ) = 0
and

F 00 (x) < 0 for all x ∈ R,

then x ∗ solves

max F (x).
x
Unconstrained Optimization

F 00 (x) < 0 for all x ∈ R and F 0 (5) = 0. F (x) is maximized

when x = 5.
Unconstrained Optimization

If F 00 (x) < 0 for all x ∈ R, then the function F is concave.

When F is concave, the first-order condition F 0 (x ∗ ) = 0 is

both necessary and sufficient for the value of x that maximizes
F (x).

And, as we are about to see, concave functions arise frequently

and naturally in economics and finance.
Unconstrained Optimization: Example 1

Consider the problem

 
1
max − (x − τ )2 ,
x 2
where τ is a number (τ ∈ R) that we might call the “target.”

The first-order condition

−(x ∗ − τ ) = 0
leads us immediately to the solution: x ∗ = τ .
Unconstrained Optimization: Example 2

Consider maximizing a function of three variables:

max F (x1 , x2 , x3 )
x1 ,x2 ,x3

Even if each variable can take on only 1,000 values, there are
one billion possible combinations of (x1 , x2 , x3 ) to search over!

This is an example of what Richard Bellman (US, 1920-1984)

called the “curse of dimensionality.”
Unconstrained Optimization: Example 2

Consider the problem:

     
1 2 1 2 1
max − (x1 − τ ) + − (x2 − x1 ) + − (x3 − x2 )2 .
x1 ,x2 ,x3 2 2 2

Now the three first-order conditions

−(x1∗ − τ ) + (x2∗ − x1∗ ) = 0

−(x2∗ − x1∗ ) + (x3∗ − x2∗ ) = 0
−(x3∗ − x2∗ ) = 0
lead us to the solution: x1∗ = x2∗ = x3∗ = τ .
Constrained Optimization

To find the value of x that solves

max F (x) subject to c ≥ G (x)

x
you can:
1. Try out every possible value of x.
2. Use calculus.

Since search could take forever, let’s use calculus instead.

Constrained Optimization

A method for solving constrained optimization problems like

max F (x) subject to c ≥ G (x)

x

was developed by Joseph-Louis Lagrange (France/Italy,

1736-1813) and extended by Harold Kuhn (US, b.1925) and
Albert Tucker (US, 1905-1995).
Constrained Optimization

Associated with the problem:

max F (x) subject to c ≥ G (x)

x

Define the Lagrangian

L(x, λ) = F (x) + λ[c − G (x)],

where λ is the Lagrange multiplier.
Constrained Optimization

L(x, λ) = F (x) + λ[c − G (x)],

Theorem (Kuhn-Tucker) If x ∗ maximizes F (x) subject to

c ≥ G (x), then there exists a value λ∗ ≥ 0 such that,
together, x ∗ and λ∗ satisfy the first-order condition

F 0 (x ∗ ) − λ∗ G 0 (x ∗ ) = 0

and the complementary slackness condition

λ∗ [c − G (x ∗ )] = 0.
Constrained Optimization

In the case where c > G (x ∗ ), the constraint is non-binding.

The complementary slackness condition

λ∗ [c − G (x ∗ )] = 0

requires that λ∗ = 0.

And the first-order condition

F 0 (x ∗ ) − λ∗ G 0 (x ∗ ) = 0

requires that F 0 (x ∗ ) = 0.
Constrained Optimization

In the case where c = G (x ∗ ), the constraint is binding. The

complementary slackness condition

λ∗ [c − G (x ∗ )] = 0

puts no further restriction on λ∗ ≥ 0.

Now the first-order condition

F 0 (x ∗ ) − λ∗ G 0 (x ∗ ) = 0

requires that F 0 (x ∗ ) = λ∗ G 0 (x ∗ ).
Constrained Optimization: Example 1

For the problem

 
1
max − (x − 5)2 subject to 7 ≥ x,
x 2
F (x) = (−1/2)(x − 5)2 , c = 7, and G (x) = x. The
Lagrangian is
 
1
L(x, λ) = − (x − 5)2 + λ(7 − x).
2
Constrained Optimization: Example 1

With  
1
L(x, λ) = − (x − 5)2 + λ(7 − x),
2
the first-order condition

−(x ∗ − 5) − λ∗ = 0

and the complementary slackness condition

λ∗ (7 − x ∗ ) = 0

are satisfied with x ∗ = 5, F 0 (x ∗ ) = 0, λ∗ = 0, and 7 > x ∗ .

