Lecture Notes on General Equilibrium and Financial Frictions

Daniel Pesch and Dimitrios P. Tsomocos

September 25, 2020

2
Contents

1 Security Markets 11
1.1 Asset Span . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.1.1 Complete and incomplete markets . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2 Answering the questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Pricing Methods 17
2.1 Arbitrage Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Definition of arbitrage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Law of one price (LOOP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.3 Replicating portfolios and redundant assets . . . . . . . . . . . . . . . . . . . 20
2.1.4 State pricing and the Fundamental Theorem of Finance . . . . . . . . . . . . 24
2.2 Equilibrium Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 Arrow-Debreu economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.2 Asset Markets Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2.3 Arrow-Debreu and complete markets equivalence . . . . . . . . . . . . . . . . 34
2.2.4 Lucas Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.4.1 Breeden’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.4.2 Consumption CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.4.3 Equity premium puzzle . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2.4.4 General equilibrium and incomplete markets . . . . . . . . . . . . . 42
2.2.4.5 Weil (1992) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.5 Risk-neutral probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3
4 CONTENTS

2.2.6 Risk-neutral valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3 General Equilibrium with Incomplete Markets 47

3.1 Fundamental Equation of Asset Pricing . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 GEI-GEI* Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Asset Span Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4 COMPLETE MARKETS THEOREM . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5 Modigliani-Miller Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.5.1 Implication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.2 Interpretation 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.3 Interpretation 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Martingale Pricing Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.7 Linear Pricing Corollary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.8 Covariance Pricing Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.8.1 Security Market Line Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4 Dynamic Completion of Markets 69

5 Options 73
5.1 Graphical Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.1.1 Profits from a European call option . . . . . . . . . . . . . . . . . . . . . . . 74
5.1.2 Profits from a European put option . . . . . . . . . . . . . . . . . . . . . . . 75
5.1.3 Payoffs from a stock and a risk-free bond . . . . . . . . . . . . . . . . . . . . 75
5.1.4 Relationships amongst elemental securities . . . . . . . . . . . . . . . . . . . 76
5.2 Bounds of a call option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Put-Call Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4 Complete Markets and Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5 American versus European Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.6 Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.6.1 Hedging sale of an option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.6.2 Replicating an option using bond and stock . . . . . . . . . . . . . . . . . . . 84
5.6.3 Binomial model of option pricing Cox et al. [1979] . . . . . . . . . . . . . . . 85
CONTENTS 5

5.6.3.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.6.3.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.6.3.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6.3.4 Intuition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.7 Black-Scholes Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

6 Portfolio Theory 95
6.1 Portfolio Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2 Markowitz Portfolios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.2.1 Case 1: Perfect positive correlation (ρ12 = 1) . . . . . . . . . . . . . . . . . . 99
6.2.2 Case 2: Imperfectly correlated assets (−1 < ρ12 < 1) . . . . . . . . . . . . . . 99
6.2.3 Case 3: Perfect negative correlation (ρ12 = −1) . . . . . . . . . . . . . . . . . 101
6.2.4 Case 4: One riskless and one risky asset . . . . . . . . . . . . . . . . . . . . . 102
6.2.5 Case 5: n risky assets (imperfectly correlated) . . . . . . . . . . . . . . . . . . 103
6.2.6 Case 6: n risky assets and a risk-free asset . . . . . . . . . . . . . . . . . . . . 105
6.2.7 Minimum variance frontier . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.2.8 Two-Fund Separation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.2.9 Capital markets line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.3 CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.3.1 Security market line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.3.2 Zero-beta CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3.3 Pareto optimality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

7 CAPM GEI 115

7.1 CAPM Efficiency Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.1.1 CAPM Mutual Fund Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.1.2 CAPM Security Market Line . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.1.3 CAPM Invariant Pricing Theorem . . . . . . . . . . . . . . . . . . . . . . . . 121
7.1.4 Multiple Commodity CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

8 APT 125
8.1 Factor Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6 CONTENTS

8.2 Single-Factor APT & CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

8.3 APT Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.4 APT formally . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.4.1 One factor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.4.2 Multi-factor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.5 APT IN GEI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.5.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.5.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
8.5.3 APT Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
List of Figures

4.1 Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.1 Payoff from a call option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.2 Payoff from a put option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3 Profits from a stock and a bond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Relationships amongst elemental securities . . . . . . . . . . . . . . . . . . . . . . . . 77
5.5 Mitigating downward risk with a call option . . . . . . . . . . . . . . . . . . . . . . . 78
5.6 Development of the share price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.7 Development of the share price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.8 Trinomial tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.1 Two perfectly positively correlated assets . . . . . . . . . . . . . . . . . . . . . . . . 100

6.2 Two imperfectly correlated assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3 Two perfectly negatively correlated assets . . . . . . . . . . . . . . . . . . . . . . . . 103
6.4 One riskless and one risky asset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.5 More than two risky assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.6 n risky assets and a risk-free asset . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.7 CML and asset j . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.8 Security market line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.9 Zero-beta CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

7
8 LIST OF FIGURES
List of Tables

2.1 From: Mehra and Prescott (1985), Summary statistics on US historical data for the
period 1889-1978. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

9
10 LIST OF TABLES
Chapter 1

Security Markets

Securities allow agents to transfer wealth from today to tomorrow and vice versa. They can also
function as an insurance mechanism. We formalise security markets below.

We assume two time periods, today (t = 0) and tomorrow (t = 1), and that the state of the
world today is known. In contrast, there are S ≡ {1, 2, ..., S} different states of the world that could
be realised at t = 1. We denote the cardinality of the set S as S.

A security j promises certain payments in each of these S states of the world (eg, a stock, a
bond etc.). Its price is denoted by qj ∈ R and its payoff at state s is denoted by ajs ∈ R. The payoff
vector aj ∈ RS for a security j is defined as

! "T
aj ≡ aj1 ··· ajS

Assume that there are J securities in the economy and each one is defined by its payoff vectors
a for all j ∈ J (J denotes the set of all securities). The payoff matrix of the economy A ∈ RS×J
j

is defined as

11
12 CHAPTER 1. SECURITY MARKETS

! "
A≡ a1 ··· aJ
! "T
≡ a1 ··· aS
# &
a11 ··· aJ1
$ '
$ .. .. .. '
≡$ . . . '
% (
a1S ··· aJS

where as ∈ RJ is a vector of payoffs from all securities in a particular state. If we hold an

amount θj ∈ R of each security, then the total holdings, θ ∈ RJ , are given by the vector

! "T
θ= θ1 ··· θJ

and the payoff in state s ∈ S is given by

J
)
as θ = ajs θj
j=1

and the payoff matrix is given by

# &
a1 θ
$ '
$ .. '
Aθ = $ . '
% (
aS θ
S×1

Looking at the portfolio holdings, two questions arise:

1. Do we need to hold an amount of every available security to construct the payoff we desire
for all states?

2. Can we construct any payoff profile via trade in available securities?

In order to answer these, we need to calculate the asset span.

1.1. ASSET SPAN 13

1.1 Asset Span

Intuitively, the asset span is the ”number” (or dimension) of different payoff profiles one can con-
struct via trade in the available assets.

Example 1.1. Assume that there are three states of the world and two assets with payoffs a1 =
(1, 0, 0)T and a2 = (0, 1, 1)T . Now, assume that one holds an amount θ1 of the first asset and θ2
of the second. The payoff matrix of this portfolio is given by

# & # &
1 0 # & θ1
$ ' θ1 $ '
$ ' (=$ '
Aθ = $ 0 1 ' % $ θ2 '
% ( θ2 % (
0 1 θ2

Notice that any portfolio with these assets would have the same payoffs in states 2 and 3. This
implies that, for example, payoff profile (1, 1, 2) cannot be achieved through trade in assets 1 and
2. We say, in this case, that the asset span is R2 since we can only achieve payout profile in the 2
-dimensional space.

Example 1.2. Suppose again that there are three states of the world and three assets with payoffs
T
a1 = (1, 0, 0)T , a2 = (0, 1, 0)T , and a3 = (0, 0, 1)T The payoff matrix of portfolio θ = (θ1 , θ2 , θ3 ) of
each asset is given by
# &# & # &
1 0 0 θ1 θ1
$ '$ ' $ '
$ '$ ' $ '
Aθ = $ 0 1 0 ' $ θ2 ' = $ θ2 '
% (% ( % (
0 0 1 θ3 θ3

In this case, the asset span is R3 since we can achieve any payout profile in the 3 -dimensional
space.

Definition 1.3. Asset span is the set of different portfolio payoffs that can be achieved in the
future via trade in the existing assets. Mathematically, asset span of payoff matrix A, denoted M,
is given by

* +
M ≡ z ∈ RS : z = Aθ for some θ ∈ RJ
14 CHAPTER 1. SECURITY MARKETS

1.1.1 Complete and incomplete markets

Markets are said to be complete if and only if payoffs of any newly issued asset can be constructed
through buy and sell orders of the existing assets; ie, if we can construct a portfolio of existing
assets that will have the same payoff profile as the newly issued one.

Definition 1.4. Markets are complete if M = RS

This means that there are S securities with independent payoff profiles; ie, the payoff matrix has

full rank.
Markets are said to be incomplete if payoffs of not every newly introduced assets can be replicated
by a portfolio of existing assets.

Definition 1.5. Markets are incomplete if M ⊂ RS

This means that there are less than S securities with independent payoff profiles; ie, the payoff

matrix has rank less than S.

1.2 Answering the questions

We attempt to answer the two questions posed at the beginning of the chapter.

1. Do we need to hold an amount of every available security to construct the payoff we desire
for all states?
It is clear from Example 1.2, that the need to hold every available security to achieve some
desired payoff in all states depends on the desired level of payoff and the structure of the
available security. For instance, in the said example, if one wishes to achieve a payoff of zero
in state 3, then he would hold zero amount of the third asset. Similarly, the need would also
depend on the payoffs of the existing assets in the economy.

2. Can we construct any payoff profile via trade in available securities?

If markets are complete (ie, asset span is equal to RS ), then one could construct securities to
have any desired payoffs. Adding any more securities will be redundant. However, if markets
are incomplete (ie, asset space is a subset of RS ), then not all payoffs can be constructed using
existing assets. Addition of securities with independent payoffs from the existing ones will
1.2. ANSWERING THE QUESTIONS 15

enable one to achieve greater range of payoffs. Once the markets become complete following
additions of securities, then any payoff can be achieved.
16 CHAPTER 1. SECURITY MARKETS
Chapter 2

Pricing Methods

This concerns the following questions:

1. How should we price securities in complete and incomplete markets?

2. Are there any differences between traded and newly issues securities?

3. If markets are complete, why are so many new products being issued and traded?

There are, generally, two methods of pricing:

ARBITRAGE PRICING has no reference to preferences, endowments, or production functions.

The only assumption required is monotonicity, that agents prefer more to less. This tells us how to
price using the concept of replicating portfolio.

EQUILIBRIUM PRICING represents an analytical solution of a general equilibrium maximi-

sation problem with specific reference to preferences, endowments, and production functions. This
tells how prices of traded securities are formed and how we should price newly issued ones.

2.1 Arbitrage Pricing

2.1.1 Definition of arbitrage

Loosely speaking, an arbitrage is a certain opportunity to obtain profit. One is said to exploit an
arbitrage opportunity when one makes a certain profit starting with zero initial capital. To do so,

17
18 CHAPTER 2. PRICING METHODS

one would form a portfolio of existing traded assets that has a positive payoff tomorrow but costs
nothing or even has negative price (ie, the trader receives money today).
We begin by recalling the definition of the product order in multi-dimension: Suppose x ∈ RS ,
then

x≥0 ⇔ xi ≥ 0 ∀i ∈ {1, . . . , S}
x≫0 ⇔ xi > 0 ∀i ∈ {1, . . . , S}
x>0 ⇔ xi ≥ 0 ∀i ∈ {1, . . . , S}

with at least one strict inequality.

Let q ∈ RJ be the price vector of the J assets so that

J
)
qθ = qj θj
j=1

is the (scalar) cost of portfolio θ . Recall that the payoff from portfolio θ is given by (vector) Aθ,
where A ∈ RS×J is the payoff matrix of the J assets in the economy.

Definition 2.1. Arbitrage exists if at least one of the following expression holds with strict in-
equality:

Aθ ≥ 0,

qθ ≤ 0.

If the first inequality holds strictly, then it says that the payoff tomorrow is positive (in some
states). If the latter inequality holds strictly, then the cost of obtaining the portfolio θ is negative.

Definition 2.2. A strong arbitrage exists when one can form a portfolio that has a positive or zero
payment tomorrow at every state but gives a positive payment today; ie,

Aθ ≥ 0, qθ < 0.

Example 2.3. Let a1 = (1, 1)T and a2 = (1, 2)T with prices q1 = q2 = 1. Then portfolio θ =
(−1, 1)T is an arbitrage but not a strong arbitrage. To see this, note
2.1. ARBITRAGE PRICING 19

# &# & # &

1 1 −1 0
Aθ = % (% (=% (>0
1 2 1 2
# &
! " −1
qθ = 1 1 % (=0
1

Let a1 = (−1, 2, 0)T and a2 = (2, 2, −1)T and θ be the portfolio. The payoff from the portfolio
is

# & # &
−1 2 # & −θ1 + 2θ2
$ ' θ1 $ '
$ ' (=$ '
Aθ = $ 2 2 '% $ 2θ1 + 2θ2 '
% ( θ2 % (
0 −1 −θ2

and the it has a positive payoff in all states if

−θ1 + 2θ2 ≥ 0

θ1 + θ2 ≥ 0

−θ2 ≥ 0

For all these inequalities to be satisfied, it must be the case that θ = 0. Thus, there is no arbitrage
for any price.
Note that arbitrage pricing assumes that prices of existing securities exist in the market through
trading and does not say how these prices are formed initially, rather, it only assumes that agents
in the economy have monotonic preferences. Arbitrage opportunities cannot consistently exist
since agents would take advantage of them and thus move prices until arbitrage opportunities are
eliminated.

2.1.2 Law of one price (LOOP)

, ∈ RJ . If Aθ = Aθ,
Definition 2.4. ((Law of one price). Take any two portfolios, θ, θ , then
,
qθ = qθ.

This says that every portfolio with the same payoff profile should be worth the same. If this
were not the case then one would sell the expensive securities and buy the cheap securities with
20 CHAPTER 2. PRICING METHODS

equivalent payoff profiles since we assume agents prefers more to less. As a result, the price of the
cheaper security increases and the other will decrease and hence prices are equated; ie, all arbitrage
opportunities would disappear eventually.

Example 2.5. Assume two securities with prices q1 and q2 . Without loss of generality, we assume
q1 > q2 and
# &
a1 a1
A=% (.
a2 a2

Now consider the two portfolios, θ 1 = (1, 0)T and θ 2 = (0, 1)T , then
# &# & # & # &# &
a1 a1 1 a1 a1 a1 0
Aθ 1 = % (% (=% (=% (% ( = Aθ 2
a2 a2 0 a2 a2 a2 1

and
# & # &
! " 1 ! " 0
qθ 1 = q1 q2 % ( = q1 > q2 = q1 q2 % ( = qθ 2
0 1

Therefore the law of one price does not hold. To see that this represents the existence of arbitrage
opportunities, consider the portfolio θ 3 = (−1, 1)T . Its payoff is then
# &# & # &
a1 a1 −1 0
Aθ 3 = % (% (=% (=0
a2 a2 1 0

and
# &
! " −1
qθ 3 = q1 q2 % ( = −q1 + q2 < 0
1

since q1 > q2 by assumption. Thus, there is an arbitrage opportunity.

Remark 2.6. If the law of one price does not hold, then there is an arbitrage opportunity.

2.1.3 Replicating portfolios and redundant assets

If the prices of the already traded securities satisfy the law of one price and are positive when they
promise positive payoffs, then arbitrage opportunities do not exist. Since arbitrage opportunities
cannot persist in the market, the above two conditions should hold for traded securities.
2.1. ARBITRAGE PRICING 21

Definition 2.7. (Redundant assets). An asset is said to be redundant if one could form a portfolio
of existing assets to replicate its payoffs in every state of the world.

Note that once we find the replicating portfolio, its price can then be deduced using the law of
one price. In order to find such a portfolio (if it exists) for an asset j ∗ with payoff profile a∗ , one
needs to find the solution (θ) to
∗
Aθ = aj .

If the system has a solution, say θ ∗ , then the asset j ∗ is redundant and its price should be given
by qθ ∗ , by the law of one price.

Example 2.8. Assume two traded assets with payoffs a1 = (3, 2)T and a2 = (2, 1)T . Their prices
are q1 = 2.5 and q2 = 1.5, respectively. Now suppose that there is another asset, with payoffs
T
a3 = (1, 1)T . This asset is said to be redundant if there exists a portfolio θ = (θ1 , θ2 ) that solves

# &# & # &

3 2 θ1 1
Aθ = % (% (=% (
2 1 θ2 1

since θ = (1, −1)T is a solution to the above, we realise that an asset with payoffs (1, 1) is redundant.
By the law of one price, the price of the third asset is
# &
! " 1
qθ = 2.5 1.5 % (=1
−1

Example 2.9. Assume that there are three traded assets 1, 2, and 3 with the payoff matrix given
by
# &
3 2 0
$ '
$ '
A=$ 2 1 1 '
% (
1 1 1

and q = (q1 , q2 , q3 ) represent the prices. These three assets can replicate an asset with payoffs
(2, 3, 0.5) since there is a solution to the problem
# &# & # &
3 2 0 θ1 2
$ '$ ' $ '
$ '$ ' $ '
Aθ = $ 2 1 1 ' $ θ2 ' = $ 3 '
% (% ( % (
1 1 1 θ3 0.5
22 CHAPTER 2. PRICING METHODS

given by θ = (1.5, −0.75, −0.25)′ . In addition, by the law of one price, the price of the forth
asset is given by
# &
1.5
! "$ '
$ '
qθ = q1 q2 q3 $ −0.75 ' = 1.5q1 − 0.75q2 − 0.25q3
% (
−0.25

Suppose we now have three assets with the payoff matrix given by
# &
3 2 1
$ '
$ '
A=$ 2 1 1 '
% (
1 1 0

Could we replicate an asset with payoffs (1, 1, 1)? If it is possible then there must exist a solution
to
# &# & # &
3 2 1 θ1 1
$ '$ ' $ '
$ '$ ' $ '
Aθ = $ 2 1 1 ' $ θ2 ' = $ 1 '
% (% ( % (
1 1 0 θ3 1

Solving yields
# &−1 # &
3 2 1 1
$ ' $ '
$ ' $ '
θ=$ 2 1 1 ' $ 1 '
% ( % (
1 1 0 1

However, the first matrix is not invertible since its determinant is given by
- -
- - - - - - - -
- 3 2 1 - - - - - - -
- - - 1 1 - - 2 1 - - 2 1 -
- - - - - - - -
- 2 1 1 - = 3- - − 2- - + 1- -
- - - 1 0 - - 1 0 - - 1 1 -
- -
- 1 1 0 -

= −3 + 2 + 1 = 0

In other words, replicating the payoff (1, 1, 1) using the said three assets is impossible. This implies
that the fourth asset is not redundant. Note that, in this case, we cannot price it via arbitrage
arguments.
2.1. ARBITRAGE PRICING 23

Derivatives are financial products whose payoffs is a function of a payoff of an underlying financial
asset, say XT , traded in the market. A forward contract is a financial derivative whose payoff is
calculated as XT − F0,T when you buy (go long on) the contract, or, F0,T − XT when you sell (go
short on) the contract.
An investor that goes long on a forward at t = 0 enters an agreement to buy at t = T for a
predetermined price of F0,T . Alternatively, an investor that goes short on a forward at t = 0 enters
an agreement to sell a stock at t = T for a predetermined price of F0,T . The cost of entering into a
forward contract today is zero.
Define the current exchange rate between the pound and the euro as S0 . meaning that 1 GBP
is worth S0 EUR. Suppose an investor goes long on a currency contract, entering an agreement to
buy the foreign currency (euros) at a predetermined exchange rate F0,T at t = T. The domestic
interest rate for the period T is denoted r0,T denominated in pounds and the foreign interest rate
∗
for the period T is denoted r0,T denominated in euros.

Proposition 2.10. (Covered interest rate parity). The relationship between the current exchange
rate, the forward rate, the domestic interest rate, and the foreign interest rate is given by
∗
1 + r0,T
F0,T = S0
1 + r0,T

Proof. The proof uses the no-arbitrage argument. Assume the following strategy:

1. borrow 1/ (1 + r0,T ) in the domestic market,

2. buy spot foreign currency at S0

∗
3. invest them in the foreign market at a rate r0,T and

4. sell a currency forward.

By the no-arbitrage argument, the payoff for this strategy at t = T has to be equal or less than the
amount you have to repay, which is 1. Hence,

1 . ∗
/ 1
S0 1 + r0,T ≤1 (2.1)
1 + r0,T F0,T

Now consider an alternative strategy:

. ∗
/
1. borrow 1/ 1 + r0,T in the foreign market,
24 CHAPTER 2. PRICING METHODS

2. buy spot domestic currency at 1/S0 ,

3. invest them in the domestic market at a rate r0,T , and

4. buy a currency forward.