Constrained Optimization: Example 1

Here, the solution has F 0 (x ∗ ) = 0 since the constraint is

nonbinding.
Constrained Optimization: Example 2

For the problem

 
1
max − (x − 5)2 subject to 4 ≥ x,
x 2
F (x) = (−1/2)(x − 5)2 , c = 4, and G (x) = x. The
Lagrangian is
 
1
L(x, λ) = − (x − 5)2 + λ(4 − x).
2
Constrained Optimization: Example 2

With  
1
L(x, λ) = − (x − 5)2 + λ(4 − x),
2
the first-order condition

−(x ∗ − 5) − λ∗ = 0

and the complementary slackness condition

λ∗ (4 − x ∗ ) = 0

are satisfied with x ∗ = 4 and F 0 (x ∗ ) = λ∗ = 1 > 0.

Constrained Optimization: Example 2

Here, the solution has F 0 (x ∗ ) = λ∗ G 0 (x ∗ ) > 0 since the

constraint is binding. F 0 (x ∗ ) > 0 indicates that we’d like to
increase the value of x, but the constraint won’t let us.
Constrained Optimization: Example 2

With a binding constraint, F 0 (x ∗ ) 6= 0 but

F 0 (x ∗ ) − λ∗ G 0 (x ∗ ) = 0. The value x ∗ that solves the problem
is a critical point, not of the objective function F (x), but
instead of the entire Lagrangian F (x) + λ[c − G (x)].
Consumer Optimization

1. Graphical Analysis
2. Algebraic Analysis
3. Time Dimension
4. Risk Dimension
Consumer Optimization

Alfred Marshall, Principles of Economics, 1890. – supply and

demand

Francis Edgeworth, Mathematical Psychics, 1881.

Vilfredo Pareto, Manual of Political Economy, 1906. –

indifference curves
Consumer Optimization

John Hicks, Value and Capital, 1939. – wealth and

substitution effects

Paul Samuelson, Foundations of Economic Analysis, 1947. –

mathematical reformulation

Irving Fisher, The Theory of Interest, 1930. – intertemporal

extension.
Consumer Optimization

Gerard Debreu, Theory of Value, 1959.

Kenneth Arrow, “The Role of Securities in the Optimal

Allocation of Risk Bearing,” Review of Economic Studies,
1964.

Extensions to include risk and uncertainty.

Consumer Optimization: Graphical Analysis

Consider a consumer who likes two goods: apples and bananas.

Y = income
ca = consumption of apples
cb = consumption of bananas
pa = price of an apple
pb = price of a banana
The consumer’s budget constraint is

Y ≥ pa ca + pb cb
Consumer Optimization: Graphical Analysis

So long as the consumer always prefers more to less, the

budget constraint will always bind:

Y = pa ca + pb cb
or
 
Y pa
cb = − ca
pb pb
Which shows that the graph of the budget constraint will be a
straight line with slope −(pa /pb ) and intercept Y /pb .
Consumer Optimization: Graphical Analysis

The budget constraint is a straight line with slope −(pa /pb )

and intercept Y /pb .
Consumer Optimization: Graphical Analysis

The budget constraint describes the consumer’s market

opportunities.

Francis Edgeworth (Ireland, 1845-1926) and Vilfredo Pareto

(Italy, 1848-1923) were the first to use indifference curves to
describe the consumer’s preferences.

Each indifference curve traces out a set of combinations of

apples and bananas that give the consumer a given level of
utility or satisfaction.
Consumer Optimization: Graphical Analysis

Each indifference curve traces out a set of combinations of

apples and bananas that give the consumer a given level of
utility.
Consumer Optimization: Graphical Analysis

Each indifference curve slopes down, since the consumer

requires more apples to compensate for a loss of bananas and
more bananas to compensate for a loss of apples, if more is
preferred to less.
Consumer Optimization: Graphical Analysis

Indifference curves farther away from the origin represent

higher levels of utility, if more is preferred to less.
Consumer Optimization: Graphical Analysis

A and B yield the same level of utility, and B and C yield the
same level of utility, but C is preferred to A if more is preferred
to less. Indifference curves cannot intersect.
Consumer Optimization: Graphical Analysis

Indifference curves are convex to the origin if consumers have

a preference for diversity.
Consumer Optimization: Graphical Analysis

A is suboptimal and C is infeasible. B is optimal.