The payoff for this strategy at t = T in foreign currency, again, has to be equal or less than the
amount you have to repay, which is 1, by the no-arbitrage argument. We thus have
1 1
∗ (1 + r0,T ) F0,T ≤ 1 (2.2)
1 + r0,T S0

From (2.1), we obtain that

∗
1 + r0,T
S0 ≤ F0,T
1 + r0,T
and from (2.2), we have
∗
1 + r0,T
F0,T ≤ S0
1 + r0,T
Combining the two above yields the covered interest parity condition:
∗ ∗ ∗
1 + r0,T 1 + r0,T 1 + r0,T
S0 ≤ F0,T ≤ S0 ⇔ F0,T = S0
1 + r0,T 1 + r0,T 1 + r0,T

2.1.4 State pricing and the Fundamental Theorem of Finance

A security which pays out in only one state, say s, is called an s th Arrow security. Denote the
payoffs of such security as

es = (0, . . . , 1, . . . , 0)T ∈ RS

with 1 in the s th place. The state price for state s is given by πs , which is equivalent to the price
of the s th Arrow security. It is the price that an investor has to pay today to receive a payment
of 1 in a specific state tomorrow. We denote the vector of state prices as

π = (π1 , . . . , πS ) ∈ RS
! "T ! "T
Assume a security j with payoffs aj = aj1 , . . . , ajS and a portfolio θ = aj1 , . . . , ajS of Arrow
securities; ie, we hold ajs of the sth Arrow security. Note that the payoff matrix A of a full set of
2.1. ARBITRAGE PRICING 25

Arrow securities can be represented by an S × S identity matrix (denoted Is). Hence, payoff of the
portfolio is

# &
aj1
$ '
$ .. '
Aθ = IS θ = θ = $ . ' = aj
% (
ajS

and the price of the portfolio is

S
)
πθ = πaj = πs ajs
s=1

Notice that the security j and the portfolio have the same payoffs, thus by the law of one price,
they should have the same price; ie,

S
)
qj = πaj = πs ajs
s=1

T
More generally, given a payoff profile z = (z1 , . . . , zS ) ∈ RS , its price, given state price π, is given
by

q(z) = πz

Example 2.11. Assume that there are three states of the world and securities, where the payoff
matrix is given by

# &
1 1 1
$ '
$ '
A=$ 2 1 1 '
% (
3 1 2

and the prices of the three securities as q = (8, 4, 5). Note that rank(A) = 3 = S and thus the
markets are complete. The fundamental pricing equation is q = πA, which, in this case, is given
26 CHAPTER 2. PRICING METHODS

by
# &
1 1 1
! " ! "$ '
$ '
8 4 5 = π1 π2 $ 2 1
π3 1 '
% (
3 1 2
# &−1
1 1 1
! " ! "$ '
$ '
⇔ π1 π2 π3 = 8 4 5 $ 2 1 1 '
% (
3 1 2
# &
−1 1 0
! "$ '
$ '
= 8 4 5 $ 1 1 −1 '
% (
1 −2 1
! "
= 1 2 1 .

Note that state prices are unique for every state and thus one can price any security and its price
will be unique too. For example, the price of a security with payoffs (5, 8, 10) will be
# & # &
5 5
$ ' ! "$ '
$ ' $ '
q = π$ 8 ' = 1 2 1 $ 8 ' = 31
% ( % (
10 10

Now consider the case when there are three states of the world, three securities with a payoff matrix
given by
# &
1 1 0
$ '
$ '
A=$ 2 1 1 '
% (
3 1 2

and prices given by q = (8, 3, 5). Note that rank(A) = 2 < S and thus markets are incomplete. If
we use the fundamental pricing equation again,
# &
1 1 0
! " ! "$ '
$ '
8 3 5 = π1 π2 π3 $ 2 1 1 '
% (
3 1 2
2.1. ARBITRAGE PRICING 27

or, written differently,

8 = π1 + 2π2 + 3π3

3 = π 1 + π2 + π 3

5 = π2 + 2π3

Using the first two equations to eliminate π1 , we obtain

5 = π2 + 2π3

It is clear that there will be infinite number of solutions satisfying1

(π1 , π2 , π3 ) = (2 − π3 , 5 − 2π3 , π3 )

In the previous example, if the state prices have to be positive, then π3 ∈ [2, 2.5]. But does it
have to be positive?

To answer this question, we appeal to the following Lemma, stated without proof.

Lemma 2.12. (Stiemke’s lemma). There does not exist a ∈ Rm such that

aY ≥ 0, ay ≤ 0

with at least one strict inequality, if and only if there exists b ∈ Rn such that

y = Yb, b ≫ 0

where y = (m × 1), Y = (m × n), b = (n × 1), a = (1 × m).

1 Recall that given a system Ax = b where there are n equations in k unknowns, if

1. rank (Ab ) = rank(A) + 1, then there is no solution;

2. rank (Ab ) = rank(A) = k, then there is a unique solution; and

3. rank (Ab ) = rank(A) < k, then there is an infinite number of solutions;

where Ab is an augmented matrix, Ab = [A : b]. In this case, we first transpose q, π, and A to obtain

q T = AT π T ⇒
! $ ! $! $
8 1 2 3 π1
" % " %" %
" % " %" %
" 3 % = " 1 1 1 %" π2 %
# & # &# &
5 0 1 2 π3
' ( ' (
where we realise that rank AT = rank AT b = 2 < 3 = k. Thus, this system has an infinite number of solutions.
28 CHAPTER 2. PRICING METHODS

Remark 2.13. Stated in a more suitable form (by taking the transpose):

∃b ∈ Rn++ : y = bY

if and only if

∄a ∈ Rm : Ya ≥ 0, ya ≤ 0

with at least one strict inequality, where now y = (1 × m), Y = (n × m), b = (1 × n), a = (m × 1).

Lemma 2.14. (Farkas’ lemma). There does not exist a ∈ Rm such that

aY ≥ 0, ay < 0

if and only if there exists b ∈ Rn such that

y = Yb, b ≥ 0

where y = (m × 1), Y = (m × n), b = (n × 1), a = (1 × m).

Remark 2.15. Stated in a more suitable form (by taking the transpose):

∃b ∈ Rn+ : y = bY

if and only if

∄a ∈ Rm : Ya ≥ 0, ya < 0

with at least one strict inequality, where now y = (1 × m), Y = (n × m), b = (1 × n), a = (m × 1).

Theorem 2.16. (Fundamental Theorem of Finance). Security prices exclude arbitrage if and only
if there exist strictly positive state prices.

Proof. Immediate from Stiemke’s lemma by setting

y = q, Y = A, b = π, a = θ

and m = J, n = S.

Corollary 2.17. Security prices exclude strong arbitrage if and only if there exist weakly positive
state prices.
2.2. EQUILIBRIUM PRICING 29

Proof. Immediate from Farkas’ lemma by setting

y = q, Y = A, b = π, a = θ

and m = J, n = S.

Remark 2.18. (Arrow securities and replicating portfolios). Since an Arrow security pays 1 in
only one state, we can choose the amount of any of the s Arrow securities to replace the payoff
structure of the asset we want to price. In complete markets, we can construct every Arrow security
via appropriate trade in existing assets (in linear algebra terms, we can construct the basis of the
payoff matrix). For some assets, i.e., those that are redundant, this argument goes through in
incomplete markets as well. However, if we wish to price an asset that is not redundant, we cannot
find a replicating portfolio of Arrow securities and, therefore, cannot price it in incomplete markets.

2.2 Equilibrium Pricing

Until now, prices were taken to be given but where do the asset prices come from?

2.2.1 Arrow-Debreu economy

The Arrow-Debreu model has two time periods, t ∈ T = {0, 1} and S different states of the world
(S ≡ {1, . . . , S}), representing the different ways in which exogenous uncertainty might be resolved
between periods 0 and 1. Consider an environment with L commodities and H agents (the set of
commodities is L and the set of agents is H). The key idea is that of a contingent commodity: a
commodity whose delivery is conditional (contingent) on the realised state of the world. We assume
an exchange economy in which agents are endowed with commodities at both t = 0 and in every
state at t = 1, and choose to trade today (t = 0) the contingent commodities. In other words, in
an Arrow-Debreu economy, assets/securities are the contingent commodities.

Definition 2.19. (Contingent commodity). A contingent commodity sl is a contract that promises

the delivery of one unit of commodity l ∈ L if state s ∈ {1, . . . , S}.

The endowment of an agent h ∈ H is given by

eh = (e1 , . . . , eS ) ∈ RL LS
+ × R+
30 CHAPTER 2. PRICING METHODS

where es = (es1 , . . . , esL ) ∈ RL , and his consumption plan is denoted

x h ∈ RL LS
+ × R+

The prices of the contingent commodities are given by

π ∈ RL × RLS

Although deliveries are contingent on the state of the world, purchase of assets occurs today (by
assumption) and thus, every agent faces the following budget set:
* . h / +
Bh (π) ≡ x ∈ RL LS
+ × R+ : π · x − e
h
≤ 0 ∀h ∈ H

Definition 2.20. An Arrow-Debreu economy is given by

* +
u h , eh h∈H
0 * + 1
Definition 2.21. An Arrow-Debreu equilibrium is a π, xh h∈H where π ∈ RL × RSL and
x h ∈ RL
+ × R+ .
SL

2 2
1. Contingent commodity markets clear: h∈H xh = h∈H eh ;

2. Budget constraint is satisfied: xh ∈ Bh (π) for all h ∈ H; and

. /
3. Maximises agent’s utility: ∀x ∈ Bh (π), uh xh ≥ uh (x) for all h ∈ H.

Remark 2.22. The Arrow-Debreu equilibrium is the solution to

max uh (x) ∀h ∈ H
x∈Bh (π)

The Lagrangian is given by

3 S )
L
4
) . /
h
L = u (x) − λ πsl xsl − ehsl ,
s=0 l=1

with the first-order conditions

∂L ∂uh (x)
=0⇔ = λπsl ∀s ∈ S, ∀l ∈ L
∂xsl ∂xsl
Thus, the price of contingent commodity {sl} in terms of the price of the numeraire commodity
{01} is
πsl ∂uh (x) /∂xsl
= .
π01 ∂uh (xx ) /∂x01
2.2. EQUILIBRIUM PRICING 31

2.2.2 Asset Markets Economy

In the Arrow-Debreu model of general equilibrium, all commodities are traded at once, no matter
when they are consumed, or under what state of nature. Therefore, all consumers face only one
budget constraint.
The distinguishing feature of the financial markets model is that consumers face a multiplicity
of budget constraints, at different times and under different states of nature. To transfer wealth
among budget constraints, consumers must hold assets.
In addition, retrading at t = 1 is not necessary for the Arrow-Debreu economy, even if spot
markets for all the L commodities were operating at t = 1 and in each state, trade in these markets
2
would not occur. In asset markets, however, retrading at t = 1 is important and it assumes that
agents correctly anticipate today the price that will prevail tomorrow using rational expectations.
Note that agents need to be more computationally capable in the asset markets.
The Arrow-Debreu equilibrium is equivalent to an asset equilibrium with complete markets
(shown formally later). In the former, there are L × (S + 1) markets. In a financial market, there
are markets for commodities today and tomorrow, 2L . In addition, there are S securities but only
one state will be realised tomorrow. Hence, the financial market has 2L + S markets in total.
Here, we concentrate on the case when there is only one good in the economy; ie, L = 1. This
. /
is referred to as a financial economy. Agent h ∈ H′ s consumption plan is denoted xh = xh0 , xh ∈
. /
R+ × RS+ , where x0 is his consumption at t = 0, and xh = xh1 , . . . , xhS is his consumption in period
1 in every state. The agent derives utility uh (xh0 ) from his consumption plan. He is endowed with
. /
eh = eh0 , eh defined in a similar manner and can transfer wealth from today to tomorrow and
across the S states via trade in J securities where A is a S × J payoff matrix of J securities (each
row corresponding to payoffs from each security). Thus, he forms a portfolio θh ∈ RJ of the J
securities. The agents objective is to maximise his utility subject to his budget constraints. There
are budget constraints for each different time period and different state of nature:

* +
Bh (q) = (x, θ) ∈ RS+1
+ × RJ : x0 + qθ ≤ eh0 , x ≤ eh + Aθ

Note that there are S + 1 number of budget constraints.

2 If it were optimal for agents to trade in period t = 1, then this could not have been an Arrow-Debreu equilibrium
because this would mean a violation of the last condition.
32 CHAPTER 2. PRICING METHODS

Definition 2.23. A financial economy is given by

0 * + 1
uh , eh h∈H , A .
5 0 1 6
Definition 2.24. A financial equilibrium is a q, xh , θ h where q ∈ RJ , xh ∈ R+ × RS+ ,
h∈H
and θh ∈ RJ :
2 2
1. Goods markets clear: h∈H xh = h∈H eh ;
2
2. Securities markets clear: θh = 0
h∈H
! "
3. Budget constraints are satisfied: xh , θ h ∈ Bh (q) for all h ∈ H; and
. /
4. Utilities are maximised: ∀(x, θ) ∈ Bh (q), uh xh ≥ uh (x) for all h ∈ H

Remark 2.25. Assume uh : RS+1

+ ,→ R is strictly increasing. The financial equilibrium is then the
solution to

max uh (x)
(θ,x)∈Bh (q)

3
subject to the two market clearing conditions. We may rewrite the budget constraints as

x0 + qθ ≤ eh0
J
)
⇔ x0 + qj θj ≤ eh0
j=1

x ≤ eh + Aθ
J
)
⇔xs ≤ ehs + ajs θjh ∀s ∈ S
j=1

The Lagrangian is then given by

# & # &
) S
) J
)
L = uh (x) − µ0 %xh0 − eh0 + qj θjh ( − µs %xhs − ehs − ajs θjh ( .
j∈J s=1 j=1

The first-order conditions applicable to interior solutions are

∂L ∂uh
=0⇔ = µs ∀s = 0, 1, . . . , S
∂xs ∂xs
)S
∂L
= 0 ⇔ µ 0 q j = µs ajs ∀j = 1, . . . J
∂θjh s=1
3 Note that there are S + 1 Walras laws, so we can normalise consumption in all states and periods
2.2. EQUILIBRIUM PRICING 33

Eliminating the Lagrange multiplier in the second expression using the first, we obtain that
S
) ∂uh /∂xs
qj = ajs
s=1
∂uh /∂x0
∂u !
h
∂uh ∂uh
"
= ∂x1 ··· ∂xS
aj
∂x0

This also implies that

! " ∂uh /∂x
q= q1 ··· qJ = A,
∂uh /∂x0
. /
where ∂uh /∂x = ∂uh /∂x1 , . . . , ∂uh /∂xS .

Definition 2.26. (Market clearing). Market clearing implies that both the goods market and the
securities markets clear. Respectively, these are implied by
) )
xhs ≤ ehs ∀s = 0, 1, . . . , S,
h∈H h∈H

and
)
θjh = 0∀j = 1, . . . , J.
h∈H

Remark 2.27. A financial equilibrium exists under very general assumptions and the observed
prices should be interpreted as equilibrium prices.

Remark 2.28. If markets are complete, we can replace the A matrix with an S -dimensional
identity matrix, Is , and the vector of prices q with the vector of state prices π . Thus, from the
first-order conditions, the state price for state s is given by

∂uh /∂xhs
πs = , ∀s ∈ S.
∂uh /∂xh0

Remark 2.29. If utilities are strictly increasing, then in a financial equilibrium, there is no arbi-
trage.
Intuition here is that with a strictly increasing utility function, the budget constraints are
binding. If there was an arbitrage opportunity, an agent could increase his utility without affecting
his budget constraint in a negative way. Thus, the budget constraints would not be binding anymore
- a contradiction. Therefore, price will adjust until equilibrium is reached.
34 CHAPTER 2. PRICING METHODS

Remark 2.30. Redundant securities will never be traded in a financial equilibrium. However, they
will be priced by the arbitrage condition.
For any redundant security, an agent can form a portfolio of the other securities to replicate
its payoff. To optimise, he does not need to take a position in a redundant security. Thus, when
solving for a financial equilibrium, we should not include redundant securities in the maximisation
problem.

2.2.3 Arrow-Debreu and complete markets equivalence

The financial market equilibrium with a full set of Arrow securities collapses to the Arrow-Debreu
4
equilibrium. This is a general result for more than one commodity and is known as the Complete
Markets Theorem (3.25). Here, we prove the special case for the financial economy with a full set
of Arrow securities.
Denote xhs , ehs , as realised consumption and endowments. Recall the budget set within a financial
economy:

! " 0 1
Bh q, xh , eh , θ h = xh0 + qθh ≤ eh0 , x ≤ eh + Aθ h .

Since we assume that there is a full set of Arrow securities, the payoff matrix A is equivalent to
the identity matrix Is . That is,
! " 0 h
1
Bh q, xh , eh , θ h = xh0 + qθ h ≤ eh0 , xh ≤ eh + IS θ̂ .

since we assume the economy has only Arrow securities, then payoffs from all J securities in state
s is given by as = is ≡ (0, . . . , 1, . . . , 0), where 1 is in the s th place. We may now write
! " 0 1
Bh q, xh , eh , θ h = xh0 + qθh ≤ eh0 , xhs ≤ ehs + is θ h ∀s ∈ S

Notice that, since the market is complete such that S = J,

is θ h = θjh = θsh ,
J
) S
)
qθh = qj θjh = qs θsh ,
j=1 s=1
4 An Arrow-Debreu economy is one in which there are only contingent goods to be traded (essentially Arrow
securities) within complete markets. A financial economy is more general in that the securities do not have to be
Arrow securities and the markets may be incomplete.
2.2. EQUILIBRIUM PRICING 35

and therefore we have

7 8
! " . / S
)
h h h h
B q, x , e , θ = xh0 − eh0 + qs θsh ≤ 0, xhs ≤ ehs + θsh ∀s ∈S .
s=1

which implies that

7 8
! " . / S
) . /
h h h h
B q, x , e , θ = xh0 − eh0 + qs xhs − ehs ≤0 .
s=1
* . / +
= π · x h − eh ≤ 0

= Bh (π),

where π = (1, q) ∈ R1+S .

A sufficient condition for Pareto optimality (given an interior solution) is that the marginal rates
of substitution across the states of the world are equal for all agents; ie, for all A, B ∈ H,

∂uA /∂xA
s ∂uB /∂xB
s
M RS A = M RS B ⇔ = ∀s ∈ S.
∂uA /∂xA
0 ∂uB /∂xB
0

Proposition 2.31. The Arrow-Debreu equilibrium is Pareto optimal.

Proof. This follows from the First Welfare Theorem (think of contingent goods as ordinary con-
sumption goods).

When markets are complete, the asset market equilibrium is also Pareto optimal since state
prices are unique and thus the MRS of agents are equated. In contrast, when markets are incomplete
the financial equilibrium is constrained optimal (where L = 1) and asset markets equilibrium is
constrained suboptimal (where L > 1).

2.2.4 Lucas Model

Lucas (1978)’s model has been widely used in the asset pricing and macroeconomics literature; it is
a model with a representative agent general equilibrium model with one commodity. In this model,
there is a continuum of identical-in-all-aspects agents (ie, representative agent), the same uncertain
environment as before, and J productive assets, which Lucas calls “trees” that payout in terms of
the one commodity at t = 1. Every agent is endowed at t = 0, the same amount of these J assets
and faces the following decision problem: how much of his holding to sell and how much to keep
36 CHAPTER 2. PRICING METHODS

and collect the payoffs (dividends) in the following period. Finally, the objective here is to derive
the equilibrium prices of these J assets at every point in time.
We assume an expected utility representation such that every agent’s utility function is given
by

u (x0 ) + δE0 u(x)

where x0 is his consumption today, x ≡ (x1 , . . . , xS ) is his contingent consumption (in t = 1), δ is
the time-discount factor and E is the expectations operator. Note that

E0 u(x) ≡ γ1 u (x1 ) + · · · + γS u (xS )

2S
where γs is the probability that state will be s tomorrow, and i=1 γi = 1. We denote by qj the
! "T
price of asset j at t = 0, aj ≡ aj1 , . . . , ajS as its payoff vector, and φj as the amount of the asset
that the representative agent chooses to sell at t = 0. We normalise the initial endowment of every
asset to 1 and assume that the initial endowment of the agent is e0 . In addition, without loss of
generality, we assume that

e ≡ (e1 , . . . , eS ) = 01×S ;

ie, we assume that the agents have zero endowments in period t = 1. The budget set of the agent
can be written as
* +
B(q) ≡ x ∈ RS+1
+ : x0 − e0 ≤ qφ, xs ≤ as (1J×1 − φ) ∀s ∈ S ,

where xs is realised consumption in state s, φ ≡ (φ1 , . . . , φJ ) ∈ RJ , and as ∈ RJ is the payoffs of

all J securities in state s.
The agent’s problem is to solve

max u (x0 ) + δE0 u(xs ).

(θ,x)∈B(q)

We assume strictly concave utilities so that the budget constraint are binding at the optimum. The
Lagrangian is
3 4 # &
S
) J
)
L = u(x0 ) + δ γs u (xs ) − µ0 %x0 − e0 − qj φ j (
s=1 j=1
9 ;
S
) J
)
− µs :xs − ajs (1 − φj )<
s=1 j=1
2.2. EQUILIBRIUM PRICING 37

The relevant first-order conditions for interior solution(s) are

u′ (x0 ) = µ0 ,
∂u (xs )
δγs = µs ∀s = 1, . . . , S
∂xs
)S
µ 0 qj = µs ajs ∀j = 1, . . . , J.
s=1

Eliminating the Lagrange multipliers in the last expression yields the Euler equation for asset prices:

3 S 4 = ′ >
1 ) ∂u (xs ) j u (xs ) j
qj = ′ δγs a ≡ δE0 ′ a . (2.3)
u (x0 ) s=1 ∂xs s u (x0 )

The ratio u′ (xs ) /u′ (x0 ) is called the marginal rate of substitution of consumption between today
and state s tomorrow and can also be called the pricing kernel or the stochastic discount factor.
The price of an asset does not only depend on how high or low its payoffs are, but also on how
“valuable” are these payoffs in the states that they occur, which is measured by the MRS for that
state.

Expanding the expectations operator and assuming that δ = 1, the Euler equation for asset
prices can be written as

u′ (x1 ) j u′ (xS ) j
qj = γ 1 a1 + · · · + γS ′ a
′
u (x0 ) u (x0 ) S
= π1 aj1 + · · · + πS ajS ,

where γs is the subjective probability belief that state s will occur and πs is the Arrow price for
state s. Note that the lower the income in states, the higher the price of the insurance contract
(Arrow security) for that state.

In equilibrium, there can be no trade given the representative agent framework so that φ = 0J×1 ,
which implies that x0 = e0 and

J
)
xs = ajs ∀s = 1, . . . , S.
j=1
38 CHAPTER 2. PRICING METHODS

2.2.4.1 Breeden’s formula

Assume that there is a riskless asset with payoff 1 in the economy and denote its price and return
by qf and Rf respectively.5 Note
1 f
Rf = , a = 1S×1 .
qf
We may apply the Euler equation (2.3) to obtain
= ′ >
1 u (x) δ
= δE0 ′ = ′ E0 [u′ (x)] . (2.4)
Rf u (x0 ) u (x0 )
Recall the definition of covariance between random variables X and Y :

Cov(X, Y ) = E(XY ) − E(X)E(Y ).

Then
? @ . / ? @
E u′ (x)aj = E [u′ (x)] E aj + Cov u′ (x), aj

(Note that a’ and x are both random variables since the agent does not know which state will be
realised in t = 1. ) Rewriting (2.3) using above, we obtain the Euler equation for any risky asset as
E [u′ (x1 )] E (aj ) + Cov [u′ (x1 ) , aj ]
qj = δ . (2.5)
u′ (x0 )
Combining equations (2.4) and (2.5), we obtain the Breeden’s formula:
? @
1 . j/ 1 Cov u′ (x), aj
qj = E a +
Rf Rf E [u′ (x)]
Notice how the Breeden’s formula decomposes the price of a risky asset into two components:

• expected income, discounted by the riskless rate; and

• a premium for risk.

2.2.4.2 Consumption CAPM

Assume a linear marginal utility u′ (x) = a − bx.6 Then, we may rewrite the Breeden’s formula as
. /
. j/ Cov x, aj
E a − qj Rf = −b .
a − bE(x)
5A riskless asset pays out an equal amount in every state.
6 Take the following quadratic utility function
(a − bx)2
u(x) = −
2b
2.2. EQUILIBRIUM PRICING 39

Normalising the price of the risky asset j to 1 gives

. /
. j/ Cov x, aj
E a − Rf = −b .
a − bE(x)
7
Expressing above in terms of rates of return, we obtain

Cov (x, rj )
E (rj ) − rf = −b .
a − bE(x)

Applying this to the market portfolio x (denoted with subscript m), we obtain

Cov (x, rm )
E (rm ) − rf = −b .
a − bE(x)

Dividing previous two equations, we obtain the CAPM equation

E (rj ) − rf Cov (x, rj )

=
E (rm ) − rf Cov (x, rm )
Cov (x, rj )
⇔ E (rj ) − rf = [E (rm ) − rf ]
Cov (x, rm )
Cov (rm , rj )
= [E (rm ) − rf ]
Var (rm )

Note that the last uses the fact that agent’s date-1 consumption and the market return are perfectly
correlated- Corr (x, rm ) = 1− since all agents have the same endowments (of zero) in period t = 1,
the consumption in t = 1 is equal to the return the agents obtain from trading. Since rate of return
is a function of the return from trading, we realise that x and rm are perfectly correlated.
Then

u′ (x) = a − bx

u′′ (x) = −b

Hence, the coefficient of absolute risk aversion is given by

u′′ (x) b
− =
u′ (x) a − bx
Recall that risk tolerance, denoted as ρ, is the inverse of above, so that
b
ρ= −x
a
We see that both b and a affects the measure of risk aversion in quadratic utility function.
Moreover, note that quadratic utility functions are mean-variance utilities.
' (
7 Rates of return of an asset j with expected payoff E aj and a price q is given by
j
' (
E aj
E (1 + rj ) =
qj
40 CHAPTER 2. PRICING METHODS

2.2.4.3 Equity premium puzzle

In the US, historical data shows that real returns to treasury bonds are about 1% per year while
return from a broad stock market index approximately 7% per year. General equilibrium theories of
asset pricing cannot account for such large difference between the two return rates. This discrepancy
between theory and historical data is referred to as the equity risk premium puzzle.
Mehra and Prescott (1985) were the first to address this issue using US historical data for the
period 18891978. Summary statistics for their dataset is given in Table 1.

US 1889-1978 Mean Variance

Stock index (total return,rm ) 0.0698 0.0274

Real interest rate (rf ) 0.0080 0.0032
Growth rate of consumption (s) 0.0183 0.0013

Table 2.1: From: Mehra and Prescott (1985), Summary statistics on US historical data for the
period 1889-1978.

It is also estimated that the coefficient of risk-aversion, ρ, in a CRRA utility function is between
1 − 4.
Consider the Euler equation as in (2.3) while assuming a CRRA utility function u(x) = x1−ρ /(1−
ρ)· In addition, assume that x follows a log-normal distribution such that

x = am = x 0 e g

. /
where g ∼ N 0, σ 2 , x0 is current-period consumption, and x = am is both future consumption
and the stock market. Now, recall that

2
E (eg ) = eµ+σ /2

. /
when g ∼ N µ, σ 2 , which implies that

2
E0 (x) = x0 E (eg ) = x0 eσ /2
2.2. EQUILIBRIUM PRICING 41

Applying the particularities to the formula for the price of risky stock gives
. /
E [u′ (x)am ] E x−ρ x
qm = δ =δ
u′ (x0 ) x−ρ
A B 0
−ρ
E (x0 eg ) x0 eg A B
=δ −ρ = δx0 E eg(1−ρ)
x
=0 >
σ2
= δx0 exp (1 − ρ)2
2
The rate of return is
E (am ) x0 E (eg )
1 + rm = = ? 2@
qm δx0 exp (1 − ρ)2 σ2
2 = 2 >
1 x0 eσ /2 1 σ 2σ
2
= ? @ = exp − (1 − ρ)
δ x0 exp (1 − ρ)2 σ22 δ 2 2
= >
1 . / σ2
= exp 2ρ − ρ2
δ 2
The risk-free rate of return is simply
af
1 + rf = ⇒ af = (1 + rf ) qf
qf
where
A B
= ′
> −ρ
E (x0 eg ) (1 + rf ) qf
u (x) f
qf = δE a =δ
u′ (x0 ) x−ρ
0
C 2 2D
ρ σ
= δqf (1 + rf ) exp
2
C 2 2D
1 ρ σ
⇒ 1 + rf = exp −
δ 2
8
Combining the expressions for the market and the risk-free asset, we obtain
A. / B
1 2 σ2
1 + rm δ exp 2ρ − ρ 2
= ! 2 2"
1 + rf 1
exp − ρ σ
δ 2
. 2
/
= exp ρσ ≈ 1 + ρσ 2

Given that ρ ∈ [1, 4] and σ 2 = 0.13% from the table, then

rm − rf ≈ ρσ 2 ∈ [0.13%, 0.52%]

It is obvious from above that in order to explain the difference, agents should be very risk averse,
which is not observed in practicethus, the puzzle.
8 We use the MacLaurin series expansion for the exponential function: ex = 1 + x + x2 /2! + x3 /3!.
42 CHAPTER 2. PRICING METHODS

2.2.4.4 General equilibrium and incomplete markets

When does a general equilibrium with multiple agents collapse to a representative agent model?
Note that we can always construct an economy with an arbitrary representative agent and solve
for prices. Yet an arbitrary representative agent is not very useful, since he has no relation to the
data of the original heterogeneous agent economy. The conditions under which we can construct a
representative agent from the data of the original economy are:

1. complete markets,

2. expected utility representation,

3. homogeneity of degree one utility functions,

4. identical utilities, and

5. no transaction costs.

2.2.4.5 Weil (1992)

Assume agent α’s endowment is given by

eα = (e0 , e1 , e2 ) ,

where the first element corresponds to t = 0, the second and third refers to t = 0 with states 1 and
2, respectively. In contrast, agent β’s endowment is assumed to be

eβ = (e0 , e2 , e1 ) .

If utilities functions are identical, states are equiprobable, and markets are complete, then both
agents will consume x = (e1 + e2 ) /2 in both states, and the price of a safe asset with payoff 1 will
be given by
u′ (e1 + e2 ) /2
qfCM = γ ,
u′ (e0 )
However, if markets are incomplete (eg, there is only a safe bond), no trade will take place, and the
price of the safe bond will be
[u′ (e1 ) + u′ (e2 )] /2
qfIM = γ .
u′ (e0 )
2.2. EQUILIBRIUM PRICING 43

In addition, if u’ is convex, then

C D
e1 + e2 u′ (e1 ) + u′ (e2 )
u′ < ⇒ qfCM < qfIM
2 2
In other words, the safe rate in incomplete markets is lower (for the same utility function) than in
complete markets. This is because of precautionary savings: Undiversifiable risk calls for higher
savings, hence the savings curve is pushed up. For a given savings rate, a model with incom-
plete markets, which would require higher savings, will therefore predict a lower safe interest rate.
Similarly, the return on equity is overstated in complete market models.
What is interesting for the equity premium puzzle is, however, the difference between the overes-
timations of the safe rate and of the equity return. Weil (1992) shows that: If u exhibits decreasing
absolute risk aversion (DARA) and decreasing absolute prudence (DAP) (ie, −u”’/u” is decreas-
ing), then a calibrator who is not aware of the existence of undiversifiable labour income risk will
understate the magnitude of the equity premium; ie, overstate the equity premium puzzle.

2.2.5 Risk-neutral probabilities

State prices are the prices of Arrow securities, which are artificial constructions that pay 1 in only
one state world. They are the solution to the following system of linear equations

q1 = π1 a11 + · · · + πS a1S
..
. ,

qJ = π1 aJ1 + · · · + πS aJS

where J is the number of already traded (ie, there is a price) securities, S is the number of states
of the world, qj is the price of security j ∈ J, ajs is the payoff of security j in the s th state of the
world, and πs is the price of the Arrow security that pays out in the sth state of the world.
The price of any security j is
qj = π1 aj1 + · · · + πS ajS . (2.6)

If the security is risk-free and has payoff of 1 in every state, then its price is
S
)
qf = πS
s=1
) S
1
⇒ = πS ,
1 + rf s=1
44 CHAPTER 2. PRICING METHODS

where qf is the price of the risk-free asset, which is equal to its expected payoff ( of 1) divided by
its return (1 + rf ).
2S
Dividing and multiplying (2.6) by s=1 πs , we obtain

S
) π1 aj1 + · · · + πS ajS
qj = πS 2S
s=1 s=1 πS
1 ! "
= µ1 aj1 + · · · + µS ajS
1 + rf
S
1 )
= µs ajS
1 + rf s=1

where

πs
µs = 2S
s=1 πs

2S
Observe that µs ≥ 0 (since state prices are positive if there is no arbitrage) and s=1 µs = 1.
These two conditions are sufficient to define a probability measure and we call µs the risk-neutral
9
probability for state sthey are simply rescaled state prices. Risk-neutral probabilities are also
called Martingale probabilities (or measure). When we price securities in continuous time (eg,
Black and Scholes methodology), we do not derive the whole Martingale measure. Since we assume
that that variables are normally distributed (or another well-defined distribution), we only need to
calculate the mean and variance under the Martingale measure. In contrast, under discrete time
and state space, there is no need to assume a specific distribution. The fundamental theorem of
finance can also be defined in terms of risk- neutral probabilities:

Theorem 2.32. Security prices exclude arbitrage if and only if there exist strictly positive risk-
neutral probabilities.

Corollary 2.33. Security prices exclude strong arbitrage if and only if there exist positive risk-
neutral probabilities.

9 First two axioms of probability theory by [Kolmogorov, 1933]; the third additivity holds trivially.
2.2. EQUILIBRIUM PRICING 45

2.2.6 Risk-neutral valuation

The fundamental relation that gives the price of any security is

1 . /
qj = E µ aj
1 + rf
1 A B
= µ1 aj1 + · · · + µS ajS
1 + rf

where s denotes the state that is realised tomorrow. The no-arbitrage pricing is translated into
calculating the value of a security as the discounted (with the risk-free rate) expected value of its
payoffs with respect to the risk-neutral probabilities. To calculate the risk-neutral probabilities, we
write down the above equation for all existing securities and solve the system of equations. We
then use these probabilities to price any redundant asset through the same equation.
Under risk-neutral valuation, an investor/trader is pricing the security as if he were risk-
neutral.10 This is due to the fact that he discounted the expected value of the payoffs with the
risk-free rate, so he does not account for any risk premium in the discount rate. However, risk is
still captured in the risk-neutral probabilities. The reason that, under the Martingale measure, the
investor prices the security as if he was risk neutral lies within arbitrage arguments: Recall that the
derivation of state prices is based on a replicating portfolio strategy. Risk neutral probabilities are
nothing more than rescaled state prices, thus, using them to price securities implies an underlying
replicating strategy and hedging of excess risk.

10 This will be more easily seen when we consider options and the discount rate of individual investors.
46 CHAPTER 2. PRICING METHODS
Chapter 3

General Equilibrium with

Incomplete Markets

Consider a world in which there are two time periods (t ∈ {0, 1}) and S + 1 states of nature, S
of which are resolved between period zero and period one. Utility for any agent h ∈ H depends
(S+1)L
on consumption of L commodities at the date-events such that uh : R+ ,→ R . Agents have
(S+1)L
endowments e ∈ h
R+ . For now, the only assumption we require on the economy is that for all
h ∈ H, uh is weakly monotonic in each state s ∈ S : ie, if xsl = xtl for all s, t ∈ S with s ∕= t and
for some l ∈ L, and xsl > xtl for s = t and for all l ∈ L, (if consumption is strictly bigger in one
state across all commodities but are otherwise the same) then, uh (x) > uh (y). 1

The difference between the GEI model and the Arrow-Debreu economy lies in how trade is
conducted; in the Arrow-Debreu model, one can exchange commodities at time 0 for any one of
the state contingent commodities in the future; however, in the GEI model, one is restricted to
exchanging commodities at time zero for assets.

An asset is represented by a vector a ∈ RSL denoting the promised payoff of each commodity
in each state in period 1 (note that the payoff can be positive or negative). The collection of all J

1 It’s weakly monotonic since uh (x) > uh (y) requires that consumption in state s is strictly large for x than y for
all commodities, rather than some commodities in state s being larger.

47
48 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

assets is given by the matrix A ∈ RSL×J , which may be written in a few ways:
! "
A= a1 ... aJ aj ∈ RSL
# &
A1
$ '
$ .. '
=$ . ' As ∈ RL×J
% (
AS
# &
a11 ··· aI1
$ '
$ .. .. .. ' j
=$ . . . ' as ∈ R L
% (
a1S ··· aIS

Agents may transfer wealth between periods and between states of nature only by holding assets.
We assume the initial endowments of assets to be zero and that agents wishing to make promises
conforming to asset j issue asset j at time zero; ie, they sell the asset by taking a negative position
on it.
Let commodity prices be p = (p0 , p1 , . . . , pS ) ∈ R(S+1)L and asset prices be q = (q1 , . . . , qJ ) ∈
RJ . We now introduce some useful notations.

Definition 3.1. Given x = (x0 , x1 , . . . , xS ) ∈ R(S+1)L , x = (x1 , . . . , xS ) ∈ RSL ie, x represents

the consumption at t = 1.

Definition 3.2. Given p, x ∈ RSL

T
p□x ≡ (p1 · x1 , . . . , pS · xS ) ∈ RS

ie, each row in p□x represent the cost of consumption goods in a particular state.

. /
Definition 3.3. Given SL × J matrix A = a1 , . . . , aI

. /
p□A ≡ p□a1 , . . . , p□al S×J
# &
p1 · a11 · · · p1 · aJ1
$ '
$ .. .. .. '
=$ . . . ' ;
% (
pS · aS · · · pS · aJS
1
S×J

ie, each column in p□A represent the monetary payoff from a particular asset.
 49

Given the notation above, the GEI budget set for agent h is written

(S+1)L . /
Bh (p, q) = {(x, θ) ∈ R+ × RJ : p0 · x0 − eh0 + qθ ≤ 0
. /
p□ x − eh ≤ (p□A)θ}

To clarify, note that

L
) J
. / . / )
p0 · x0 − eh0 + qθ = p0l x0l − eh0l + qj θj
l=1 j=1
. h
/
p□ x − e ≤ (p□A)θ
# . / & # &# &
p1 · x1 − eh1 p1 · a11 · · · p1 · aI1 θ1
$ ' $ '$ '
$ .. ' $ .. .. .. '$ .. '
⇔$ . '≤$ . . . '$ . '
% ( % (% (
. h
/ 1 J
p S · x S − eS p S · aS · · · p S · a S θJ
L J
E L F
) . / ) ) j
⇔ psl xsl − ehsl ≤ psl asl θj ∀s ∈ S
l=1 j=1 l=1

Definition 3.4. GEI economy is defined by

0* + 1
u h , eh h∈H
,A

where uh : RL
+ × R+ ,→ R represents the utility for consumption plans of agent h, e ∈ R+ × R+
LS h L LS

represents the endowment of agent h, and A ∈ RSL×J represents the payoffs of the assets j ∈ J in
the L commodities across the S states in period 1.
0* 1 5 0 1 6
+ h
Definition 3.5. A GEI equilibrium of GEI economy u h , eh h∈H
, A is a p, q, x h
, θ
h∈H
where xh ∈ RL
+ × R+ ∀h ∈ H, p ∈ R × R
SL L SL
, q ∈ RJ , and θ h ∈ RJ :
2 2
1. Commodities markets clear: h∈H xh = h∈H eh
2
2. Securities markets clear: θh = 0J×1
h∈H
0 1
3. Budget constraints are satisfied: xh , θ h = Bh (p, q) for all h ∈ H; and
. /
4. Utilities are maximised: ∀{x, θ} ∈ Bh (p, q), uh xh ≥ uh (x) for all h ∈ H.

Remark 3.6. A few observations can be made about the GEI model:

• There are separate budget constraints for each date-event.

50 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

• Expenditures on commodities and assets can exceed receipts from the sale of commodity
endowments and the issue of new asset in a state only if the difference can be made up from
the receipts form asset holdings.

2
– As a result, there will be not 1, but S + 1 Walras Laws.

– Moreover, there are S + 1 independent homogeneity properties for demand in prices; ie,
3
doubling the price vector ps for any s does not affect market clearing.

• Once the equilibrium prices p and q are given, they matter only through the amount of value
they promise in each state.

• In the budget set, we allowed every individual access to the same markets, the GEI model
includes situations where some agents do not care for consumption in state 0 but they still
are able to trade assets at that time.

• Weak monotonicity implies that, in GEI equilibrium, ps ≥ 0, and for each s ∈ S, there exists
l with psl > 0.

Remark 3.7. Consider the case where there is only one financial security that delivers one dollar in
both states 1 and 2. Agents can save by purchasing the asset, giving up wealth at t = 0 in exchange
for money at t = 1. Similarly, they can borrow by selling the asset at date t = 0, collecting money
qθ, and later paying back θ dollars in each state, where q is the price of the asset. Trade in securities
which are not in positive supply cannot take place unless some agents are allowed to go short The
2 This implies that at equilibrium, if
H
) H
)
xh
sl = eh
sl , ∀s, l = 1, . . . , L − 1
h=1 h=1

then the same necessarily holds for all s and l = L.

3 This can be easily seen if we recall the budget set:

(S+1)L
Bh (p, q) = {(x, θ) ∈ R+ × RJ :
* +
p 0 · x 0 − eh
0 + qθ ≤ 0
* +
p□ x − eh ≤ (p□A)θ}

The budget constraints relevant to t = 1 is the inequality in the second line, and since p is on both sides of the
inequality, the implication is obvious.
3.1. FUNDAMENTAL EQUATION OF ASSET PRICING 51

definition of GEI explicitly allows for all possible short sales4 by lettings agents choose θ ∈ RJ ,
instead of restricting their choices to RJ+ .

Remark 3.8. In GEI equilibrium, all agents optimise with respect to budget sets defined by the
same prices. This has several noteworthy implications:

• It implies that all agents have perfect (conditional) expectations, in keeping with the rational
expectations hypothesis (that agents are correct in their conditional forecasts of ps )

• All agents keep all their promises to deliver on assets. There is no bankruptcy or default.

• As in the Arrow-Debreu model, there is implicitly a perfect credit market from the beginning
of a date-event to its end. Agents can keep their promises to deliver goods even if their
endowments do not include those particular goods, provided they can afford to trade for the
promised goods on the spot markets.

• Agents know the spot prices and the state of nature when they carry out their spot market
trade.

3.1 Fundamental Equation of Asset Pricing

Definition 3.9. We say that a sequence (. . . , xt , xt+1 , . . .) of random variables all defined on the
same probability space {Ω, µ} satisfies the Martingale property, if for µ, almost all ω,

Eµ (xt+1 | . . . xt (ω)) = xt (ω).

This property states that past behaviour of the variable allows one to predict the future that
variable. Note that the agents must agree on the state space. The Martingale property of asset
prices cannot possibly hold true without some correction for inflation. In a multiple commodity
world, that means depicting an index of prices.

Definition 3.10. We say that a commodity bundle x∗s is a numeraire for state s if ps · x∗s = 1
4 Short sales is when one sells a promise to give something. In this case, an agent borrowing by selling the risk-free
asset at t = 0, is said to be short-selling.
52 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

Before stating the next theorem, it is useful to state what the condition for the existence of
arbitrage is in the context of GEI.

Definition 3.11. Arbitrage exists if

qθ ≤ 0, (p□A)θ ≥ 0

holds with at least one strict inequality.

5 0 1 6
h h
Theorem 3.12. (Fundamental Equation of Asset Pricing). Let p, q, x , θ be a GEI
h∈H
0* + 1
equilibrium of GEI economy uh , eh h∈H , A .

1. Fix any bundle x∗s for s ∈ S. Then, there exists a strictly positive probabilities over states µ,
and a discount rate δ > 0 such that
S C D
qj 1 ) ps · ajs
= µ s ∀j ∈ J
p0 · x∗0 1 + δ s=1 ps · x∗s

5 0 1 6
′ ′ h
2. There is another GEI equilibrium p ,q , x ,θ h
for which x∗s is a numeraire for
h∈H
every state s = 0, 1, . . . , S such that
S
1 ). /
qj′ = µs p′s · ajs ∀j ∈ J (3.1)
1 + δ s=1

Above is known as the Fundamental Equation of Asset Pricing.