Consumer Optimization: Graphical Analysis

At B, the optimal choice, the indifference curve is tangent to

the budget constraint.
Consumer Optimization: Graphical Analysis

Recall that the budget constraint

Y = pa ca + pb cb
or
 
Y pa
cb = − ca
pb pb
has slope −(pa /pb ).
Consumer Optimization: Graphical Analysis

Suppose that the consumer’s preferences are also described by

the utility function

u(ca ) + βu(cb ).
The function u is increasing, with u 0 (c) > 0, so that more is
preferred to less, and concave, with u 00 (c) < 0, so that
marginal utility falls as consumption rises.

The parameter β measures how much more (if β > 1) or less

(if β < 1) the consumer likes bananas compared to apples.
Consumer Optimization: Graphical Analysis

Since an indifference curve traces out the set of (ca , cb )

combinations that yield a given level of utility Ū, the equation
for an indifference curve is

Ū = u(ca ) + βu(cb ).
Use this equation to define a new function, cb (ca ), describing
the number of bananas needed, for each number of apples, to
keep the consumer on this indifference curve:

Ū = u(ca ) + βu[cb (ca )].

Consumer Optimization: Graphical Analysis

The function cb (ca ) satisfies Ū = u(ca ) + βu[cb (ca )].

Consumer Optimization: Graphical Analysis

Differentiate both sides of

Ū = u(ca ) + βu[cb (ca )]

to obtain

0 = u 0 (ca ) + βu 0 [cb (ca )]cb0 (ca )

or

u 0 (ca )
cb0 (ca ) = − .
βu 0 [cb (ca )]
Consumer Optimization: Graphical Analysis

This last equation,

u 0 (ca )
cb0 (ca ) = − ,
βu 0 [cb (ca )]
written more simply as

u 0 (ca )
cb0 (ca ) = − ,
βu 0 (cb )
measures the slope of the indifference curve: the consumer’s
marginal rate of substitution.
Consumer Optimization: Graphical Analysis

Thus, the tangency of the budget constraint and indifference

curve can be expressed mathematically as

pa u 0 (ca )
= .
pb βu 0 (cb )
The marginal rate of substitution equals the relative prices.
Consumer Optimization: Graphical Analysis

Returning to the more general expression

u 0 (ca )
cb0 (ca ) = − ,
βu 0 [cb (ca )]
we can see that cb0 (ca ) < 0, so that the indifference curve is
downward-sloping, so long as the utility function u is strictly
increasing, that is, if more is preferred to less.
Consumer Optimization: Graphical Analysis

u 0 (ca )
cb0 (ca ) = −
βu 0 [cb (ca )]
Differentiating again yields

βu 0 [cb (ca )]u 00 (ca ) − u 0 (ca )βu 00 [cb (ca )]cb0 (ca )
cb00 (ca ) = − ,
{βu 0 [cb (ca )]}2

which is positive if u is strictly increasing (more is preferred to

less) and concave (diminishing marginal utility). In this case,
the indifference curve will be convex. Again, we see how
concave functions have mathematical properties and economic
implications that we like.
Consumer Optimization: Algebraic Analysis

Graphical analysis works fine with two goods.

But what about three goods? That depends on how good an

artist you are!

And what about four or more goods? Our universe won’t

accommodate a graph like that!

But once again, calculus makes it easier!

Consumer Optimization: Algebraic Analysis

Consider a consumer who likes three goods:

Y = income
ci = consumption of goods i = 0, 1, 2
pi = price of goods i = 0, 1, 2
Suppose the consumer’s utility function is

u(c0 ) + αu(c1 ) + βu(c2 ),

where α and β are weights on goods 1 and 2 relative to good
0.
Consumer Optimization: Algebraic Analysis
The consumer chooses c0 , c1 , and c2 to maximize the utility
function

u(c0 ) + αu(c1 ) + βu(c2 ),

subject to the budget constraint

Y ≥ p0 c0 + p1 c1 + p2 c2 .
The Lagrangian for this problem is

L = u(c0 ) + αu(c1 ) + βu(c2 ) + λ(Y − p0 c0 − p1 c1 − p2 c2 ).

Consumer Optimization: Algebraic Analysis

L = u(c0 ) + αu(c1 ) + βu(c2 ) + λ(Y − p0 c0 − p1 c1 − p2 c2 ).