Proof. The proof relies on the fact that in GEI equilibrium, there can be no arbitrage. Stiemke’s
Lemma implies that if there exists

b ∈ RS++ : q = b(p□A),

then there can be no arbitrage. Define

1
b≡ (v1 , v2 , . . . , vS ) ,
v0

where vs > 0 for all s = 0, 1, 2, . . . S. Then

1 ). /
q = b(p□A) ⇔ qj = vs ps · ajs ∀j ∈ J (3.2)
v0
s∈S
3.1. FUNDAMENTAL EQUATION OF ASSET PRICING 53

Define
1 1 )
≡ ∗ vs (ps · x∗s )
1+δ v0 p0 · x 0
s∈S
vs ps · x s
µs ≡ (1 + δ) >0 (3.3)
v0 p0 · x∗0
2
Note that s∈S µs = 1 so that µs may be interpreted as probabilities. Substituting for vs in (3.2)
using (3.3) yields that

C D
qj 1 ) ps · ajs
= µ s ∀j ∈ J.
p0 · x∗0 1+δ ps · x∗s
s∈S

Recall that there are S + 1 homogeneity properties of demand. So, for each state, adjust ps and q
so that x∗s is the numeraire. Then

1 ). /
qj′ = µs p′s · ajs ∀j ∈ J.
1+δ
s∈S

Remark 3.13. Some remarks are in order.

• µ is typically not uniquely determined. Even when it is, it usually does not correspond to any
objective probability over the state space.
. /
• It is clear that the price of a security is the discounted expected value of future payoffs ps · ajs
with risk-neutral probabilities (µs ) . Hence, asset prices have the Martingale property.

• Martingale property of the security prices are equivalent to the equilibrium of no arbitrage.

• The Martingale pricing theorem implies that prices are linear (in payoffs); ie, the price of an
. /
asset obtained by multiplying payoffs by λ for asset j λps · ajs is simply λqj .
5 0 1 6
Corollary 3.14. (Linear Pricing Corollary). Let p, q, xh , θ h be a GEI equilibrium of
h∈H
0* + 1
GEI economy uh , eh h∈H , A . Suppose that for some scalars λ1 , . . . , λJ , it holds that

J
) . /
λj p□aj = 0.
j=1
54 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

Then, it follows that

J
)
q·λ= qj λj = 0.
j=1

Proof. First, note that

J
) J
)
. / . /
λj p□aj = 0 ⇔ λj ps · ajs = 0∀s ∈ S
j=1 j=1

From the Fundamental Equation of Asset Pricing, there exists δ and µs such that

J J
E S
F
) ) 1 )
λj qj = λj µs ps · ajs
j=1 j=1
1 + δ s=1
S J
1 ) )
= µs λj ps · ajs
1 + δ s=1 j=1
J
)
Noting that λj ps · ajs = 0.
j=1

=0

Note that µs can be ”taken out” since it does not depend on the asset but only on the state.

Remark 3.15. Suppose that we have add a new asset into the GEI economy with payoffs an , and
moreover, suppose that there exists λj s such that
J
) . /
λj p□aj = (p□an ) .
j=1

Then, the linear pricing corollary says that the price of this new asset (which is redundant) is given
2J
by qn = j=1 qj λj .

3.2 GEI-GEI* Equivalence

0* + 1 0 * + 1
Definition 3.16. GEI* equilibrium of GEI economy u h , eh h∈H
,A is π, xh h∈H where
π ∈ RL × RLS and xh ∈ RL
+ × R+ ∀h ∈ H :
SL

2 2
1. Commodity markets clear: h∈H xh = h∈H eh
3.2. GEI-GEI* EQUIVALENCE 55

2. Budget constraints are satisfied:

0 . /
xh ∈ β h (π) ≡ x ∈ RL SL
+ × R+ : π · x − e
h
≤ 0.
. / 1
and π□ x − eh ∈ Span[π□A]

. /
3. Utilities are maximised: ∀x ∈ β h (π), uh xh ≥ uh (x) for all h ∈ H

Note, in particular, the a GEI economy has no assets. In this section, we show that GEI and
GE* are equivalent. This result is useful in proving what is to follow in this section.

Lemma 3.17. Suppose that the prices (p, q) and π ≥ 0 can be related to each other by

µs p s
π0 = p 0 , πs ≡ ∀s ∈ S, (3.4)
1+δ

where µs and 1 + δ are strictly positive. Suppose further that

S S
1 ) )
q= µs p s As = πs A s .
1 + δ s=1 s=1

Then, x ∈ β h (π) with all constraints binding ⇔ ∃θ ∈ RI : (x, θ) ∈ Bh (p, q) with all constraints
binding.

Proof. First, we prove ”⇒” so suppose that x ∈ β h (π) with all the constraints holding with equality.
. /
Since π□ x − eh ∈ Span[π□A] and
* +
Span(V) = z ∈ RS : z = (π□A)θ for some θ ∈ RI

there must exist θ such that

. /
π□ x − eh = (π□A)θ
. / . /
since π□ x − eh is the price of consumption x − eh and (π□A)θ is the monetary payoff from
holding portfolio θ, above implies that
. /
πs · xs − ehs = πs As θ∀s ∈ S. (3.5)

Since ps is a scalar multiple of πs for each s (by assumption) and the homogeneity properties of
demands yields that the last S inequalities of Bh must hold with equality (ie, budget constraints
for t = 1 must all bind). It remains to show that the budget constraint for t = 0 binds.
56 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

Recall the Martingale Pricing Theorem:

1 ). / 1 )
qj = µs ps · ajs ∀j ∈ J ⇔ q = (µs ps As ) .
1+δ 1+δ
s∈S s∈S

Hence,
4 3
. / . 1 ) /
p0 · x0 − eh0
+ qθ = p0 · x0 − + eh0
(µs ps As ) θ
1+δ
s∈S
. / 1 )
= p0 · x0 − eh0 + µs s s As θ
1+δ
s∈S
. / )
[U sing(3.4)] = p0 · x0 − eh0 + πs As θ
s∈S
. / ) . /
h
[U sing(3.5)] = π0 · x0 − e0 + πs · xs − ehs
s∈S
. h
/
=π· x−e =0

where that fact that it equals zero comes from the fact that the constraint in β h (π) is binding.
Hence, we have shown that {x, θ} ∈ Bh (p, q).
Now, we prove ”⇐” so suppose that {x, θ} ∈ Bh (p, q) with all the constraints holding with
equality. Thus, we immediately have that
. /
p□ x − eh = p□Aθ

from the last S constraints. Pre-multiplying both vectors by a diagonal matrix with µs /(1 + δ) in
the s th diagonal place give
# &
µ1
1+δ ··· 0
$ ' . / . /
$ .. .. .. '
$ . . . ' p□ x − eh = π□ x − eh
% (
µs
0 ··· 1+δ
# &
µ1
1+δ ··· 0
$ '
$ .. .. .. '
$ . . . ' p□Aθ = π□Aθ
% (
µs
0 ··· 1+δ

Therefore,
. / . /
π□ x − eh = π□Aθ ⇒ π□ x − eh ∈ Span[π□A].
. /
To show that π · x − eh = 0, notice that the steps in the proof for ” ⇒ ” may be reversed.
3.2. GEI-GEI* EQUIVALENCE 57
0* + 1
Theorem 3.18. (GEI-GEI* Equivalence Theorem). Consider the GEI economy u h , eh h∈H
, A :
5 0 1 6
1. Let p, q, xh , θ h be a GEI equilibrium. Then, there exists π (state prices) such that
h∈H
0 * + 1
π, xh h∈H is a GEI equilibrium. Moreover, for each asset j, qj = π · aj .
0 * + 1 0 1
2. Let π, xh h∈H be a GEI equilibrium. Then, there exists p, q, and θ h such that
5 0 1 6 h∈H

p, q, xh , θ h is a GEI equilibrium. Moreover, for each asset j, it holds that qj = π·aj .

h∈H
0 0 11
Proof. Part (i): Let p, q, xh , θ h be a GEI equilibrium. From weak monotonicity, it follows
that p ≥ 0 and furthermore that all the inequalities defining the budget set, Bh (p, q), hold with
equality. From the Fundamental Equation of Asset Pricing, there exists 1 + δ, µs ∀s ∈ S strictly
positive such that

1 )
q= µs p s As
1+δ
s∈S

Define

µs p s
π 0 = p 0 , πs ≡ ∀s ∈ S
1+δ

This yields the conditions for Lemma 3.17 to hold. Hence, it follows that xh ∈ β h (π) and is optimal.
0 * + 1
Part (ii): Let π, xh h∈H be a GEI ∗ equilibrium. Define
)
p ≡ π, q ≡ ps As , δ = 0, µs = 1∀s ∈ S
s∈S

Once again, the conditions of the Lemma 3.17 are satisfied so that there exists θ h such that
* h h+
x ,θ ∈ Bh (p, q). Note that it is possible to choose a maximal set of linearly independent
columns of j of the matrix π□A and θ h such that θjh ∕= 0 only if j is one of those columns. It
2
remains to show h∈H θ h = 0. since

) )
xh = eh
h∈H h∈H

then
). / )
0 = π□ xh − eh = (π□A) θh
h∈H h∈H
58 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS
! "
Because of the assumed independence of the columns of π□A that gets nonzero weight ie, θ hj ∕= 0 ,
2
we must have h∈H θ h = 0.

Remark 3.19. The proof of the theorem suggests that the assets matter only through their col-
lective span; ie, they matter only through the span of their money payoffs at the prices p. This is
made explicit in the next theorem.

Remark 3.20. Note that π is the vector of state prices in the economy; ie, the prices of Arrow
securities.

3.3 Asset Span Theorem

We can immediately draw inference from this theorem that assets matter only through their col-
lective span, and not through their individual names, or their number etc. More precisely, they
matter only through the span of their money payoffs at the prices p. We make this explicit in the
next theorem.
0* + 1
Theorem 3.21. (Asset Span Theorem). Let uh , eh h∈H , A be a GEI econ- omy with a GEI
5 0 1 6
equilibrium at p, q, xh , θ h . Suppose that B is an SL × K matrix representing the payoffs
h∈H
of K alternative assets, and that Span[p□A] = Span [p□B]. Then there are q′ , ϕ ∈ RK such that
0 * + 1 0* + 1
p, q′ , xh , ϕh h∈H is a GEI equilibrium for the GEI economy uh , eh h∈H , B .
Moreover, if

J
) . /
p□bk = λj p□aj
j=1

for some asset k, then

J
)
qk′ = λj qj .
j=1

5 0 1 6
h h
Proof. From the GEI-GEI* Equivalence Theorem, we know that the GEI equilibrium p, q, x , θ
h∈H
0 * + 1
is equivalent to a GEI equilibrium π, xh h∈H since Span[p□A] = Span[p□B], and we construct
π by scaling p, we must have Span[π□A] = Span[π□B]. Once again from the last theorem, this
3.3. ASSET SPAN THEOREM 59
0* + 1
must correspond to a GEI equilibrium for the economy u h , eh h∈H
, B . The second part follows
from the Linear Pricing Corollary. Suppose that
J
) J
)
. / . /
λj p□aj = p□bk ⇔ λj ps · ajs = ps · bks ∀s ∈ S
j=1 j=1

Then
J J
E S
F
) ) 1 )
λj qj = λj µs ps · ajs
j=1 j=1
1 + δ s=1
S
1 )
= µs ps · bks
1 + δ s=1

= qj′ ′

where we used the Fundamental Equation of Asset Pricing twice.

Remark 3.22. As long as two economies have the same monetary payoffs with some p vector, then
one can find asset prices etc. that support the same equilibrium allocation. Note, in particular,
that the economies need not have the same dimension of the asset payoffs (ie, different number of
assets) but the span of the monetary payoffs must be the same.

Remark 3.23. From this last theorem, we learn that not only are the prices of markets assets that

• are spanned by other basic assets are determined by prices of those basic assets, and

• assets which have not yet been priced but are spanned by existing assets are, in effect, already
priced.

Adding them to the economy would not change the equilibrium allocation of consumption goods.
5 0 1 6 0* 1
h h
+
Exercise 3.24. Let p, q, x , θ be a GEI equilibrium for the GEI economy uh , eh h∈H , A .
h∈H
. /
Let B = b1 , . . . , bJ+1 be an SL × (J + 1) matrix whose first J columns are identical to
. /
A = a1 , . . . , aj . Suppose that
)
bJ+1 = λ j aj ,
j∈J

Argue directly from the definition of equilibrium that

G G J J
H H ) K 0 0 11 K
p, q, λj qj , x h , θ h , 0
I I L h∈H L
j∈J
60 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS
0* + 1
is a GEI equilibrium for the economy u h , eh h∈H
, B .

Proof. By assumption, B′ s J + 1 th column is a linear combination of the columns from A such

that Span[p□A] = Span[p□B]. Hence, the consumption xh and portfolio θ h in the two economies
2
must coincide. Since h∈H θ h = 0 by definition, θJ+1 in economy B must be zero. The price of
the redundant asset J + 1 is immediate from the second result of the previous theorem since
# &
) J
) . /
p□bJ+1 = p□ % λ j aj ( = λj p□aj
j∈J j=1

3.4 COMPLETE MARKETS THEOREM

This states that the Arrow-Debreu economy and the GEI are equivalent if and only if the asset span
is full. In such cases, we can find state and commodity prices such that we replicate the equilibrium
of the GEI (or Arrow-Debreu) equilibrium.

Theorem 3.25. (Complete Markets Theorem). Recall that the market is complete if Span [p□A] =
RS
5 0 1 6 0* 1
h h
+
1. Let p, q, x , θ be a GEI equilibrium for the GEI economy u h , eh h∈H
,A and
h∈H
suppose that the market is complete. Then, there exists π such that is an Arrow-Debreu
* +
equilibrium for the Arrow-Debreu economy uh , eh h∈H .
0 * + 1 * +
2. Let π, xh h∈H be an Arrow-Debreu equilibrium for the economy uh , eh h∈H Then, there
5 0 1 6 5 0 1 6
exists an SL × S asset matrix A and p, q, θ h such that p, q, xh , θ h is a
h∈H h∈H
0* + 1
GEI equilibrium for the GEI economy uh , eh h∈H , A

Proof. (i) Follows immediately from the GEI-GEI∗ equivalence, since when the span of the assets
is all of RS , the GEI∗ budget set and the Arrow-Debreu budget sets are identical. (ii) Consider an
0 * + 1
Arrow-Debreu equilibrium π, xh h∈H . By weak monotonicity, we know that for each state s,
there must be some commodity l such that πsl > 0. For each s ∈ S, let asset s promise delivery of
one unit of commodity sl and nothing of every other commodity. Let the asset matrix A consist
of precisely these S assets. Then, it must be that Span[p□A] = RS . Hence, the Arrow-Debreu
3.5. MODIGLIANI-MILLER THEOREM 61
0* + 1
equilibrium is a GEI for the economy uh , eh h∈H , A . From the GEI-GEI∗ equivalence, we can
* +
make xh h∈H part of a GEI equilibrium.

Remark 3.26. Markets:

• In an Arrow-Debreu economy, there are L + LS markets at t = 0.

• In a GEI economy, there are L + J markets in t = 0.

The theorem shows that GEI can implement any Arrow-Debreu equilibrium:

• with only L + S markets in t = 0

• since at t = 1, only one state will actually occur, trades need to take place in only L + S + L
markets, which is likely to be smaller than L + LS.

In other words, fewer trades on assets are required. However, GEI places larger burden on the
rationality of expectations hypothesis.

3.5 Modigliani-Miller Theorem

Modigliani and Miller (1958) claimed that the financial policy of firms was irrelevant to real out-
comes.
0* + 1
Theorem 3.27. (Modigliani-Miller Theorem). Consider two GEI economies u h , eh h∈H
, A and
0* + 1
uh , eh h∈H , B where aj = bj for j = 3, . . . , J.
Suppose that there is some λ > 0 such that b1 = λa1 (first columns
5 of A and B6are collinear)
0 1
and that a1 + a2 = b1 + b2 . Then, for any GEI equilibrium p, q, xh , θ h for the first
h∈H
economy, there exists q1′ , q2′ , ψ1h , and ψ2h with q1′ + q2′ = q1 + q2 such that
0 * * ++ 1
p, {q1′ , q2′ , q3 , . . . , qI } , xh , ψ1h , ψ2h , θ3h , . . . , θJh h∈H

is a GEI equilibrium for the second economy.

Proof. This follows immediately from the Asset Span Theorem. Observe that because the first
two columns of A and B sum to the same vector, the same must be true of p□A and p□B, and
similarly the first columns are collinear. Hence, the span of the first two columns are the same, and
hence Span[p□A] = Span[p□B]
62 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

3.5.1 Implication

An implication of MMT: Suppose that z ∈ RSL

+ is a vector (to be thought of as the state-contingent

output of some firm). Suppose that a1 + a2 = z = b1 + b2 are two ways of diving the output of
the firm (say between debt and equity). If

? @ ? @
Span p□a1 , p□a2 = Span p□b1 , p□b2 (3.6)

then the distribution of firm profits between debt and equity will not disturb the original equilibrium.

3.5.2 Interpretation 1

The MMT says that if there are two different firms in the same economy, with the same random
output z , that finance their operations differently between debt and equity, then the sum of the
values of debt and equity must be the same for both firms.
If the values of the firms were unequal, there would be an arbitrage profit to be made by buying
the cheaper firm and selling short the more expensive clone (the proof of linearity of prices is based
precisely on no arbitrage). Secondly, a whole pie should cost the same amount, no matter how it is
sliced. This is the linearity of pricing assertion.

3.5.3 Interpretation 2

If some industry changed the mix of its financing between debt and equity, would it lead to a
different equilibrium? (No.) This involves a comparison across equilibria instead of a comparison
across firms in the same equilibrium.
To take the central special case, let
# & # &
k k
$ ' $ '
$ .. ' $ .. '
p□a1 = $ . ' , p□b1 = λ $ . '
% ( % (
k k

We interpret p□a1 as the money payoff to the bondholders when the firm issues k bonds and delivers
all that it promises and p□b1 as the money payoff when the firm delivers all that it promises on λk
bonds. Then (3.6) holds such that Modigliani-Miller theorem implies that the change is irrelevant
to real comes.
3.6. MARTINGALE PRICING THEOREM 63

There is a converse to the asset span theorem which will not be formally stated but, if Span[p□A] ∕=
* +
Span[p□B], then ”except for accidents”, the equilibrium allocation xh h∈H for the economy with
asset structure A will not be part of any equilibrium with asset structure B. This has an important
consequence for the the Modigliani-Miller Theorem (MMT) when we consider limited liability.
Suppose that
# & # &
min {k, p1 · z1 } min {λk, p1 · z1 }
$ ' $ '
$ .. ' $ .. '
p□a1 = $ . ' , p□b1 = $ . ' (3.7)
% ( % (
min {k, pS · zS } min {λk, pS · zS }

Recall that zs is the firm’s output in state s and k, λk is the money-worth of bonds issued.
So, suppose we’re in state s, and the firm’s output, zs is lower than what they need to pay to the
bond holders, in which case the bond holders (who owns security 1), is paid only the output. The
payment to the shareholder (who owns security 2) is whatever is left over after paying the debt
holders so

p□a2 ≡ p□z − p□a1

p□b2 ≡ p□z − p□b1

Given (3.7), note that p□a2 and p□b2 are not allowed to be negative. If S > 2 and for some
s, λk > ps · zs then almost surely

? @ ? @
Span p□a1 , p□a2 ∕= Span p□b1 , p□b2

and the firms’ financial policy will have a real effect.

Hence, when there is limited liability and the possibility of bankruptcy, MMT is generally invalid,
even ignoring the distortions caused by institutional factors.

3.6 Martingale Pricing Theorem

We start by defining some notation: If x ∈ RS and y ∈ RS and γ ∈ Pr(S), then

S
)
x ·γ y ≡ γ s x s ys
s=1
64 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

Note that this, given this notation, we may write Covγ (x, y) = x ·γ y − (Eγ x) (Eγ y) Moreover,

Eγ x = x · γ 1

where 1 = (1, . . . , 1) ∈ RS .
0* + 1
Definition 3.28. A GEI economy u h , eh h∈H
,A has weakly monotonic preferences if and
(S+1)L
only if for each state s = 0, 1, . . . , S, there exists an agent h ∈ H such that if x, y ∈ R+ x≥
h h
y, xs ≫ ys ⇒ u (x) > u (y) and if for all agents h ∈ H,

x ≫ y ⇒ uh (x) > uh (y)

0* + 1
Theorem 3.29. (Martingale Pricing Theorem). Let uh , eh h∈H , A be a GEI economy with
5 0 1 6
weakly monotonic preferences. Let p, q, xh , θ h be a GEI equilibrium. Then,
h∈H

∃π ∈ RS++ : q = π · [p□A]

Let γ ∈ Pr(S) and suppose γ ≫ 0. Then, there exists µ ∈ RS++ such that

∃µ ∈ RS++ : q = µ ·γ [p□A]

where π is the vector of state prices, and µ is the vector of stochastic discount factors (ie, marginal
utility, recall Lucas Model).

Proof. Given weakly monotonic preferences, there can be no arbitrage in a GEI equilibrium. In
other words, there cannot exist a solution to

qθ ≤ 0, [p□A]θ ≥ 0

with at least one strict inequality. By Stiemke’s Lemma, arbitrage can be ruled out if (and only if)

∃π ∈ RS++ : q = π[p□A] = π · [p□A]

Next, note that q = π · [p□A] implies that

. /
q = π · p□a1 , . . . , p□aJ
)
⇔ qj = πs ps · ajs ∀j ∈ J
s∈S
3.6. MARTINGALE PRICING THEOREM 65

Then defining
C D
π1 πs
µ= ,..., ≫0
γ1 γs
it follows that

q = µ ·γ [p□A]

Remark 3.30. The theorem states that the price of an asset is given by the sum of monetary
payoffs of the asset in every state weighted by the multiplier µ ≫ 0. The same multiplier can be
used for all the assets.