First-order conditions:

u 0 (c0∗ ) − λ∗ p0 = 0
αu 0 (c1∗ ) − λ∗ p1 = 0
βu 0 (c2∗ ) − λ∗ p2 = 0
Consumer Optimization: Algebraic Analysis

The first-order conditions

u 0 (c0∗ ) − λ∗ p0 = 0
αu 0 (c1∗ ) − λ∗ p1 = 0
βu 0 (c2∗ ) − λ∗ p2 = 0
imply

u 0 (c0∗ ) p0 u 0 (c0∗ ) p0 αu 0 (c1∗ ) p1

∗
= and ∗
= and ∗
= .
αu 0 (c1 ) p1 βu 0 (c2 ) p2 βu 0 (c2 ) p2
The marginal rate of substitution equals the relative prices.
Consumer Optimization: The Time Dimension

Irving Fisher (US, 1867-1947) was the first to recognize that

the basic theory of consumer decision-making could be used to
understand how to optimally allocate spending
intertemporally, that is, over time, as well as how to optimally
allocate spending across different goods in a static, or
point-in-time, analysis.
Consumer Optimization: The Time Dimension

Following Fisher, return to the case of two goods, but

reinterpret:
c0 = consumption today
c1 = consumption next year
Suppose that the consumer’s utility function is

u(c0 ) + βu(c1 ),
where β now has a more specific interpretation, as the
discount factor, a measure of patience.
Consumer Optimization: The Time Dimension

A concave utility function implies that indifference curves are

convex, so that the consumer has a preference for a
smoothness in consumption.
Consumer Optimization: The Time Dimension

Next, let
Y0 = income today
Y1 = income next year
s = amount saved (or borrowed if negative) today
r = interest rate
Consumer Optimization: The Time Dimension

Today, the consumer divides his or her income up into an

amount to be consumed and an amount to be saved:

Y0 ≥ c0 + s.
Next year, the consumer simply spends his or her income,
including interest earnings if s is positive or net of interest
expenses if s is negative:

Y1 + (1 + r )s ≥ c1 .
Consumer Optimization: The Time Dimension
Divide both sides of next year’s budget constraint by 1 + r to
get
Y1 c1
+s ≥ .
1+r 1+r
Now combine this inequality with this year’s budget constraint

Y0 ≥ c0 + s.
to get
Y1 c1
Y0 + ≥ c0 + .
1+r 1+r
Consumer Optimization: The Time Dimension

The “lifetime” budget constraint

Y1 c1
Y0 + ≥ c0 +
1+r 1+r
says that the present value of income must be sufficient to
cover the present value of consumption over the two periods.
It also shows that the “price” of consumption today relative to
the “price” of consumption next year is related to the interest
rate via
p0
= 1 + r.
p1
Consumer Optimization: The Time Dimension

The slope of the intertemporal budget constraint is −(1 + r ).

Consumer Optimization: The Time Dimension

At the optimum, the intertemporal marginal rate of

substitution equals the slope of the intertemporal budget
constraint.
Consumer Optimization: The Time Dimension

We now know the answer ahead of time: if we take an

algebraic approach to solve the consumer’s problem, we will
find that the IMRS equals the slope of the intertemporal
budget constraint:

u 0 (c0 )
= 1 + r.
βu 0 (c1 )
But let’s use calculus to derive the same result.
Consumer Optimization: The Time Dimension
The problem is to choose c0 and c1 to maximize utility

u(c0 ) + βu(c1 )
subject to the budget constraint
Y1 c1
Y0 + ≥ c0 + .
1+r 1+r
The Lagrangian is

 
Y1 c1
L = u(c0 ) + βu(c1 ) + λ Y0 + − c0 − .
1+r 1+r
Consumer Optimization: The Time Dimension
 
Y1 c1
L = u(c0 ) + βu(c1 ) + λ Y0 + − c0 − .
1+r 1+r

The first-order conditions

u 0 (c0∗ ) − λ∗ = 0
 
0 ∗ ∗ 1
βu (c1 ) − λ = 0.
1+r
lead directly to the graphical result

u 0 (c0∗ )
= 1 + r.
βu 0 (c1∗ )
Consumer Optimization: The Time Dimension

At first glance, Fisher’s model seems unrealistic, especially in

its assumption that the consumer can borrow at the same
interest rate r that he or she receives on his or her savings.

A reinterpretation of saving and borrowing in this framework,

however, can make it more applicable, at least for some
consumers.
Investment Strategies and Cash Flows

Cash Flow Cash Flow

Investment Strategy at t = 0 at t = 1

Saving −1 +(1+r)

Buying a bond −1 +(1+r)

(long position in bonds)
Investment Strategies and Cash Flows
Cash Flow Cash Flow
Investment Strategy at t = 0 at t = 1

Borrowing +1 −(1 + r )

Issuing a bond +1 −(1 + r )

Short selling a bond +1 −(1 + r )

(short position in bonds)

Selling a bond +1 −(1 + r )

(out of inventory)
Investment Strategies and Cash Flows
Cash Flow Cash Flow
Investment Strategy at t = 0 at t = 1

Buying a stock −P0s +P1s

(long position in stocks)

Short selling a stock +P0s −P1s

(short position in stocks)

Selling a stock +P0s −P1s

(out of inventory)
Consumer Optimization: The Time Dimension

Someone who already owns bonds can “borrow” by selling a

bond out of inventory. In fact, theories like Fisher’s work better
when applied to consumers who already own stocks and bonds.