Remark 3.31. The Martingale Pricing Theorem states that in a GEI equilibrium, we account for
risk aversion not by adjusting the rate of return of each asset, but by adjusting the probabilities
on all the states, the same way for each asset, and then taking the expected value of the monetary
payoffs with respect to those probabilities discounted at the risk-free rate of interest.

Remark 3.32. A risk-free asset, f, pays out one in every state so that p□af = 1 . Define the
risk-free rate, r, as
1 )
≡ πs
1+r
s∈S

so that
2
πS
q = 2s∈S (π · [p□A])
s∈S πS
C D
1 π πS
= 2 1 ,..., 2 · [p□A]
1+r s∈S πS s∈S πS
2
Define µs ≡ πs / s∈S πs so that
µ · [p□A] Eµ [p□A]
q= = ,
1+r 1+r
where µs may be interpreted as risk-neutral probabilities over the states and µ is the Martingale
measure or the vector of risk-neutral probabilities.

,∗ ∈ Span[p□A] and unique µ∗ ∈ Span[p□A] with

Corollary 3.33. There exists unique µ

, ∗ · (p□A) = µ∗ · γ(p□A).
q=µ

Proof. Take the projection of π and µ (in γ geometry ) onto Span[p□A].

66 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

3.7 Linear Pricing Corollary

5 0 1 6
h h
Corollary 3.34. (Linear Pricing Corollary). Let p, q, x , θ be a GEI equilibrium for
h∈H
0* + 1
the GEI economy uh , eh h∈H , A that has weakly monotonic preferences.

1. Then, the asset prices are a linear function on Span[p□A].

2. In particular, if [p□A]θ = 0, then q · θ = 0.

3. It follows that, for any marketed monetary payoff m ∈ M = Span[p□A] there is a uniquely
determined price q(m).

Proof. Use the fact that q = π · [p□A]. Two also follows from this since

[p□A]θ = 0 ⇒ π · ([p□A]θ) = 0,

and

0 = π · ([p□A]θ) = (π · [p□A])θ = qθ.

3.8 Covariance Pricing Theorem

0* + 1
Theorem 3.35. (Covariance Pricing Theorem). Let uh , eh h∈H , A be a GEI economy with
5 0 1 6
weakly monotonic preferences. Let p, q, xh , θ h be a GEI equilibrium and γ ∈ Pr(S)
h∈H
strictly positive. Then, there exists

∃m∗ ∈ M ≡ Span[p□A] ⊂ RS

and ∃r > −1 such that

Eγ m − Covγ (m∗ , m)
q(m) = ∀m ∈ M
1+r

Proof. From the Martingale Pricing Theorem, we know that there exists µ∗ ∈ M such that for all
m ∈ M,

q(m) = µ∗ ·γ (m) ≡ (Eγ µ∗ ) (Eγ m) + Covγ (µ∗ , m)

3.8. COVARIANCE PRICING THEOREM 67

If (Eγ µ) ∕= 0, let
1
(Eγ µ∗ ) = , m∗ = −(1 + r)µ∗
1+r
Then

(Eγ m)
q(m) = + Covγ (µ∗ , m)
1+r
C D
(Eγ m) −m∗
= + Covγ ,m
1+r 1+r
Eγ m − Covγ (m∗ , m)
= ∀m ∈ M
1+r

Remark 3.36. This theorem states that the riskiness of an asset’s promised payments is not
measured by something intrinsic, such as its variance, but by its covariance with some specified
asset payoff m∗ .

Remark 3.37. Remark 3.37. Furthermore, any m∗ ∈ M satisfying above (for some r > −1) can
be written as
! "
∗
m∗ = α1 − µ∗ (1 + r)

∗
where r ∈ R+ and 1 = Proj∗A (1). If 1 ∈ M, then
1
= π(1)
1+r
and
1 ! ∗ "
m∗ = α1 − µ∗
q(1)

3.8.1 Security Market Line Theorem

0* +
Theorem 3.38. (Security Market Line Theorem). Let xh , eh h∈H be a GEI economy with
5 0 1 6
monotonic preferences. Let p, q xh , θ h be a GEI equilibrium where q ∕= 0. Let γ ∈ Pr(S)
h∈H
and γ ≫ 0. Then, we can find some asset payoff

m∗∗ ∈ M ≡ Span[p□A] ⊂ RS
68 CHAPTER 3. GENERAL EQUILIBRIUM WITH INCOMPLETE MARKETS

and r > −1 such that

q (m∗∗ ) = 1

and

Covγ (m∗∗ , m)
Eγ (m) − (1 + r) = [Eγ (m∗∗ ) − (1 + r)]
Varγ (m∗∗ )

for all m ∈ M with q(m) = 1.

Chapter 4

Dynamic Completion of Markets

Consider the case in which there are five states of the world in t = 2. Uncertainty is resolved as
in Figure 1 . There are three assets with payoffs and prices shown at each node and the risk-free
interest rate is assumed to be 0. We would like to find the conditional and unconditional risk-neutral
probabilities.

Figure 4.1: Tree

.

Recall that the state prices are the solution to

q = πA

Since the risk-free rate is zero, state prices and risk-neutral probabilities coincide. Thus, for the

69
70 CHAPTER 4. DYNAMIC COMPLETION OF MARKETS

upper half of the tree, it is the solution to

G
M
M 1 = π1∗ + π2∗ + π3∗
M
H
5 = 3π1∗ + 4π2∗ + 8π3∗ ⇒ π1∗ = π2∗ = π3∗ = 13 .
M
M
M
I 3 = 2π ∗ + 3π ∗ + 4π ∗
1 2 3

For the lower half of the tree

G
M
M 1 = π4∗ + π5∗
M
H
7 = 6π4∗ + 8π5∗ ⇒ π4∗ = π5∗ = 12 .
M
M
M
I 4.5 = 4π ∗ + 5π ∗
1 2

Notice that the third asset is redundant here. We need the unconditional probabilities for the whole
tree. For the first part,
G
M
M 1=π ,1 + π
,2
M
H
6 = 5,
π1 + 7,π2 ⇒ π ,2 = 12 .
,1 = π
M
M
M
I 3.75 = 3,
π + 4.5,
π 1 2

Note that the third asset is redundant. We deduce the unconditional probabilities for the whole
tree

π = (π1 , π2 , π3 , π4 , π5 )

= (π1∗ π
,1 , π2∗ π
,1 , π3∗ π
,1 , π4∗ π
,2 , π5∗ π
,2 )
C D
1 1 1 1 1 1 1 1 1 1
= × , × , × , × , ×
2 3 2 3 2 3 2 2 2 2
C D
1 1 1 1
= , , ,
6 6 6 4
Suppose we have another asset that pays

(0, 0.25, 4.25, 1.5, 3.5)

in each of the five states, respectively. Then, its price at t = 0 is

1 1 1 1 1
p0 = · 0 + · 0.25 + · 4.25 + · 1.5 + · 3.5 = 2
6 6 6 4 4
Its prices at t = 1, in the upper and lower half of the tree, respectively, are
1 1 1
p1,1 = 3 ·0+ 3 · 0.25 + 3 · 4.25 = 1.5
1 1
p1,2 = 2 · 1.5 + 2 · 3.5 = 2.5
 71

Notice that from the conditional probabilities, we’d expect the prices at t = 1 from the perspective
of t = 0 to match that of p0 :

1 1
p1 = · p1,1 + · p1,2 = 2 = p0
2 2

Note that, here, there are three securities and five states of the world and thus markets are in-
complete. However, we managed to find a unique Martingale measure and price any security!
Essentially, we manage to complete the markets through retrading at the second node. We call this
dynamic completion of the markets.

Theorem 4.1. Uniqueness of Martingale measure ⇔ Dynamic completion of markets.

Remark 4.2. For dynamic completion of markets, the number of securities traded must be no less
than the maximum number of branches emanating from each note on the event tree. This property
is used to price options. This is a necessary but not a sufficient condition.

Remark 4.3. It is also possible to complete markets via options or quasicomplete markets.
72 CHAPTER 4. DYNAMIC COMPLETION OF MARKETS
Chapter 5

Options

Definition 5.1. A call option gives you the right to buy the stock at a certain price (called
strike/exercise price, K), but not the obligation. Thus, you exercise it only when the stock’s price
is higher than the strike price K

Definition 5.2. A put option gives you the right to sell the stock at a certain price (again, called
strike/exercise price, K) , but not the obligation. Thus, you exercise it only when the stock’s price
is lower than the strike price K.

Definition 5.3. European options can be exercised only at their expiration whilst American options
can be exercised at any time before their expiration or at their expiration.

Some notations:

c : European call option price, C : American call option price,

p : European put option price, P : American put option price,
T : expiration date, St : stock price in period t,
K : strike price, rf : risk-free rate
B : price of a risk-free bond.

73
74 CHAPTER 5. OPTIONS

Figure 5.1: Payoff from a call option

.

5.1 Graphical Representations

5.1.1 Profits from a European call option

If we sell a call option, then we receive the call price c now. If the stock price stays the same or falls,
the option will mature expired, and we keep the future value of the sale price. If the stock price
rises, we lose a dollar for each dollar it rises (∆S > 0 with slope − 1) , Buying a call is the exact
opposite. If you are writing a call, you make maximum profit (c) when the stock price exceeds the
exercise price as you pay the call option holder ∆S > 0 and as ∆S increases, you make smaller and
smaller profit. If you are buying a call, you make minimum profit (loss) when ∆S ≤ 0. As soon as
∆S > 0, then you can buy the stock for less than the market price, hence profits start to increase.
Thus,

payoff = max (ST − K, 0)

5.1. GRAPHICAL REPRESENTATIONS 75

Figure 5.2: Payoff from a put option

.

5.1.2 Profits from a European put option

If you sell a put option, you obtain maximum profit if the stock price is equal or above the exercise
price.Otherwise you would have to pay the holder ∆S. If, on the other hand, you are buying a put,
you make greater profit if the stock price fell below the exercise price but you can only make a
minimum loss equal to the exercise price. Thus,

payoff = max (K − ST , 0)

5.1.3 Payoffs from a stock and a risk-free bond

Having a short position in stock imply that the profit is higher if the stock price falls where as
having a long position in stocks imply that profit is higher if the stock price increases. A risk-free
bond is presumed to have identical payoffs irrespective of the changes in the stock price.
76 CHAPTER 5. OPTIONS

Figure 5.3: Profits from a stock and a bond

.

5.1.4 Relationships amongst elemental securities

Elemental securities may be combined in various ways to obtain the following relationship

S+p=B+c

This implies that buying a share of stock (S) and buying a put written on that share (p) yields the
same payoff as holding a bond (B) and buying a call (C). This is clear if we realise that holding a
portfolio made up of long positions in the stock (ie, buy a stock) and the put (ie, sell a put) and a
short position in the call (ie, sell a call) is equivalent to the risk-free payoff from the bond.
Notice that having a call option can limit the downward risk from holding a stock.

5.2 Bounds of a call option

By using first-order stochastic dominance (ie, one asset will be preferred by all investors, regardless
of their type, if the return that it offers is superior in every state to the return offered by a second
1
asset), we can put both upper and lower bounds for the price of a call option, c.
1 The property of first-order stochastic dominance also implies that the value of call option does not depend on
preferences of the investors.
5.2. BOUNDS OF A CALL OPTION 77

Figure 5.4: Relationships amongst elemental securities

.

Proposition 5.4. c < S0

Proof. Suppose not, c > S0 . Consider the following trading strategy:

1. At t = 0, you sell the option and receive c and buy the stock depositingc − S0 > 0 at the
risk-free rate rf .

2. At t = 1 :

• if ST ≥ K, then the call option you sold is exercised so that you return the stock that
you own and be paid K; ie, (c − S0 ) (1 + rf ) + K > 0

• If ST < K, then the buyer will not exercise the option and you are left with (c − S0 ) (1 + rf )+
ST > 0

Above implies that, for a negative price (ie, you receive money today), you could set up a trading
strategy that gives strictly positive payoff in any case tomorrow-this is strong arbitrage and thus
cannot hold.
78 CHAPTER 5. OPTIONS

Figure 5.5: Mitigating downward risk with a call option

.

Proposition 5.5. c < S0 − K/ (1 + rf )

Proof. Suppose not, c < S0 − K (1 + rf ) . Consider the following trading strategy:

1. At t = 0, short sell the stock receiving S0 whilst buying the option and depositing ω =
S0 − c > K/ (1 + rf ) at the risk-free rate rf

2. At t = 1 :

• If ST ≥ K, then you exercise the option, pay K, and return the stock you sold short and
be left with ω (1 + rf ) − K > 0.

• If ST < K, then you do not exercise the option and buy the stock in the market at
ST < K and be left with ω (1 + rf ) − ST > 0.

Above implies that, for a negative price (ie, you receive money today), you could set up a trading
strategy that gives stricly positive payoff in any case tomorrow-this is strong arbitrage and thus
cannot hold.
5.3. PUT-CALL PARITY 79

5.3 Put-Call Parity

We show that for European options, there is a fixed relationship between the price of put and call
options with the same maturity date that are written on the same asset. Thus, if we know the price
of a European call option on an asset, we can derive the price of a put on the same asset.

Proposition 5.6. Assume no dividends, then

K
c+ = p + S0
1 + rf
Proof. Suppose a portfolio consists of:

1. One share of stock,

2. One put option,

3. One short call option.

Assume that the options have the same maturity date and exercise price K. There are two possible
states ST ≤ K and ST ≥ K.

If ST ≤ K :

you hold the stock worth + ST

call option is worthless + 0

put option is worth + K − ST

net position = K
If ST > K :

you hold the stock worth + ST

call option is worth + (ST − K)

put option is worthless + 0

net position = K
Hence, this portfolio is risk-free and thus we can discount its value at the risk-free rate:
K K
S0 + p − c = ⇔c+ = p + S0
1 + rf 1 + rf
80 CHAPTER 5. OPTIONS

5.4 Complete Markets and Options

Suppose there exists a payoff profile z = (z1 , . . . , zs ) with zs ∕= zs′ for all s ∕= s. A call option with
strike price K matures out-of-the money for all states with zs < K.

Proposition 5.7. Ross [1976] If the payoff z and S − 1 options with strike prices zs (except the
greatest) are traded, then markets are complete.

Remark 5.8. If the payoff z takes the same values in two states, then the markets are not complete.

Example 5.9. Let there be three states and the payoff of an asset be (1, 3, 6). Consider two
call options with strike prices 3 and 1, respectively. Then Proposition 5.7 says that markets are
complete. Now, let there be four states and the payoff of the asset (1, 3, 3, 6). Can we complete the
markets? Since payoffs in state 2 and 3 are the same, the options on this asset must also have the
same payoffs in those states.

Remark 5.10. Markets are incomplete even if all options with arbitrary strike prices are traded.

5.5 American versus European Options

Since American options can be exercised at any time, they must be worth at least as much as the
corresponding European options; ie,
C ≥ c, P ≥p

Proposition 5.11. An American call option on a non-dividend paying stock is never worth being
exercised early.

Proof. Let Γn be the value of an American call option at t = n. Obviously, Γn ≥ Vn , where Vn is

the value of the respective European call, and Un be the value of the respective put option. Then
put-call parity for t = n gives
K
Vn + = U n + Sn
(1 + r)N −n
Then
K K
Γn ≥ Vn ⇒ Γn ≥ Un + Sn − ≥ Sn − > Sn − K
(1 + r)N −n (1 + r)N −n
Thus, the value of an American call at t = n is greater than what we will get from immediate
exercise. It follows that an American call (on a non-dividend paying stock should never be exercised
early. Consequently, Γn = Vn for all n.
5.6. PRICING 81

Proposition 5.12. An American put may be worth being exercised early.

Proof. Let Φn be the value of an American put option at t = n. Let Vn be the value of the respective
European call and Un be the value of the respective European put option, then Φn ≥ Un . Put-call
parity for t = n yields
K
Vn + = U n + Sn
(1 + r)N −n
Then
K K
Φn ≥ U n = V n + − Sn ≥ − Sn
(1 + r)N −n (1 + r)N −n
But
K
− Sn ≥ K − Sn
(1 + r)N −n
so that it may not always be the case that Φn ≥ K − Sn

5.6 Pricing

5.6.1 Hedging sale of an option

Consider a binomial model such that there are two states of the world. Stock price today (S0 ) is
£20 and in three months (S1 , S2 ) = (£22, £18). The riskfree rate is assumed to be rf = 12% .
Consider a call option with strike price K = £21 and maturity 1/4 (ie, three months). Then the
three-month compounding interest rate is given by

1/4
1 + r4 = (1 + rf ) = 1.029

Example 5.13. Consider a strategy in which you hold ∆ stocks long and short 1 option. The
payoff of the portfolio in the first state is 22∆ − 1 and in the second 18∆. For the portfolio to be
riskless, we need that
1
22∆ − 1 = 18∆ ⇒ ∆ =
4
The payoff of the portfolio is then £4.5. The price today is given by 4.5/1.029 = £4.374. The value
of the option is thus worth
1
£20 × − £4.374 = £0.626
4
82 CHAPTER 5. OPTIONS

Notice that ∆ is the ratio of the change in the price of a stock option to the change in the price
of the underlying stock. It is the number of stock we have to go short (long) to hedge our position
if we go long (short) on an option position; ie, to lock on the risk-free rate. This is called delta
hedging.

We now generalise the previous example. Now suppose that your portfolio consists of ∆ stocks
long and 1 option short, with payoff profile (fu , fd ) . The price tomorrow is either uS0 or dS0 where
u > 1 + rf > 1 and 0 < d < 1 < 1 + rf .

Figure 5.6: Development of the share price

.

The payoff of the portfolio in the first state is uS0 ∆ − fu and dS0 ∆ − fd . For the portfolio to
be riskless, we need ∆∗ such that

fu − fd
uS0 ∆∗ − fu = dS0 ∆∗ − fd ⇒ ∆∗ =
S0 (u − d)

The payoff of the portfolio is then

fu − fd
uS0 ∆∗ − fu = uS0 − fu
S0 (u − d)

and its price today is

= >
1 ∗ 1 fu − fd
(uS0 ∆ − fu ) = uS0 − fu
1 + rf 1 + rf S0 (u − d)
5.6. PRICING 83

Then the value of the option, f, will be

1
f = S0 ∆ ∗ − (S0 u∆∗ − fu )
1 + rf
1
= [(1 + rf ) S0 ∆∗ − S0 u∆∗ + fu ]
1 + rf
1
= [(1 + rf − u) S0 ∆∗ + fu ]
1 + rf
= >
1 fu − fd
= (1 + rf − u) S0 + fu
1 + rf S0 (u − d)
= >
1 (1 + rf − u) (fu − fd )
= + fu
1 + rf (u − d)
5 >
1 u − (1 + rf ) (fu − fd ) − d
= fu
1 + rf u−d

Define the risk-neutral probabilities as

1 + rf − d
π∗ ≡
u−d

Then
1
f= [π ∗ fu + (1 − π ∗ ) fd ] (5.1)
1 + rf
π ∗ is referred to as the risk-neutral probabilities for the first state; it is the probability of the first
state occurring if investors were risk neutral. This also implies that the price of the option is simply
2
the expectation of its discounted future value in a risk-neutral world.
Suppose that the actual probability that S moves to uS is given by q. Given risk-neutral in-
vestors, the expected rate of return on the stock would be the riskless interest rate; ie,

(1 + rf ) S = quS + (1 − q)dS
(1 + rf ) − d
⇔q= = π∗
u−d
Hence, the value of an option can be interpreted as the expectation of its discounted future value
in a risk-neutral world.
Note that we can price the option by finding the risk-neutral probabilities instead of finding a
replicating portfolio. Some properties of the option pricing formula 5.1:
2 Note
' (
that 1/ 1 + rf imply that the cashflow at t = 1 is being discounted at the risk-free rate, which indicates
that the cashflow itself must be risk-neutral. π ∗ is called risk-neutral probabilities since it ensures that cashflow is
indeed risk-neutral.
84 CHAPTER 5. OPTIONS

1. The price does not depend on the probability of a state occurring: investors with heteroge-
neous expectations about such probability will still agree on the option value relative to its
other parameters.

2. Individual attitude toward risk are irrelevant: we only need monotonicity in wealth so that
arbitrage profits are eliminated in equilibrium.

3. Only random variable on which the option value depends is the stock itself; ie, the is it
independent of the market portfolio.

4. Irrelevance of the stock’s return: When we are valuing an option in terms of the underlying
stock, the expected return on the stock is irrelevant.

5.6.2 Replicating an option using bond and stock

We will arrive at the same pricing formula of an option as in the previous subsection but using a
slightly different way.
Suppose

Figure 5.7: Development of the share price

.

We wish to create a replicating portfolio consisting of ∆ amount of stocks and B amount of bonds.
5.6. PRICING 85

Thus, we wish to find ∆ and B such that

fu = uS∆ + (1 + rf ) B
fd = dS∆ + (1 + rf ) B

From the first expression, we may obtain

fu − (1 + rf ) B fu − uS∆
∆= ,B =
uS 1 + rf

Then
fu − (1 + rf ) B
fd = dS + (1 + rf ) B
uS
⇔ ufd = dfu − d (1 + rf ) B + u (1 + rf ) B
ufd − dfu
⇔B=
(1 + rf ) (u − d)
fu − uS∆
fd = dS∆ + (1 + rf )
1 + rf
fu − fd
⇔∆=
S(u − d)
Thus, the replicating portfolio is given by holding the following portfolio

ufd − dfu fu − fd
B= ,∆ =
(1 + rf ) (u − d) S(u − d)

No arbitrage implies that the price of the option must be the same as the price of this portfolio,
which is given by
p = ∆S + B
fu − fd ufd − dfu
= S+
S(u − d) (1 + rf ) (u − d)
= >
1 (1 + rf ) (fu − fd ) + ufd − dfu
=
1 + rf u−d
= >
1 (1 + rf ) − d u − (1 + rf )
= fu + fd
1 + rf u−d u−d

5.6.3 Binomial model of option pricing Cox et al. [1979]

5.6.3.1 Method 1

Recall that a necessary and sufficient condition for no arbitrage is the existence of a unique Mar-
tingale probability measure given that markets are dynamically complete with the price processes
86 CHAPTER 5. OPTIONS

we have assumed. Define R ≡ 1 + rf . Let s∗ (t) = s(t)R−t and B ∗ (t) = Rt R−t = 1 be the discount
values of the stock and the bond at each t. We need to find π ∗ such that

= >
∗ s∗ (t + 1)
E = 11 , ∀t < T.
s∗ (t)
By backward reasoning,

π ∗ uR−1 + (1 − π ∗ ) dR−1 = 1.

At any node t < T :

R−d
π∗ = .
u−d
since π ∗ is a probability, π ∗ ∈ (0, 1), which implies that

d<R<u

as we had previously. We realise that the probability of any of nature un- der the Martingale
probabilities is completely determined by the number of upward moves. This implies that for n
upwards moves
n T −n
π ∗ (nup) = (π ∗ ) (1 − π ∗ )

and

s(t) = s(0)un dT −n .

Define, as usual,
T!
CTn = .
n!(T − n)!
In general, given s(t), the probability that there will be n ≤ T − t upward moves after t is given by

n T −t−n
CTn−t (π ∗ ) (1 − π ∗ )

and the stock price is then

S(T ) = s(t)un dT −t−n .

Now, recall that the payoff of a European call option at T is given by

max[s(T ) − K, 0].
5.6. PRICING 87

Given the information of our binomial tree, the price is

0 1
ct = E∗ max[s(T ) − K, 0]R−(T −t)
T
) −t
n T −t−n ? @
= R−(T −t) CTn−t (π ∗ ) (1 − π ∗ ) max s(t)un dT −t−n − K, 0 .
n=0

Let j ≡ minimum number of upward moves such that s(t)uj dT −t−j ≥ K ie, the call option finishes
in-the-money), then taking logs,

ln[s(t)] ln(u)(T − t − j) ln(d) ≥ ln(K)

ln(K)
⇒ j[ln(u) − ln(d)] ≥
(T − t) ln(d) ln[s(t)]
? @
ln K/s(t)dT −t
⇒j≥ , j ∈ Z++
ln(u/d)

Then
T
) −t
n T −t−n ? @
ct = R−(T −t) CTn−t (π ∗ ) (1 − π ∗ ) s(t)un dT −t−n − K
n=j
T
) −t C Dn = >T −t−n
π∗ u (1 − π ∗ ) d
= s(t) CTn−t ]T −t−n
n=j
R R
2T −t n T −t−n
−KR−(T −t) n=j CTn−t (π ∗ ) (1 − π ∗ )
,) − KR−(T −t) φ (j|T − t, π ∗ ),
= s(t)φ(j|T − t, π
where
T
) −t
n T −t−n
φ (j|T − t, π ∗ ) = CTn−t (π ∗ ) (1 − π ∗ )
n=j

is the complementary binomial distribution, which implies that the binomial distribution equals
1 − φ (j|T − t, π ∗ ) , and ? @
ln K/s(t)dT −t
j ∈ Z++ : j ≥
ln(u/d)
u ∗
,= π
π
R
d
, = (1 − π ∗ )
⇒1−π
R

5.6.3.2 Method 2

We can come to the same formulation using a different approach. The T-period generalisation of
the binomial call (European) pricing formula is the probability of each final outcome multiplied by
88 CHAPTER 5. OPTIONS

the value of the outcome and discounted at the risk-free rate for T time periods. The general form
of the payoff is then
? @
max un dT −n s(t) − K, 0 ,

where, as previously, we assumed that stocks in each period can move up by a multiplicative factor
u or down by a multiplicative factor d, T is the total number of periods, n measures the number of
upward movements in the stock. The general form of the probabilities of each payoff is given by
the binomial distribution:
(T − t)! n T −t−n
φ (n|T − t, π ∗ ) = (π ∗ ) (1 − π ∗ )
(T − t − n)!n!
T −t−t
= CnT −t (π ∗ ) (1 − π ∗ ) .

Then the call price in the initial period is given by

T
) −t
1 n T −t−n ? @
ct = CTn−t (π ∗ ) (1 − π ∗ ) max un dT −t−n s(t) − K, 0 (5.2)
RT −t n=0

To simplify the expression, use the fact that many of the final payoffs for a call option will be zero
because the option finishes out-of-the-money (ie, the strike price is higher than the stock price).
Denote j as the integer that bounds these states of nature when the option has a nonnegative value.
Put differently, note that when n increases, un dTS −t−n (t) − K also increases. We choose

j = min{n} : un dTS −t−n (t) − K > 0,

which implies that for n < j, un dT −t−n s(t) − K is negative so that

? @
max un dT −t−n s(t) − K = 0.

Note that 7 ? @8
ln K/s(t)dT −t
j = min j ∈ Z++ :j≥ .
ln(u/d)

We can now express (5.2) as

3T −t 4
) un dT −t−n
∗ n ∗ T −t−n
ct = s(t) CTn−t (π ) (1 − π )
n=a
R
3T −t 4
K ) n T −t−n
− T −t CTn−t (π ∗ ) (1 − π ∗ ) .
R n=a
5.6. PRICING 89

The second term is the discounted value of the exercise price. Define the complementary binomial
distribution Φ (n ≥ j|T − t, π ∗ ) as the cumulative probability of having in-the-money options where
the probabilities are the risk-neutral probabilities. Define
u ∗ d
,≡
π , ≡ (1 − π ∗ ) .
π ⇒1−π
R R
We then have that
un dT −t−n ! u "n = d >T −t−n
∗ n ∗ T −t−n ∗ ∗
(π ) (1 − π ) = π (1 − π )
R R R
n T −t−n
= (π ∗ ) (1 − π ∗ ) .

Finally the model can be summarised as

K
ct = s(t)Φ(n ≥ j|T − t, π
,) − Φ (n ≥ j|T − t, π ∗ ) ,
RT −t
where
R−d
π∗ = ,
u − d′
u
, = π∗ ,
π
R 7
? @8
ln K/s(t)dT −t
j = min j ∈ Z++ :j≥ ,
ln(u/d)
T
) −t
Φ(n ≥ j|T − t, π) = CTn−t π n (1 − π)T −t−n .
n=j

5.6.3.3 Properties

• The price of a call option increases as the price of the underlying stock, s(t), increases.

• The price of a call option decreases as the strike price K increases.

• If R increases, the main effect is to decrease the discounted value of the exercise price, KRT −t ,
which increases the price of the call option. However, there are secondary effects on π ∗ and
, causing them to decrease as R increases.
π

• The price of a call option increases as the number of total time periods, T, increases. Note
that T does not change the hedging probabilities, π ∗ , however, it does increase the number
of positive payoffs.

• The price of a call option increases with the size of the change, σ 2 = T π ∗ (1 − π)∗
90 CHAPTER 5. OPTIONS

5.6.3.4 Intuition

The value of the call is equal to the today’s stock price, converted into a risk-adjusted position of
one call by multiplying it by one over the hedge ratio, then subtracting the present value of the
exercise price weighted by the probability that the option will mature in-the-money. π̂ is used for
the risk-adjusted position and π ∗ is used for the objective probability of finishing in the-money.

Example 5.14. Consider the binomial model with u = 1.2, d = 0.9, R = 1.1 and initial asset price
s(t) = 100. We can then calculate the prices of a European call and a European put option with
exercise price K = 100 and n = 4 periods left until expiry. First, the risk-neutral probability of the
’up’ state is
R−d 1.1 − 0.9 2 1
π∗ = = = ⇒ 1 − π∗ =
u−d 1.2 − 0.9 3 3
Then from (5.2),
T
) −t
n T −t−n ? @
ct = R−(T −t) CTn−t (π ∗ ) (1 − π ∗ ) max s(t)un dT −t−n − K, 0
n=0

1
24 4!
. 2 /n . 1 /4−n ? @
= 1.14 n=0 n!(4−n)! 3 3 max 100(1.2)n (0.9)4−n − 100, 0
100
24 4! 2n
? n 4−n
@
= 1.14 n=0 n!(4−n)! 34 max (1.2) (0.9) − 1, 0
100
. 4! 1 4! 2 4! 4 104 4! 8 8 4! 671 4!
/
= 1.14 0!4! 81 ×0+ ×0+
1!3! 81 2!2! 81 × 625 + 3!1! 81 × 625 + 4!0! 81 × 625
100
. 2496 11104
/
= 1.14 50625 + 50625 + 10736
50625 =
100 24336 24336
1.14 1.14 50725

= 32.7
The put option can then be priced using the put-call parity

pt = ct − s(t) + KR−T −t
100
= 32.7 − 100 +
1.14
= 1.0.

5.7 Black-Scholes Model

Assumptions:

• The stock price follows a geometric Brownian motion with constant drift and volatility.

• The short selling of securities with full use of proceeds is permitted.

5.7. BLACK-SCHOLES MODEL 91

• There are no transaction costs or taxes.

• All securities are perfectly divisible.

• There are no dividends during the life of the derivative.

• There are no riskless arbitrage opportunities.

• Security trading is continuous.

• The risk-free rate of interest is constant and is the same for all maturities.

The main underlying argument is that you can dynamically complete the markets with just the
stock and the bond. Thus, you can price an option on the stock using risk-neutral pricing (ie, you
can derive the unique state price that exclude arbitrage). In particular, the stock price and the
option price depend on the same underlying source of uncertainty. We can then form self-financing
portfolios through continuous retrading such that, at the expiration date, the portfolio of the bond
and the stock replicate exactly the payoffs of the option. Thus, according to the law of one price, the
portfolio and the option should have the same price today. In other words, we dynamically complete
the markets by means of self-financing portfolio that replicates exactly the payoffs in t + δt in any
state of the world. This happens even if markets are in general incomplete (ie, at the expiration
date, we don’t have as many assets as states). Hence, we can calculate the unique risk-neutral
probabilities (ie, a unique risk-neutral probability measure) and apply risk-neutral valuation in the
pricing of the option. Put another wayy we can price the option as a redundant security even
though markets may be’ typically’ incomplete but dynamically complete.
Implied volatility is the volatility implied by option prices observed in the market as if they were
determined by the Black-Scholes model. Using the Black-Scholes formula, we use the option prices
that we observe as inputs and solve for the volatility. Traders like to calculate implied volatilities
from actively traded options on a certain asset and interpolate between them to calculate the ap-
propriate volatility for pricing a less actively trade option on the same stock.

The Black-Scholes model is an extension of the binomial pricing model. It is a continuous time
equivalent if we hold the amount of time constant and divide it into more and more binomial trials.
Define r as the rate of return for one year as usual and let a be the rate that is compounded n
92 CHAPTER 5. OPTIONS

3
times in the interval T, then

C D Tn
j
lim 1+ = ej = 1 + r.
n→∞ n/T
In continuous time, u and b in a single binomial trial relate to the annual standard deviation of a
stock’s rate of return as E N F E N F
T T
u = exp σ , d = exp −σ .
n n
Invoking the central limit theorem as we allow for continuous time trading, the pricing formula
becomes
c = s(t)N (xt ) − K exp[−r(T − t)]N (yt ) ,

where
ln[s(t)/K] + rf (T − t) 1 √
xt = √ + σ T − t,
σ T −t 2
√
yt = xt − σ T − t,
O x
N (x) = f (z)dz, z ∼ N (0, 1).
−∞
Cox et al. [1979] proved that as the number of binomial jumps per year, n, becomes large, the 2
formulas converge; ie,
Φ(n ≥ j|T − t, π,) → N (xt ) ,
! √ "
Φ (n ≥ j|T − t, π ∗ ) → N xt − σ T − t .
In other words, the binomial option pricing formula contains the Black-Scholes as a limiting case.
What does this mean? since the exercise will occur when s(t) > K, the first term represents the
present value of receiving the stock if (and only if) s(t) > K. The second term is the present value
of paying the exercise price if (and only if) s(t) > K.
If s(t) ≫ K, then the option will be exercised with probability 1, implying that N (xt ) , N (yt ) →
1. Thus,
c = s(t) − K exp[−r(T − t)].

Alternatively, if we view the European call option as a portfolio with a long position in stock and a
short position in bond then the first term can be interpreted as the amount invested in stock and
the second term, the amount borrowed.
3 Recall the definition of the exponential function:
exp (x) ≡ limm→∞ (1 + x/m)m .
Define m ≡ n/T then as n → ∞, m → ∞. Thus, the expression [1 + j/(n/T )]n/T becomes ej in the limit.
5.7. BLACK-SCHOLES MODEL 93

Note that this methodology cannot be used for all derivatives (refer to the assumptions) and
when closed form solutions are not available, we have to resort to numerical procedures (ie, com-
putationally expensive). We use the formula to obtain the implied volatility σ ∗ for the stock that
the exotic derivative depends on, then decide the time interval δt for the recombining of the tree
and calculate the up and down movements as
! √ " ! √ "
u = exp σ ∗ δt , d = exp −σ ∗ δt .

The risk-neutral probabilities are calculated as usual

exp(rδt) − d
π∗ = .
u−d

Finally, we calculate the payoff at every node and implement risk-neutral pricing via backwards
induction.
When the holder has the right to exercise early, binomial trees are particularly useful for pricing.
4
For fixed income securities, we use trinomial trees (see Figure 5.8). We use the binomial tree
procedure to price:

• American calls on dividend paying stock.

• American puts on non-dividend and dividend paying stocks.

• Lookback options which gives its holder the right to purchase the underlying asset at the
minimum price realised over the life of the option.

• Barrier options whose payoff depends on whether the underlying asset’s price reaches a certain
level during a certain period of time.

• Asian options whose payoff is not determined by the underlying price at maturity but by the
average underlying price over some preset period of time.

• Compound options which is simply an option on an option. The exercise payoff of a compound
option involves the value of another option.

4 Fixed income refers to any type of investment that is not equity which obligates the borrower to make payments
on a fixed schedule even if the number of the payments may be variable. For example, if you lend money to a
borrower and the borrower has to pay interest once a month, then you have been issued a fixed-income security.
94 CHAPTER 5. OPTIONS

Figure 5.8: Trinomial tree

.
Chapter 6

Portfolio Theory

6.1 Portfolio Selection

Define Y0 as the initial wealth, rP as the risky asset’s rate of return, rf as the risk-free asset’s rate
of return, and a as the amount of investment in the risky asset. Then investor’s wealth at the end
of the period, Ỹ1 , is given by

YP1 = (1 + rf ) (Y0 − a) + a(1 + rP)

r − rf ) .
= Y0 (1 + rf ) + a (P
At time 0, he wishes to maximises his expected end-of-the-period payoff. Thus, he solves
! "
arg max Eu YP1 = Eu [Y0 (1 + rf ) + a (P
r − rf )] (6.1)
a

The first-order condition is given by

E {U ′ [Y0 (1 + rf ) + a (P
r − rPf )] (P
r − rPf )} = 0.

Example 6.1. Let u(Y ) = ln Y and the risk asset’s returns be either r2 or r1 , r2 > rf > r1 , and
r) = πr2 + (1 − π)r1 > rf . The first-order condition then comes
E(P
= >
rP − rf
E = 0.
r − rf )
Y0 (1 + rf ) + a (P
Note that, with probability π, rP = r2 and with probability 1 − π, rP = r1 Thus, above is equivalent
to
π (r2 − rf ) (1 − π) (r1 − rf )
+ = 0.
Y0 (1 + rf ) + a (r2 − rf ) Y0 (1 + rf ) + a (r1 − rf )

95
96 CHAPTER 6. PORTFOLIO THEORY

Rearranging yields

π (r2 − rf ) [Y0 (1 + rf ) + a (r1 − rf )]

= −(1 − π) (r1 − rf ) [Y0 (1 + rf ) + a (r2 − rf )]

⇔ π (r2 − rf ) Y0 (1 + rf ) + aπ (r2 − rf ) (r1 − rf )

= −(1 − π) (r1 − rf ) Y0 (1 + rf )

− a(1 − π) (r1 − rf ) (r2 − rf )

⇔ a (r2 − rf ) (r1 − rf )

= −Y0 (1 + rf ) [(1 − π) (r1 − rf ) − π (r2 − rf )]

a (1 + rf ) [πr2 + (1 − π)r1 − rf ]
⇔ =−
Y0 (r2 − rf ) (r1 − rf )
(1 + rf ) [E(,r ) − rf ]
=− >0
(r2 − rf ) (r1 − rf )

Theorem 6.2. Assume u′ (·) > 0 and u′′ (·) < 0 and let ,
a denote the solution to the maximisation
problem (6.1). Then

a > 0 ⇔ E(P
, r ) > rf

a = 0 ⇔ E(P
, r ) = rf

a < 0 ⇔ E(P
, r ) < rf

In words, given an increasing, concave utility function:

• it is optimal to hold risky assets if and only if the expected rate of return on the risky asset
is higher than that of the risk-free asset,

• it is never optimal to hold risky assets if and only if the expected rate of return on the risky
asset is equivalent to that of the risk-free asset, and

• it is optimal to take a short position on the risky asset if and only if the expected rate of
return on the risky asset is less than that of the risk-free asset.

Remark 6.3. Let ,

a=,
a (Y0 ) be the solution to the maximisation problem (6.1). For all Y0 , for
6.2. MARKOWITZ PORTFOLIOS 97

two investors 1 and 2, we have

ARA1 > ARA2 ⇒ ,

a1 (Y0 ) < ,
a (Y0 )
d [,
a (Y0 )]
RRA′ (Y ) < 0 ⇒ <0
dY0
d [,
a (Y0 )]
RRA′ (Y ) = 0 ⇒ =0
dY0
d [,
a (Y0 )]
RRA′ (Y ) > 0 ⇒ >0
dY0
where
u′′i (Y ) Y u′′i (Y )
ARAi = − , RRA = , ∀i ∈ {1, 2}
u′i (Y ) u′i (Y )
Only for utility functions with constant ARA or RRA does the composition of the investors’
optimal portfolio not vary with his wealth Y ([Cass]) . More specifically, unctions with constant
ARA implies that the investor will always hold the same amount of risk-free assets where as utility
functions with constant RRA implies that the investor will always hold the same proportionate
amount of risk-free assets and risky assets.
r1 , . . . , rPN ) . The investor’s maximisation problem
Suppose we have N risky assets with returns (P
now becomes: 3 4
N
)
arg max Eu (Y1 ) = Eu Y0 (1 + rf ) + ri − rf ) .
ai (P
a1 ,...,aN
i=1

Define wi ≡ ai /Y0 such that it represents the proportion of wealth invested in the risky asset i.
Equivalently, 7 3 48
N
)
arg max Eu Y0 (1 + rf ) + ri − rf )
wi (P
w1 ,...,wN
i=1
A B
P
⇔ arg max Eu Y0 (1 + R)
w1 ,...,wN
! "
⇔ arg max Eu YP1 ,
w1 ,...,wN

where YP1 is the end-of-period wealth and R

P is the rate of return of the entire portfolio.

6.2 Markowitz Portfolios

From now on, we work with mean-variance utility function; ie, utility depends only on the mean
(denoted with µ) and the variance of the portfolio (denoted with σ 2 ) : U (µp , σp ) with U ′ > 0 and
U ′′ < 0
98 CHAPTER 6. PORTFOLIO THEORY

1
Definition 6.4. Portfolio A mean-variance dominates portfolio B if

(µA ≥ µB ) ∧ (σA < σB )

or

(µA > µB ) ∧ (σA ≤ σB )

Definition 6.5. The efficient frontier is the locus of all undominated portfolios in the (σ, µ) space.

Consider a portfolio, p, consisting of two assets (1 and 2 ) with weights w1 on asset 1 and w2 =
1 − w1 on asset 2. Expected rates of return are given by r1 and r2 for assets 1 and 2, respectively.
The two assets have variance σ12 and σ22 , respectively. Then expected return from the portfolio, µp ,
and its variance, σp , are given by

µp ≡ E (w1 rP1 + w2 rP2 )

= w1 r1 + (1 − w1 ) r2 (6.2)

σp2 ≡ Var (w1 rP1 + w2 rP2 )

= w12 Var (P
r1 ) + w22 Var (P r1 , rP2 )
r2 ) + 2w1 w2 Cov (P
2
= w12 σ12 + (1 − w1 ) σ22 + 2w1 (1 − w1 ) σ1 σ2 ρ12 (6.3)

where

ri ≡ E (P
ri ) ∀i ∈ {1, 2}

σi2 ≡ Var (P
ri ) ∀i ∈ {1, 2}
r1 , rP2 )
Cov (P r1 , rP2 )
Cov (P
r1 , rP2 ) ≡ Q
ρ12 ≡ Corr (P =
Var (P
r1 ) Var (P
r2 ) σ1 σ2

Without loss of generality, we assume that

r 2 ≥ r 1 , σ2 ≥ σ1

1 Recall that ∧ means ”and”.

6.2. MARKOWITZ PORTFOLIOS 99

6.2.1 Case 1: Perfect positive correlation (ρ12 = 1)

Substituting ρ12 = 1 into (6.3) yields

2
σp2 = w12 σ12 + (1 − w1 ) σ22 + 2w1 (1 − w1 ) σ1 σ2
2
= [w1 σ1 + (1 − w1 ) σ2 ]

⇒ σp = ± [w1 σ1 + (1 − w1 ) σ2 ]
σp − σ2
⇒ w1 = ∵ σp ≥ 0
σ1 − σ2

which implies that the standard deviation of the portfolio is given by

σp = ± [w1 σ1 + (1 − w1 ) σ2 ] ,

which, in turn, implies the weight

σp − σ2
w1 = ,
σ1 − σ2

while noting that σp > 0 by definition. Substituting the expression for w1 into the expected return
equation, (6.2), yields
C D
σp − σ2 σp − σ2
µp = r1 + 1− r2
σ1 − σ2 σ1 − σ2
σ2 − σp σ1 − σp
= r1 + r2
σ2 − σ1 σ1 − σ2
σ2 σp σ1 σp
= r1 − r1 + r2 −
σ2 − σ1 σ2 − σ1 σ1 − σ2 σ1 − σ2
σ2 σ1 r2 − r1
= r1 − r2 + σp
σ2 − σ1 σ2 − σ1 σ2 − σ1

This gives the equation for the portfolio frontier in (σ, µ) space. Figure 6.1 plots this frontier. Note
that the efficient frontier in this case is the whole line.
Notice that, in general, the efficient frontier will not be linear due to the covariance term in σp2 .