Greg Mankiw and Stephen Zeldes, “The Consumption of

Stockholders and Nonstockholders,” Journal of Finance, 1991.

Annette Vissing-Jorgensen, “Limited Asset Market

Participation and the Elasticity of Intertemporal Substitution,”
Journal of Political Economy, 2002.
Consumer Optimization: The Risk Dimension

In the 1950s and 1960s, Kenneth Arrow (US, 1921-2017,

Nobel Prize 1972) and Gerard Debreu (France, 1921-2004,
Nobel Prize 1983) extended consumer theory to accommodate
risk and uncertainty.

To do so, they drew on earlier ideas developed by others, but

added important insights of their own.
Building Blocks of Arrow-Debreu Theory

1. Fisher’s (1930) intertemporal model of consumer

decision-making.
2. From probability theory: uncertainty described with
reference to “states of the world.” (Andrey Kolmogorov,
1930s).
3. Expected utility theory (John von Neumann and Oskar
Morgenstern, 1947).
4. Contingent claims – stylized financial assets – a powerful
analytic device of their own invention.
Consumer Optimization: The Risk Dimension
To be more specific about the source of risk, let’s suppose
that there are two possible outcomes for income next year,
good and bad:
Y0 = income today
Y1G = income next year in the “good” state
Y1B = income next year in the “bad” state
where the assumption Y1G > Y1B makes the “good” state
good and where
π = probability of the good state
1 − π = probability of the bad state
Consumer Optimization: The Risk Dimension

An event tree highlights randomness in income as the source

of risk.
Consumer Optimization: The Risk Dimension

Arrow and Debreu used the probabilistic idea of states of the

world to extend Irving Fisher’s work, recognizing that under
these circumstances, the consumer chooses between three
goods:
c0 = consumption today
c1G = consumption next year in the good state
c1B = consumption next year in the bad state
Consumer Optimization: The Risk Dimension

Under uncertainty, the consumer chooses consumption today

and consumption in both states next year.
Consumer Optimization: The Risk Dimension

Suppose that the consumer’s utility function is

u(c0 ) + βπu(c1G ) + β(1 − π)u(c1B ),

so that the terms involving next year’s consumption are
weighted by the probability that each state will occur as well
as by the discount factor β.
Consumer Optimization: The Risk Dimension

In probability theory, if a random variable X can take on n

possible values, X1 , X2 , . . . , Xn , with probabilities
π1 , π2 , . . . , πn , then the expected value of X is

E (X ) = π1 X1 + π2 X2 + . . . + πn Xn .
Consumer Optimization: The Risk Dimension

Hence, by assuming that the consumer’s utility function is

u(c0 ) + βπu(c1G ) + β(1 − π)u(c1B ),

we are assuming that the consumer’s seeks to maximize
expected utility

u(c0 ) + βE [u(c1 )].

Consumer Optimization: The Risk Dimension

But by writing out all three terms,

u(c0 ) + βπu(c1G ) + β(1 − π)u(c1B ),

we can see that concavity of the function u, which in the
standard microeconomic case represents a preference for
diversity, represents here a preference for smoothness in
consumption over time and across states in the future – the
consumer is risk averse in the sense that he or she does not
want consumption in the bad state to be too much different
from consumption in the good state.
Consumer Optimization: The Risk Dimension

Suppose next that today, the consumer can buy and sell
contingent claims for both future states.

A contingent claim for the good state costs q G today, and

delivers one unit of consumption next year in the good state
and zero units of consumption next year in the bad state.

A contingent claim for the bad state costs q B today, and

delivers one unit of consumption next year in the bad state
and zero units of consumption next year in the good state.
Consumer Optimization: The Risk Dimension

Payoffs for the contingent claim for the good state.

Consumer Optimization: The Risk Dimension

Payoffs for the contingent claim for the bad state.

Consumer Optimization: The Risk Dimension

Today, the consumer divides his or her income up into an

amount to be consumed and amounts used to purchase the
two contingent claims:

Y0 ≥ c0 + q G s G + q B s B ,
where s G and s B denote the number of each contingent claim
purchased or sold.

If either s G or s B is negative, the consumer can increase c0

today, but is promising to deliver goods to someone else next
year – a sophisticated kind of borrowing.
Consumer Optimization: The Risk Dimension

Next year, the consumer simply spends his or her income,

including payoffs on contingent claims:

Y1G + s G ≥ c1G
in the good state and

Y1B + s B ≥ c1B
in the bad state.
Consumer Optimization: The Risk Dimension

Y0 ≥ c0 + q G s G + q B s B
Y1G + s G ≥ c1G
Y1B + s B ≥ c1B
Multiply both sides of the second equation by q G and both
sides of the third equation by q B , Then add them all up to get
the lifetime budget constraint

Y0 + q G Y1G + q B Y1B ≥ c0 + q G c1G + q B c1B .

Consumer Optimization: The Risk Dimension

The problem is to choose c0 , c1G , and c1B to maximize

expected utility

u(c0 ) + βπu(c1G ) + β(1 − π)u(c1B ),

subject to the budget constraint

Y0 + q G Y1G + q B Y1B ≥ c0 + q G c1G + q B c1B .

This was Arrow and Debreu’s key insight: that finance is like
grocery shopping. Mathematically, making decisions over time
and under uncertainty is no different from choosing apples,
bananas, and pears!
Consumer Optimization: The Risk Dimension
The Lagrangian is

L = u(c0 ) + βπu(c1G ) + β(1 − π)u(c1B )

+λ Y0 + q G Y1G + q B Y1B − c0 − q G c1G − q B c1B ,

and the first-order conditions are

u 0 (c0∗ ) − λ∗ = 0
βπu 0 (c1G ∗ ) − λ∗ q G = 0
β(1 − π)u 0 (c1B∗ ) − λ∗ q B = 0
Consumer Optimization: The Risk Dimension
The first-order conditions

u 0 (c0∗ ) − λ∗ = 0
βπu 0 (c1G ∗ ) − λ∗ q G = 0
β(1 − π)u 0 (c1B∗ ) − λ∗ q B = 0
imply that marginal rates of substitution equal relative prices:

u 0 (c0∗ ) 1 u 0 (c0∗ ) 1
0 G∗
= G
and 0 B∗
= B
βπu (c1 ) q β(1 − π)u (c1 ) q
πu 0 (c1G ∗ ) qG
and = .
(1 − π)u 0 (c1B∗ ) qB
Consumer Optimization: The Risk Dimension

Do we really observe consumers trading in contingent claims?

Yes, if we think of financial assets as “bundles” of contingent

claims.

This insight is also Arrow and Debreu’s.

Consumer Optimization: The Risk Dimension

A “stock” is a risky asset that pays dividend d G next year in

the good state and d B next year in the bad state.

These payoffs can be replicated by buying d G contingent

claims for the good state and d B contingent claims for the bad
state.
Consumer Optimization: The Risk Dimension

Payoffs for the stock.

Consumer Optimization: The Risk Dimension

A “bond” is a safe asset that pays off one next year in the
good state and one next year in the bad state.

These payoffs can be replicated by buying one contingent claim

for the good state and one contingent claim for the bad state.
Consumer Optimization: The Risk Dimension

Payoffs for the bond.

Consumer Optimization: The Risk Dimension

If we start with knowledge of the contingent claims prices q G

and q B , then we can infer that the stock must sell today for

q stock = q G d G + q B d B .

Since if the stock cost more than the equivalent bundle of

contingent claims, traders could make profits for sure by short
selling the stock and buying the contingent claims; and if the
stock cost less than the equivalent bundle of contingent
claims, traders could make profits for sure by buying the stock
and selling the contingent claims.
Consumer Optimization: The Risk Dimension

“Pricing” the stock.

Consumer Optimization: The Risk Dimension
Likewise, if we start with knowledge of the contingent claims
prices q G and q B , then we can infer that the bond must sell
today for
q bond = q G + q B .
Since the bond pays off one for sure next year, the interest
rate, defined as the return on the risk-free bond, is
1 1
1+r = = .
q bond qG + qB
The bond price relates to the interest rate via
1
q bond = .
1+r
Consumer Optimization: The Risk Dimension

Pricing the bond.

Consumer Optimization: The Risk Dimension

We’ve already seen how contingent claims can be used to

replicate the stock and the bond.