6.2.2 Case 2: Imperfectly correlated assets (−1 < ρ12 < 1)

Consider how the variance of the portfolio is affected by the correlation term:

∂σp2
= 2w1 (1 − w1 ) σ1 σ2 > 0
∂ρ12
100 CHAPTER 6. PORTFOLIO THEORY

Figure 6.1: Two perfectly positively correlated assets

.

Hence, when ρ12 < 1 then the variance of the portfolio, σp2 is smaller than in the case when ρ12 = 1;
ie,

. / . /
σp2 |ρ12 ∈ [−1, 1) < σp2 |ρ12 = 1

⇒ (σp |ρ12 ∈ [−1, 1)) < (σp |ρ12 = 1) ∵ σp ≥ 0

⇒ σp ≤ w1 σ1 + (1 − w1 ) σ2

Intuitively, the standard deviation of the portfolio, σp is smaller than the weighted average of σ1
and σ2 due to gains from diversification. Now, let µp be fixed, which implies that w1 is also fixed.
Then as ρ12 becomes smaller σp2 also becomes smaller. Hence, the portfolio frontier lies to the left
of Case 1 except at the point where

w1 = 0 ⇒ (σ2 , r2 )

w1 = 1 ⇒ (σ1 , r1 )

The figure shows the minimum variance frontier in this case. Note that the efficient frontier is the
upper contour of the minimum variance frontier.
6.2. MARKOWITZ PORTFOLIOS 101

Figure 6.2: Two imperfectly correlated assets

.

6.2.3 Case 3: Perfect negative correlation (ρ12 = −1)

Substituting ρ12 = −1 into (6.3) yields

2
σp2 = w12 σ12 + (1 − w1 ) σ22 − 2w1 (1 − w1 ) σ1 σ2
2
= [w1 σ1 − (1 − w1 ) σ2 ]

⇒ σp = ± [w1 σ1 − (1 − w1 ) σ2 ]

Then there are two cases depending on the sign of w1 σ1 − (1 − w1 ) σ2 :

1. if w1 σ1 − (1 − w1 ) σ2 ≥ 0, then

σp = w1 σ1 − (1 − w1 ) σ2
σp + σ2
⇔ w1 =
σ1 + σ2
which, in turn, implies that
C D
σp + σ2 σp + σ2
µp = r1 + 1 − r2
σ1 + σ2 σ1 + σ2
σ2 r1 σ1 − σp
= r 1 + σp + r2
σ1 + σ2 σ1 + σ2 σ1 + σ2
σ2 σ1 r2 − r1
= r1 + r2 − σp
σ1 + σ2 σ1 + σ2 σ1 + σ2
102 CHAPTER 6. PORTFOLIO THEORY

2. if w1 σ1 − (1 − w1 ) σ2 < 0, then

σp = − [w1 σ1 − (1 − w1 ) σ2 ]
σ2 − σp
⇔ w1 =
σ1 + σ2

which, in turn, implies that

C D
σ2 − σp σ2 − σp
µp = r1 + 1 − r2
σ1 + σ2 σ1 + σ2
σ2 r1 σ1 + σp
= r 1 − σp + r2
σ1 + σ2 σ1 + σ2 σ1 + σ2
σ2 σ1 r2 − r1
= r1 + r2 + σp
σ1 + σ2 σ1 + σ2 σ1 + σ2

More succinctly, we have G

M
M σp +σ2
if w1 ≥ σ2
M
H σ1 +σ2 σ1 +σ2
σ2 −σp σ2
w1 = if w1 <
M
M σ1 +σ2 σ1 +σ2
M
I σ +σ σ1
2 2 if w1 + σ1 +σ2 r 2
G
H r 2 −r 1
if w1 ≥ σ2
σ1 +σ2 σ1 +σ2
+σp
I r 1 −r 2
if w1 < σ2
σ1 +σ2 σ1 +σ2

Note that
σ2 r 2 − µp σ2
w1 ≥ ⇔ ≥
σ1 + σ2 r2 − r1 σ1 + σ2
(r2 − r1 ) σ2
⇔ r2 − ≥ µp
σ1 + σ2
σ2 σ1
⇔ r1 + r 2 ≥ µp
σ1 + σ2 σ1 + σ2
Graphically, the efficient frontier is shown in bold in Figure 6.3.

6.2.4 Case 4: One riskless and one risky asset

Suppose asset 1 is riskless and asset 2 is risky such that r1 = rf , σ1 = 0 with r2 > rf . Substituting
the values into (6.3) yields
2
σp2 = (1 − w1 ) σ22

⇒ σp = (1 − w1 ) σ2
σ2 − σp
⇒ w1 =
σ2
6.2. MARKOWITZ PORTFOLIOS 103

Figure 6.3: Two perfectly negatively correlated assets

.

Substituting above into (6.2) gives

C D C D
σ2 − σp σ2 − σp
µp = r1 + 1 − r2
σ2 σ2
r1 r2
= r 1 − σp + σp
σ2 σ2
r2 − rf
= rf + σp
σ2
Graphically, the efficient frontier is shown in bold in Figure 6.4.

6.2.5 Case 5: n risky assets (imperfectly correlated)

T
Suppose the returns of n risky assets are given by a vector r = (r1 , . . . , rn ) which are jointly
T
normal with mean µ = (µ1 , . . . , µn ) and variance-covariance matrix Σ = {σij }n×n′ σii = σi2 . The
T
portfolio is given by the vector ω = (ω1 , . . . , ωn ) , where ωi indicates the portfolio share of asset i
2n
and i=1 ωi = 1. Then

• Portfolio return, Rp , is given by

n
)
Rp = ω i Ri = ω T r
i=1

• Expected return, µp , is given by

. /
µp = E ω T r = ω ′ E(r) = ω T µ
104 CHAPTER 6. PORTFOLIO THEORY

Figure 6.4: One riskless and one risky asset

.

• Portfolio variance, σp2 , is given by

?. /. /@
σp2 = E ω T r − ω T µ rT ω − µT ω
? . /@
= ω T E (r − µ) rT − µT ω

= ω T Σω

Again, we see that diversification leads to lower portfolio variance. To see this, let ω = (1/n . . . , 1/n)T .
Then # &# &
σ11 ··· σ1n 1
1 ! "$ '$ ' 1
$ .. .. .. '$ .. '
σp2 = 1 ... 1 $ . . . '$ . '
N % (% (N
σn1 · · · σnn 1
# &
n n
1 %) ) (
= 2 σji
N j=1 i=1

Let σvmax = max σii and σcmax = max σij , then

1 max N − 1 max
σp2 ≤ σ + σc .
N v N
Thus, as n → ∞, σp2 tends to the covariance. In other words, risks can be fully diversified with
sufficiently large portfolio but the systematic risks (covariance) cannot be diversified.
6.2. MARKOWITZ PORTFOLIOS 105

Suppose initially there are 2 assets, A and B. Adding another asset C would (not always)
expand the investment opportunity set as can be seen from Figure 6.5. In general, a portfolio with
n risky assets would have a ’bullet’ shaped opportunity set and typically moves the efficient frontier
to the left.

Figure 6.5: More than two risky assets

.

6.2.6 Case 6: n risky assets and a risk-free asset

Suppose A and B are the bounds of the investment set of all risky assets. Then adding a risk-free
asset would imply that the efficient set would be the line M-T in Figure 6.6, a line tangential to the
risky asset frontier at T. If we allow short sales of the risk-free asset, the efficient frontier expands
beyond T(dotted line).

6.2.7 Minimum variance frontier

The efficient frontier of n assets (one of them may be risk-free asset) is the part of the undominated
portfolios of the minimum variance frontier. The latter is the solution to the following maximisation
problem
n )
) n
min wi wj σij
w1 ,...,wn
j=1 i=1
106 CHAPTER 6. PORTFOLIO THEORY

Figure 6.6: n risky assets and a risk-free asset

.

subject to
n
) n
)
wi ri = µ, wi = 1
i=1 i=1

It is usual to form the Lagrangian as

n )
n
E n
F
) )
L = wi wj σij − λ1 wi r i − µ
j=1 i=1 i=1

2n
and then to use the constraint i=1 wi = 1 on the resulting first-order conditions to obtain the
optimal weights.
Alternatively, using duality theory, one can maximise the mean for given levels of standard
deviation. Other constraints may be imposed to the portfolio problem:

• non-negativity constraints (wi ≥ 0)

• stock holdings should not fall below a certain level (wj ≥ Vj /Vp )

• home bias puzzle,

• transaction costs, etc.

6.3. CAPM 107

6.2.8 Two-Fund Separation Theorem

The optimal portfolio is naturally defined, as in microeconomics, as the portfolio that maximises the
investor’s mean-variance utility. Since the indifference curves are convex to the origin, for the case
of risky assets and a risk-free asset, all optimal portfolios should be one the linear efficient frontier.
Given different preferences, the maximum will be at different points on this linear frontier. Crucially,
all optimal portfolios would consists of combinations of the risk-free asset and the tangency portfolio
T as in Figure 6.6; this is known as the two-fund separation theorem. Consequently, optimal portfolio
of risky assets may be identified independent of the knowledge of risk preferences.

Theorem 6.6. (Two-fund separation). If every agent’s risk tolerance is linear with common slope
γ, then date- 1 consumption plans at any Pareto optimal allocation lie in the span of the risk-free
payoff and the aggregate endowment.

6.2.9 Capital markets line

In the presence of a risk-free asset, the locus of optimal portfolio is a line passing through two
points: the risk-free asset and the tangency portfolio. The slope of this line is called the Sharpe
ratio and is given by
r T − rf
SR =
σT
It gives the price of risk, ie, how much one should be compensated in expectation for undertaking
an additional unit of risk. The lower the Sharpe ratio for an asset, the less efficient it is by itself.

6.3 CAPM
CAPM is a theory of financial equilibrium. There is no attempt to relate returns with production
and thus, there is no interaction between the real and the nominal sectors of the economy. It
involves no derivation of supply and demand functions of assets and assumes that supply equals
demand of assets; ie, that observed prices are equilibrium prices (economy-wide asset holdings are
investors’ optimal asset demands). CAPM is able to determine the asset returns relationships so
that equilibrium prices coincide with observed asset prices and provides a specific characterisation
of risk-the only risk that matters is the one coming from the movement of the market as a whole.
Assumptions:
108 CHAPTER 6. PORTFOLIO THEORY

• Each investor maximises a mean-variance utility U (µp , σp ) with U ′ > 0 and U ′′ < 0

• All investors have common time horizon and homogenous belief about expected returns and
variances of existing assets.

• Short sales of the risk-free assets are allowed.

• If there is no risk-free asset present, no-short-sales restrictions on risky assets.

• The endowments of all agents are traded.

Proposition 6.7. All existing assets must belong to T, hereafter called the market portfolio.

Proof. Suppose not. In equilibrium, if some assets were not in T , there would be no demand for
them. However, all assets exist in positive supply. Recall that agents should hold T in equilibrium
or combination of T and the risk-free asset by the two-fund separation result.

Proposition 6.8. The share of assets in T is equal to

pq
2i j
j pj q j

where pj is the price of security j and qj is the quantity traded. Thus, the tangency portfolio is the
market portfolio.

Proof. If a market equilibrium is to exist, the prices of all assets must adjust until all are held
by investors. Thus, prices must be established so that the supply of all asset equals the demand
for holding them. Consequently, the market portfolio will consist of all marketable assets held in
proportion to their value weights.

Note that the market portfolio is efficient since it is on the efficient frontier.

Definition 6.9. The locus of all optimal portfolios are called the capital market line, which is
linear when there are n risky assets and a risk-free asset. The CML is given by

r M − rf
r p = rf + σp
σM

Notice that CAPM identifies the T portfolio as being the market portfolio M .
6.3. CAPM 109

6.3.1 Security market line

Consider a portfolio where a proportion (1 − a) of wealth is invested in j and the rest in the market
portfolio, M . Then

µp = arM + (1 − a)rj
? @1
σp = a 2 σM
2
+ (1 − a)2 σj2 + 2a(1 − a)σjM 2

As we vary a, we would trace the locus of the efficient frontier when there is only asset j in the
market (the dotted line in Figure 6.7).

Figure 6.7: CML and asset j

.

In equilibrium, the market portfolio already has exposure to risky assets (Proposition 6.8).
Thus, (1 − a) represents the excess demand for an individual risky asset j, which must be 0 at the
110 CHAPTER 6. PORTFOLIO THEORY

equilibrium; ie, a = 1 at equilibrium. Then

-
∂µp --
= rM − rj
∂a -a=1
-
∂σp -- 1? 2
@
- = 2aσM − 2(1 − a)σj2 − 2aσjM + 2(1 − a)σjM
∂a a=1 2
? 2 2 @− 1 --
× a σM + (1 − a)2 σj2 + 2a(1 − a)σjM 2 -
a=1
1. 2 / . 2 /− 12
= 2σM − 2σjM σM
2
σ 2 − σjM
= M
σM
We want to know the slope of this line in (σ, µ) space:
C D−1
∂µp ∂µp ∂σp rM − rj
= = 2
∂σp ∂a ∂a σM − σjM /σM
Recall that this must have the same slope as the capital market line:
r M − rf rM − rj
= 2
σM σM − σjM /σM
rM − rj 1
⇔ 2 = (rM − rf ) 2
σM − σjM σM
σjM
⇔ rM − rj = (rM − rf ) − (rM − rf ) 2
σM
σjM
⇔ rj = rf + (rM − rf ) 2
σM
2
Above is the equation for the security market line. Alternatively, define βj ≡ σjM /σM , then
C D
r M − rf
r j = rf + β j σM
σM
where βj σM represents the systematic/undiversifiable risk. βj itself is a quantity of risk and in
particular, it measures the contribution of asset j to the portfolio risk. We can also interpret
rM − rf as the price of risk.

Remark 6.10. Every investor holds the market portfolio. The relevant risk for the investor is
thus the variance of the market portfolio. Consequently, what is important to the investor is the
contribution of asset j to the risk of the market portfolio; that is, the extent to which the inclusion
of asset j into the overall portfolio increases the latter’s variance. This marginal contribution is
measured by βjM σM . Investors, need to be compensated to take this portfolio which higher risk by
taking higher expected returns.
6.3. CAPM 111

Remark 6.11. The CML and SML suggest that an efficient portfolio is one for which all diversifi-
able risks have been eliminated. In other words, for an efficient portfolio total risk equals systematic
risk.

Figure 6.8: Security market line

.

6.3.2 Zero-beta CAPM

Definition 6.12. Efficient portfolios are those frontier portfolios for which the expected return
exceed the expected return of the minimum variance portfolio.

Proposition 6.13. Any convex combination of frontier portfolio is also a frontier portfolio.

Proposition 6.14. For any frontier portfolio p, except the minimum variance portfolio, there
exists a unique frontier portfolio with which p has zero covariance. We call this portfolio the zero
covariance portfolio relative to p and denote its vector of portfolio weights by ZC(p)

Figure 6.9 shows ZC(p) graphically.

Our purpose is to find a similar equation to the SML equation found in the regular CAPM
model. To do so, we assume that both the expected returns and the variance-covariance matrix are
112 CHAPTER 6. PORTFOLIO THEORY

Figure 6.9: Zero-beta CAPM

.

given (as in Markowitz portfolio theory). After some calculations, we obtain that for any portfolio
q , the following holds

? @ ? . /@
E (P rp ) − E r̃ZC(p) .
rq ) = E r̃ZC(p) + βpq E (P

In equilibrium, the expected return vector is endogenous, in contrast to what we assumed. Thus,
we need to identify a particular portfolio as being a frontier portfolio without specifying a priori
the expected return vector and variance-covariance matrix of its constituent assets. Recall that,
in equilibrium, investors will not be maximising utility unless they hold mean-variance efficient
portfolios. However, this implies that, in equilibrium, the market portfolio, which is a convex
combination of individual portfolios is also on the efficient frontier. Therefore p can be chosen to
be the market portfolio M. Then the equation becomes
? @ ? . /@
rM ) − E rPZC(M )
rq ) = E rPZC(M ) + βM q E (P
E (P

and for any individual asset j :

? @ ? . /@
E (P rM ) − E r̃ZC(M ) .
rj ) = E r̃ZC(M ) + βM j E (P

Note that these only hold in equilibrium; zero-beta CAPM is an equilibrium theory. The name
6.3. CAPM 113

’zero-beta’ CAPM comes from the fact that

? @
Cov rPM , rPZC(m)
βZC(M ),M = 2 =0
σM

by construction.

6.3.3 Pareto optimality

Given that a risk-free rate exists, all agents optimise on the capital market line. Thus, the marginal
rates of substitution (the slope of the indifference curve) are all equal to the Sharpe ratio, which
is the slope of the CML. Consequently, the allocation is Pareto optimal. Given their risk-aversion,
they will choose a different exposure to the market portfolio.
However in the zero-beta CAPM, we do not have a riskless asset and thus agents will optimise
on the minimum variance frontier. Since the frontier is concave (bullet shape), their MRSs will not
be equal and hence the allocation is not Pareto optimal.
114 CHAPTER 6. PORTFOLIO THEORY
Chapter 7

CAPM GEI

We focus on the CAPM-GEI model, which is essentially a GEI economy with the following assump-
tions:

• L=1

• Mean-variance utility; eg,

S
) C D
αh x2s
uh (x) = uh0 (x0 ) + γsh xs −
s=1
2

This also says that people have v.N-M. quadratic (expected) utilities with state probabilities
γsh . This allows us to rewrite utility of anything as some function of its mean and its variance
(mean-variance utility). In particular, note that the marginal utility is linear (for contingent
consumption).

• γsh = γs for all h and s

This ensures that the objective probabilities are not subjective; ie, every agent believe with
the same probability that the state will be s.

• eh ∈ Span(A) for all h

States that agents’ initial endowments are in the span of the assets (ie, they are tradable).
This is the non-generic assumption that is necessary to ensure Pareto optimality.

115
116 CHAPTER 7. CAPM GEI

• Preferences are monotonic. This is guaranteed if

)
1 − αh ehs > 0∀h ∈ H, ∀s ∈ S
h∈H

where 1 − αh xs is the marginal utility from period- 1 consumption. (Since quadratic utility
eventually declines with more consumption, we must suppose that the utilities are chosen
such that agents still want to eat more even if they consume the whole endowment.)

• 1 ∈ Span(A)

This states that there is a riskless asset in the economy. In this one-good model, the natural
definition of riskless asset is that of an asset which pays on unit of the good in every state of
nature.

Note that the first three assumptions allows us to argue agent’s choice in terms of mean and variance.
Such mean-variance preference is achieved by assuming quadratic utilities, which can be written as
C D
h 1 2
EU = u0 (x0 ) + E xs − αh xs
2
1 . /
= uh0 (x0 ) + E (xs ) − αh E x2s
2
h 1 A 2
B
= u0 (x0 ) + E (xs ) − αh E (xs ) − Var (xs )
2
5 0 1 6
h
Definition 7.1. A CAPM-GEI equilibrium is a q, x , θ h
such that q ∈ RJ , xh ∈ R×RS ,
h∈H
h
and θ ∈ R satisfying;
J

2H 2H
1. Goods market clear: h=1 xh = h=1 eh
2H
2. Securities market clear: h=1 θh = 0

3. Budget constraints are satisfied:

0 1 0 ∗
1
xh , θ h ∈ Bh (q) = xh ∈ RS : xh0 + qθ ≤ eh0 , xh ≤ eh + Aθ h

for all h ∈ H, and

. /
4. Utilities are maximised: x ∈ Bh (q) ⇒ uh (x) ≤ uh xh for all h ∈ H.

Remark 7.2. The CAPM-GEI equilibrium solves

max U h (x) (7.1)

{x,θ}∈B′′ (q)
7.1. CAPM EFFICIENCY THEOREM 117

7.1 CAPM Efficiency Theorem

Remark 7.3. This theorem says that even if the number of assets is much smaller than the number
of the states (so that it is impossible to insure most risks), equilibrium allocations are fully Pareto
optimal, if the riskless asset is tradeable, and if every agent can sell shares in his future income.
Without the riskless asset, equilibrium allocations may be terrible. By adding the riskless asset,
equilibrium jumps to the Pareto frontier, even though there are many other missing asset markets
that have not been added.

Theorem 7.4. Suppose at a CAPM-GEI equilibrium, it holds that xh ≫ 0. Suppose, in addition,

* +
that there is a riskless asset. Then xh h∈H is Pareto optimal.