Now let’s see how the stock and the bond can be used to
replicate the contingent claims.
Consumer Optimization: The Risk Dimension

Consider buying s shares of stock and b bonds, in order to

replicate the contingent claim for the good state.

In the good state, the payoffs should be

sd G + b = 1
and in the bad state, the payoffs should be

sd B + b = 0
since the contingent claim pays off one in the good state and
zero in the bad state.
Consumer Optimization: The Risk Dimension

To replicate the contingent claim for the good state:

sd G + b = 1
sd B + b = 0 ⇒ b = −sd B
Substitute the second equation into the first to solve for

1 −d B
s= and b =
dG − dB dG − dB
Since s and b are of opposite sign, this requires going “long”
one asset and “short” the other.
Consumer Optimization: The Risk Dimension

To replicate the contingent claim for the good state:

1 −d B
s= and b =
dG − dB dG − dB
If we know the prices q stock and q bond of the stock and bond,
we can infer that in the absence of arbitrage, the claim for the
good state would have price

q stock − d B q bond
q G = q stock s + q bond b = .
dG − dB
Consumer Optimization: The Risk Dimension

Consider buying s shares of stock and b bonds, in order to

replicate the contingent claim for the bad state.

In the good state, the payoffs should be

sd G + b = 0
and in the bad state, the payoffs should be

sd B + b = 1
since the contingent claim pays off one in the bad state and
zero in the good state.
Consumer Optimization: The Risk Dimension

To replicate the contingent claim for the bad state:

sd G + b = 0 ⇒ b = −sd G
sd B + b = 1
Substitute the first equation into the second to solve for

−1 dG
s= and b =
dG − dB dG − dB
Once again, this requires going long one asset and short the
other.
Consumer Optimization: The Risk Dimension

To replicate the contingent claim for the bad state:

−1 dG
s= and b =
dG − dB dG − dB
Once again, if we know the prices q stock and q bond of the stock
and bond, we can infer that in the absence of arbitrage, the
claim for the bad state would have price

d G q bond − q stock
q B = q stock s + q bond b = .
dG − dB
Consumer Optimization: The Risk Dimension

What makes it possible to go back and forth between traded

assets, like stocks and bonds, and contingent claims is that
there are the same number of traded assets as there are
possible states of the world next year.

More generally, asset markets are complete if there are as

many assets (with linearly independent payoffs) as there are
states next year.
Consumer Optimization: The Risk Dimension

If asset markets are complete, then we can use the prices of

traded assets to infer the prices of contingent claims.

Then we can use the contingent claims prices to infer the price
of any newly-introduced asset.
General Equilibrium

An allocation of resources is Pareto optimal if it is impossible

to reallocate those resources without making at least one
consumer worse off.

A competitive equilibrium is an allocation of resources and a

set of prices such that, at those prices: (i) each consumer is
maximizing utility subject to his or her budget constraint and
(ii) the supply of each good equal the demand for each good.

The two welfare theorems of economics link optimal and

equilibrium allocations.
Optimal Allocations

In an economy with two consumers, 1 and 2, and two goods, a

and b, the key properties of Pareto optimal allocations can be
illustrated using an Edgeworth box.
ca1 = 1’s consumption of good a
cb1 = 1’s consumption of good b
ca2 = 2’s consumption of good a
cb2 = 2’s consumption of good b
Optimal Allocations

The Edgeworth box contains the entire set of feasible

allocations.
Optimal Allocations

Both consumers prefer allocations in the green region to A.

Optimal Allocations

Both consumers prefer B to A, but still there are allocations

that are even more strongly preferred.
Optimal Allocations

At C, there is no way to make one consumer better off without

making the other worse off. C is Pareto optimal.
Optimal Allocations

There are many Pareto optimal allocations, but each is

characterized by the tangency of the two consumers’
indifference curves.
Optimal Allocations

Note that Pareto optimality is a welfare criterion that accounts

for efficiency but not equity: an allocation may be Pareto
optimal even though it provides most of the goods to one
consumer.

But since the slope of the indifference curves is measured by

the marginal rate of substitution, the mathematical condition
associated with all Pareto optimal allocations is

1 2
MRSa,b = MRSa,b . (PO)
Optimal Allocations

Suppose that consumer 1 has utility function

u(ca1 ) + αu(cb1 )

and consumer 2 has utility function

v (ca2 ) + βv (cb2 ).
Optimal Allocations
Consider a benevolent “social planner,” who divides Ya units
of good a and Yb units of good b up between the two
consumers, subject to the resource constraints

Ya ≥ ca1 + ca2

and
Yb ≥ cb1 + cb2 ,
so as to maximize a weighted average of their utilities:

θ[u(ca1 ) + αu(cb1 )] + (1 − θ)[v (ca2 ) + βv (cb2 )],

where 1 > θ > 0.

Optimal Allocations
Since there are two constraints, the Lagrangian for the social
planner’s problem requires two Lagrange multipliers:

L = θ[u(ca1 ) + αu(cb1 )] + (1 − θ)[v (ca2 ) + βv (cb2 )]

+λa (Ya − ca1 − ca2 ) + λb (Yb − cb1 − cb2 ).

The first-order conditions are:

θu 0 (ca1 ) − λa = 0

θαu 0 (cb1 ) − λb = 0
(1 − θ)v 0 (ca2 ) − λa = 0
(1 − θ)βv 0 (cb2 ) − λb = 0.
Optimal Allocations
The first-order conditions

θu 0 (ca1 ) − λa = 0

θαu 0 (cb1 ) − λb = 0
(1 − θ)v 0 (ca2 ) − λa = 0
(1 − θ)βv 0 (cb2 ) − λb = 0.
imply that
u 0 (ca1 ) λa v 0 (ca2 )
= = ,
αu 0 (cb1 ) λb βv 0 (cb2 )
a restatement of (PO) that must hold for any value of θ.
Equilibrium Allocations

Now let’s see what happens when markets, instead of a social

planner, allocate resources:
Ya1 = consumer 1’s endowment of good a
Yb1 = consumer 1’s endowment of good b
Ya2 = consumer 2’s endowment of good a
Yb2 = consumer 2’s endowment of good b
pa = price of good a
pb = price of good b
Equilibrium Allocations

Consumer 1 chooses ca1 and cb1 to maximize utility

u(ca1 ) + αu(cb1 )

subject to the budget constraint

pa Ya1 + pb Yb1 ≥ pa ca1 + pb cb1 ,

taking the prices pa and pb as given.

Equilibrium Allocations
The Lagrangian for consumer 1’s problem is

L = u(ca1 ) + αu(cb1 ) + λ1 (pa Ya1 + pb Yb1 − pa ca1 − pb cb1 ).

The first-order conditions

u 0 (ca1 ) − λ1 pa = 0

αu 0 (cb1 ) − λ1 pb = 0
imply that
u 0 (ca1 ) pa
0 1
= . (CE-1)
αu (cb ) pb
Equilibrium Allocations

Similarly, consumer 2 chooses ca2 and cb2 to maximize utility

v (ca2 ) + βv (cb2 )

subject to the budget constraint

pa Ya2 + pb Yb2 ≥ pa ca2 + pb cb2 ,

taking the prices pa and pb as given.

Equilibrium Allocations
The Lagrangian for consumer 2’s problem is

L = v (ca2 ) + βv (cb2 ) + λ2 (pa Ya2 + pb Yb2 − pa ca2 − pb cb2 ).

The first-order conditions

v 0 (ca2 ) − λ2 pa = 0

βv 0 (cb2 ) − λ2 pb = 0
imply that
v 0 (ca2 ) pa
0 2
= . (CE-2)
βv (cb ) pb
Equilibrium Allocations
Hence, in any competitive equilibrium

u 0 (ca1 ) pa
0 1
= . (CE-1)
αu (cb ) pb

and
v 0 (ca2 ) pa
0 2
= (CE-2)
βv (cb ) pb
must hold, so that

u 0 (ca1 ) pa v 0 (ca2 )
= = .
αu 0 (cb1 ) pb βv 0 (cb2 )
Equilibrium Allocations

There will be different equilibrium allocations associated with

different patterns for the endowments Ya1 , Yb1 , Ya2 , and Yb2 .

In addition, different equilibrium allocations may require

different prices pa and pb to equate the supply and demand of
each good.

But all equilibrium allocations must satisfy

1 pa 2
MRSa,b = = MRSa,b . (CE)
pb
Equilibrium Allocations

The Pareto optimal allocation C is supported in a competitive

equilibrium with prices pa and pb , and the equilibrium
allocation C is Pareto optimal.
General Equilibrium

The coincidence between (PO) and (CE) underlies results that

extend Adam Smith’s (Scotland, 1723-1790) notion of an
“invisible hand” that guides self-interested individuals to
choose resource allocations that are Pareto optimal.

First Welfare Theorem of Economics The resource allocation

from a competitive equilibrium is Pareto optimal.

Second Welfare Theorem of Economics A Pareto optimal

resource allocation can be supported in a competitive
equilibrium.