Proof. The agent’s maximisation problem is given by

. /
max U h xh
xh ∈Bh (p,q)

The Lagrangian is then

9 ; 9 ;
J
) S
) J
)
. /
L =U h
x h
− µh0 :xh0 − eh0 + qj θjh < − µhs :xhs − ehs − ajs θjh <
j=1 s=1 j=1

The first-order conditions are

∂L ∂U h (xh )
∂xs =0⇔ = µhs ∀s ∈ {0, 1, . . . , S}
∂xs
∂L
2S
∂θjh
= 0 ⇔ µh0 qj = s=1 µhs ajs

Note that µs > 0 for all s = 0, 1, . . . , S (constraints are binding). We thus obtain
S
1 ) h j 1
qj = µ a = h µ h · γ aj
µ0 s=1 s s µ0

where C D
h µh1 µh . /
µ = ,..., S , µhs = γs 1 − αh xhs
γ1 γS
The projection of µh /µh0 onto Span[A] is given by a unique vector µ∗ ∈ RS (by the Projection
Theorem). Notice, in particular, that the same unique vector is obtained for all h ∈ H.
We may write
µh = 1 − α h xh
118 CHAPTER 7. CAPM GEI

It’s clear from the fact that the budget constraints all bind (by weak monotonicity) that xh is a
linear combination of e ∈ Span[A] and Aθ h ∈ Span[A] hence xh ∈ Span[A]. since 1 ∈ Span[A] by
assumption, it follows that

µh ∈ Span[A]

But we already established that µ∗ is the unique projection of µh /µh0 onto Span[A], so

λh µ h = µ ∗

where λh = 1/µh0 . Finally, note that Pareto optimality in the case of interior solution is the same as
the ratio of marginal utilities being equal across agents. It’s clear from the expression about that
this would be the case.

7.1.1 CAPM Mutual Fund Theorem

Theorem 7.5. Suppose at a CAPM-GEI equilibrium, it holds that xh ≫ 0. Then

xh ∈ Span[1, e]∀h ∈ H

2
where e = h∈H eh

Proof. From the proof of the CAPM Efficiency Theorem, we have

. /
µ ∗ = λh 1 − α h x h

We may rearrange above to obtain that

1 1
xh = 1 − h h µ∗
αh α λ

Thus, every agent’s final consumption is a combination of just two payoff vectors, neither of which
are riskless. It remains to show that xh ∈ Span[1, e] Summing over all agents, we have
7.1. CAPM EFFICIENCY THEOREM 119

) )
e≡ eh = xh
h∈H h∈H
)C 1 1
D
∗
= 1 − µ
αh α h λh
h∈H
E F E F
) 1 ) 1
∗
=1 −µ
αh α h λh
h∈H h∈H

∗
. −1
/
= 1(α) − µ β (7.2)

⇔ µ∗ = β(α1 − βe)

with α > 0 and β > 0. Hence, µ∗ is a linear combination of e and 1 . Therefore, it must be that
µ∗ ∈ Span[1, e]

Remark 7.6. We need not assume that 1 belongs to the span of A. We could, instead, assume
∗
that we can always take the projection of the riskless asset onto the span of A obtaining 1 in
which case we would have
A B
∗
xh ∈ Span 1 , e ∀h ∈ H

However, we do still assume that initial endowments are tradable; ie, eh ∈ Span[A] for all h ∈ H.
To see this, recall that
µh = 1 − αh x−h

Denote µ∗ as the projection of µh onto Span[A], then, by definition

∗
µ∗ = 1 − α h xh

since xh ∈ Span[A]. To see the latter, recall from the budget constraint that

xh − eh = Aθ

Hence, xh is a linear combination of eh ∈ Span[A] and Aθ ∈ Span[A]. Hence, xh ∈ Span[A].

Rearranging yields
1 ∗ 1
xh = h
1 − h µ∗
α α
Summing across agents and using the market clearing condition yields
! "
∗
e = α 1 − µ∗
120 CHAPTER 7. CAPM GEI

2 A B
h ∗ ∗ ∗ ∗
where α = h 1/α . That is, µ is a linear combination of 1 and e, such that µ ∈ Span 1 , e .
∗
Finally, recall that xh is a linear combination of 1 and µ∗
. /−1 T
, = AT A
Remark 7.7. To project 1 onto A, we regress 1 = Aγ by OLS. Recall that γ A 1
∗
so that 1 (which is the OLS prediction of 1) is given by

∗ . /−1 T
1 ≡ Proj∗A (1) = A,
γ = A AT A A 1

∗ ∗
since 1 is a linear combination of columns of A, it follows that 1 ∈ Span[A]. Obviously, if
∗
1 ∈ Span[A], then 1 = 1

7.1.2 CAPM Security Market Line

Proposition 7.8. Suppose at a CAPM-GEI equilibrium, it holds that xh ≫ 0 for all h. Then,
there exists r > −1 and δ > 0 such that, for all j ∈ J
. /
Eγ aj − δ Covγ e, aj
qj =
1+r
More generally, for any z ∈ Span[A]
Eγ z − δ Covγ (e, z)
q(z) =
1+r
Proof. Recall from the proof of the CAPM Efficiency Theorem:
1 h
qj = µ ·γ aj = µ∗ · γaj
µh0
and from the proof of the Mutual Fund Theorem:

µ∗ = β(α1 − βe)

Combining the two yields qj = β(α1 − e) · γaj . By definition, we have

Covγ (x, y) = x ·γ y − Eγ (x)Eγ (y)

where x, y ∈ RS . Thus,
. / A ! " B
∗
qj = Eγ [β(α1 − e)]Eγ aj + Covγ β α1 − e , aj
? @ . / . /
= β αEγ (1) − Eγ (e) Eγ aj − β Covγ e, aj
. / . /
= β (α − Eγ (e)) Eγ aj − β Covγ e, aj
7.1. CAPM EFFICIENCY THEOREM 121

where
S
)
Eγ (1) = 1 · γ1 = γs = 1
s=1

Define
1
α − Eγ (e) ≡ > 0, δ ≡ β(1 + r) > 0
1+r
Then, . / . /
. i / Eγ ai − δ Covγ e, aj
qj a =
1+r
Eγ (z) − δ Covγ (e, z)
⇒ q(z) = ∀z ∈ Span[A]
1+r

7.1.3 CAPM Invariant Pricing Theorem

Suppose our economy dos not feature a riskless asset; then bad allocations may occur. Now, imagine
somebody introducing a new asset. Clearly, equilibrium allocations will change too but the question
is: what happens to the price of the old asset? One might think that this depends on whether the
new asset is a substitute or a complement to the original asset. In fact, we get a very clear answer in
the CAPM model. The prices of the old asset simply do not change. Put in a slightly more general
form, if we have two CAPM economies with at least two common assets, then at equilibrium, the
price ratios between the common assets in the two economies will be the same. We do no even
suppose that initial endowments are traded. In particular, we take the original economy A and add
one more asset to it, the relative prices are going to remain unchanged. This is a very interesting
knife-edged case in real-world markets, where new assets are continuously introduced; if a new kind
of mortgage asset, say, is created, what will that do the stock market or the bond market? If we
believe that utilities are quadratic, then we know that relative prices are not going to change.
0* + 1 0* + 1
Theorem 7.9. Let u h , eh h∈H
,A and u h , eh h∈H
,B be two GEI satisfying:

1. one good (L = 1),

2. quadratic utilities,
. /
3. common priors γsh = γs for all h ∈ H , and

4. monotonicity.
122 CHAPTER 7. CAPM GEI
5 0 1 6
j′ j′
j j
Suppose it happens that a = b and a = b . Suppose finally that at the respective GEI, p, q , A
xhA , θ hA
h∈H
5 0 1 6
and p, qB , xhB , θ hB with xhA , xhB ≫ 0 for all h ∈ H. Then,
h∈H

qjA qjB
=
qjA′ qjB′
Proof. From the first-order condition:
. / µh 1
µh = 1 − α h x h ⇔ h = h 1 − x h
α α

and it must also be the case that (by monotonicity)

E F
) µh ) 1 )
= 1 − xh = α1 − e
αh αh
h∈H h∈H h∈H

which does not depend on anything but preferences (ie, is the same under A. and B). Recall that

qjA = µ ·γ aj = β(α1 − e) · γaj

1 A ) µh
⇔ qj = (α1 − e) · γaj = · γaj
β αh
h∈H
) µh
qjB = µ ·γ bj = · bj
αh
h∈H

It follows that 2 µh
qjA h∈H αh · γa
j
(α1 − e) · γaj
= 2 =
qjA′ µ h
h∈H αh · γa
i′ (α1 − e) · γai
2 µh
(α1 − e) · γbj h∈H αh · γb
j qjB
= = 2 =
(α1 − e) · γb′ µh
h∈H h · γb
j′ qjB′
α
j j i ′
since a = b and a = b , by assumption.

Remark 7.10. We need not assume that initial endowments are traded (e ∈ Span[A]).

Remark 7.11. If time-zero consumption exists, then price relative to today’s consumption might
change, given the possibility of saving more as a new asset is available. In this case, introducing a
new asset might make somebody worse off than before; where no utility for today’s consumption
is present, adding a new asset is always going to make everyone better off (there is always the
possibility of replicating the allocation corresponding to the old, poorer asset structure, given that
the relative prices of the old assets stay the same).
7.1. CAPM EFFICIENCY THEOREM 123

Remark 7.12. The main lesson of CAPM is that if only we add a riskless asset, everybody is
better off. Yet, in the US, there really is no riskless asset, or even an attempt at creating one (when
inflation is taken into account).

7.1.4 Multiple Commodity CAPM

Once we move over to a multi-commodity world, the question is that of what a riskless asset should
be.

• Taking an asset which pays the same amounts of goods in each state does not make sense as
relative prices do matter in this new setup.

• Maybe the right thing to do is to have a bundle of commodities in which this asset pays off;
we pick the right index and then have an asset which guarantees the correct purchasing power
which allows us to buy the same bundle in every state. How could this possibly be a sensible
riskless asset, given that, after all, prices could be so different across states that supplies of
the goods would differ from one state to the other?

In this setting, we have to revise two of the assumptions we made earlier:

2’. Let
S
) 5 6
1 2
uh (x) ≡ uh0 (x0 ) + γs vs (xs ) − αh [vs (xs )]
s=1
2

where vs : RL ,→ R is smooth, concave, and homogeneous of degree 1 for all h ∈ H. Note

that vsh = vs for all h ∈ H.

5’. 1 − αh vs (es ) > 0. That is, in every state s, there is a common vs which is like an index
that says, given a bundle of goods xs′ (vs (xs ) is just some number, that way we are led
back (in a sense to be specified later) to the one commodity world.

Theorem 7.13. 5 (Multiple Commodity

6 CAPM Theorem). Consider a GEI satisfying (2′ ) , (3), (4),
0 1
and (5′ ) . Let p, q, xh , θ h be a GEI for which p□xh ≫ 0 for all h ∈ H. Suppose there is
h∈H
an asset r ∈ Span(A) of the form
! "
r= e1 eS
v1 (e1 ) ··· vS (eS )

* +
Then xh h∈H is Pareto optimal.
124 CHAPTER 7. CAPM GEI
* +
Remark 7.14. We can see from xh h∈H that the riskless asset does reflect what is available; ie,
it is not a fixed index. What is also interesting in this model is that the riskless asset is not going
to yield the same purchasing power in terms of a fixed commodity basket in every state.
Chapter 8

APT

In many applications (mainly in corporate valuation), we do not calculate the expected value under
the risk-neutral measure but use the subjective probabilities to get an expectation about the future
payoff (cashflows) of a security (or a project). We then discount this expectation with an adjusted
return for risk. The most common way to calculate this return is via CAPM, which is an financial
equilibrium model.

Arbitrage pricing theory (APT) provides an expression for deriving the risk-adjusted return
via arbitrage arguments without considering explicitly an equilibrium model. APT relies in the
identification of risk factors in the economy that capture the risk-premium for all securities in the
economy. Intuitively, the importance of each factor will defer from security to security due to
different risk profiles / exposures. The underlying logic in APT is very similar to the fundamental
logic of the Arrow securities. It replaces the underlying structure based on Arrow securities that
payout exclusively in a certain state of the world with other fundamental securities exclusively
remunerating some form of risk taking.APT replaces the concept of the state of nature with the
hypothesis that there exists a stable set of factors that are essential and exhaustive determinants
of all asset returns. The primitive security is then defined as a security whose risk is exclusively
determined by its association with one specific risk factor and totally immune from the association
with any other risk factor. Thus, the difference from equilibrium model is that the prices of
fundamental securities are not derived from primitives but will be deduced empirically from observed
asset returns without attempting to explain them.

125
126 CHAPTER 8. APT

8.1 Factor Pricing

Suppose there are K contingent (ie, the payoff depends on the state of the world realised) claims
f1 , . . . , fk called factors, which capture the total risk in the economy. Factors such as GDP growth,
government bond yield with different maturities, the return on the market portfolio (ie, the portfolio
of all assets traded in the economy), the small caps or the book-value-to-price portfolios (Fama-
French three-factor model) have been used in the empirical literature. Many of those factors are
not appropriate to use when implementing the APT. We will comment on the properties that these
factors should have later and for the time being, we assume that there are K factors that capture
the total risk in the economy and can be used to calculate the risk premium an investor would
demand to buy a security due to risk aversion. Let xj be the payoff of security j and pj its price.
We define by rj = xj /pj − 1, the rate of return of security j. APT tells us that

rj = aj + b1j f1 + · · · + bKj fK + ej

where bkj is the loading factor (ie, how important factor k is for security j) , ej is the error (or
the realised return not captured by factors) and E (ej ) = 0. The expected return which is required
when discounting expected cashflow is

E (rj ) = aj + b1j E (f1 ) + · · · + bKj E (fK ) = aj

where we have normalised the expected value of the factors to o by subtracting any factor means.
The reason why E (ej ) is quitive: The factors, by definition, capture all the risks in the economy;
thus if one forms a portfolio of the factors with weights given by their respective loading factor, one
can wipe out the risk arising from ej (essentially a hedging argument). Under some calculations
shown later, we obtain that

E (rj ) = rf + β1j [E (rf1 ) − rf ] + · · · + βKj [E (rfK ) − rf ]

The weights βkj is now given by

Cov (rj , rfk )
βkj =
Var (rfk )

which can be estimated by a linear regression. Thus, loadings may be estimated using market data.
8.2. SINGLE-FACTOR APT & CAPM 127

8.2 Single-Factor APT & CAPM

Recall that CAPM is an equilibrium asset pricing model. Its main prediction is capture by the
Security Market Line (SML), given by the equation

E (rj ) = rf + βj [E (rM ) − rf ]

where rM is the return on the market portfolio (eg, S&P 500 index ) and βj capture the covariance-
sensitivity of the return on the market portfolio with the return on asset j :

Cov (rM , rj )
βj =
Var (rM )

It is obvious that if we assume that the movement (or return) of the aggregate stock index is the
only factor affecting asset returns, CAPM is a single-factor APT model. Nevertheless, we have to
point out that CAPM suggests that the market portfolio is the only factor as an equilibrium result
starting from fundamentals and under certain assumption. On the other hand, APT says nothing
about which factor one should use, or more importantly, how many.

8.3 APT Example

Assume a two-factor model where security j ′ ’s factor loadings on f1 and f2 are β1j = 0.8 and
β2j = 0.4, respectively. This is like a portfolio with proportions of 0.8 of the portfolio f1 , o.4 of
the pure portfolio f2 and consequently, proportion −0.2 in the riskless asset. The expected rate of
return on security j should then be

E (rj ) = −0.2rf + 0.8E (rf1 ) + 0.4E (rf2 )

= −0.2rf + 0.8 [E (rf1 ) − rf ] + 0.4 [E (rf2 ) − rf ] + 0.8rf + 0.4rf

= rf + 0.8 [E (rf1 ) − rf ] + 0.4 [E (rf2 ) − rf ]

Thus, the APT equation can be seen as the immediate consequence of the linkage between pure
factor portfolios and complex securities in an arbitrage-free context. Note that diversifiable risk is
not priced in a complete or effectively complete world.
128 CHAPTER 8. APT

8.4 APT formally

Assume K factors and J assets. The APT can be written conveniently in vector notations:

r = a + Bf + e

where
E(e) = E(f ) = E (ef ′ ) = 0

E (f ′ ) = I

E (ee′ ) = D
! "T
a = a1 ··· aJ ∈ RJ
! "
B = b1 ··· bK ∈ RI×K
! "T
bk = bk1 ··· bkJ ∈ RI
! "T
f = f1 ··· fK ∈ RK
! "T
e = e1 ··· eJ ∈ RK

Thus, a is the vector of alphas for each asset, B is the J × K matrix of factor loadings for each
asset, f is the vector of factors, and eis the vector of diversifiable-idiosyncratic asset risks.

8.4.1 One factor Model

Assume a one-factor model (K = 1) without asset-specific (idiosyncratic) risk ( ie, ej = 0) , which

implies that
Cov (rj , f ) = Cov (aj + bj f, f )

To price assets with different values of b, consider a portfolio of two assets j and j ′ with bj ∕=
bj ′ , bj , bj ′ ∕= 0 formed by investing a fraction w in asset j and (1 − w) in asset j ′ . The return on this
portfolio is
r = w (aj + bj f ) + (1 − w) (aj ′ + bj ′ f )

= w (aj − aj ′ ) + aj ′ + [w (bj − bj ′ ) + bj ′ ] f

If we choose w∗ = −bj ′ / (bj − bj ′ ) , then

aj − aj ′
r ∗ = bj ′ + aj ′
bj ′ − bj
8.4. APT FORMALLY 129

This return is certain, thus it must equal the risk-free return; ie,
aj − aj ′
rf = bj ′ + aj ′
bj ′ − bj
By symmetry, we also have that
aj ′ − aj
rf = bj + aj
bj − bj ′
aj − aj ′
= bj + aj
bj ′ − bj
Hence,
a j ′ − rf aj − aj ′ a j − rf
= =
bj ′ bj − bj
′ bj
We define
a j ′ − rf a j − rf
λ≡ =
bj ′ bj
since E (rj ) = aj ′ 1
E (rj ) = rf + bj λ (8.1)

where λ is some constant known as the factor risk premium. Using the market portfolio to express
λ (ie, notationally, set j = M ) and rearranging, we obtain that
E (rM ) − rf
λ=
bM
Note that bj in a one-factor model can be obtained by running the following OLS regression

rj = a j + bj f

which yields an estimate of bj , ,bj , as

Cov (rj , f )
b,j =
Var(f )
since Var(f ) = 1 (ie, it is not a random variable) and that there is only factor fM in this case, the
factor loading for the market portfolio, bM , is given by

bM = Cov (rM , f )
C D
rM − a M
= Cov rM ,
bM
1
= Var (rM )
bM
= pM Var (rM )
1 Note that
aj − r f
E (rj ) = rf + bj λ = rf + bj = aj
bj
130 CHAPTER 8. APT

which implies that

E (rM ) − rf
λ=
pM Var (rM )

where pM is the price of the market portfolio. Finally, substituting above into (8.1), we have

E (rj ) = rf + bj λ
E (rM ) − rf
= rf + Cov (rj , fM )
pM Var (rM )
Cov (rj , rM ) E (rM ) − rf
= rf +
bM pM Var (rM )
E (rM ) − rf
= rf + pm Cov (rj, , rM )
pM Var (rM )
Cov (rj , rM )
= rf + [E (rM ) − rf ]
Var (rM )
= rf + βj [E (rM ) − rf ]

which is the CAPM equation.

8.4.2 Multi-factor model

For multi-factor models, a similar analysis can be performed, provided that the factor loadings are
sufficiently different. Assume two factors and no idiosyncratic risk, then

rj = aj + b1j f1 + b2j f2

Choose a portfolio of the three securities, say, j = 1, 2, 3, then the return on this portfolio, p, is
)
rp = w j rj
j
)
= wj (aj + b1j f1 + b2j f2 )
j
) ) )
= w j aj + f1 wj b1j + f2 wj b2j
j j

2 2
If we select j wj b1j = j wj b2j = 0, then the portfolio is riskless. Thus, to exclude arbitrage

) )
w j a j = rf ⇒ wj (aj − rf ) = 0
j j
8.4. APT FORMALLY 131

The model can then be written as

# &# & # &
a − rf a 2 − rf a 3 − rf w1 0
$ 1 '$ ' $ '
$ '$ ' $ '
$ b11 b12 b13 ' $ w2 ' = $ 0 '
% (% ( % (
b21 b22 b23 w3 0

For the no-arbitrage condition to hold for any portfolios, the determinant of the matrix has to equal
zero (ie, no solution). Because the last two rows are not colinear, we have

E (rj ) − rf = aj − rf = λ1 b1j + λ2 b2j

2
The only remaining issue is to identify λ1 and λ2 . If we choose a portfolio with j wj b1j = 1 and
2
j wj b2j = 0, then
)
r f1 = w j aj + f1
j

= rf + f 1

⇒ E (rp1 ) − rf = λ1
2 2
Similarly, when j wj b1j = 0 and j wj b2j = 1

E (rf2 ) − rf = λ2

Thus, λ1 and λ2 can be thought of as the excess return per unit risk associated with factors f1 and
f2 :
E (rj ) − rf = λ1 b1j + λ2 b2j

= b1j [E (rf1 ) − rf ] + b2j [E (rf2 ) − rf ]

APT holds even when we allow for idiosyncratic risk though derivation is complicated and beyond
the scope of this course.

Theorem 8.1. If the returns on the risky assets are given by a K-factor linear model with bounded
residual risk and there are no (asymptotic) arbitrage opportunities,then there exists a linear pricing
model which gives expected returns with a mean-square error (defined as vj ) of zero. That is, there
are factors λ1 , . . . , λK such that
)
v j = a j − rf − bkj λj
k
and
1) 2
lim vj = 0
J→∞ J
j
132 CHAPTER 8. APT

8.5 APT IN GEI

The question is whether or not we can say something interesting about µ∗ in this setup. We shall
see that if we restrict attention to v.N-M preferences with common priors, then µ∗ is not arbitrary.
Knowing Span(p□A) and p□e), we can limit the location of µ∗ ∈ Span(p□A). CAPM will be
recognised to be a special case of APT, with special v.N-M quadratic preferences.

8.5.1 Assumptions

1. L = 1
2S
2. uh (x) = s=1 γs v h (x0 , xs ) , where v h is smooth, strictly concave, and monotonic;

3. γsh = γs for all h and s

4. eh ∈ Span(A) for all h

8.5.2 Definitions

Definition 8.2. We say that . ∈ RS is strong orthogonal to x ∈ RS , written x6γ & if and only if
Eγ (&|x) = 0

Remark 8.3. Note that G

H E & =0
γ
x6γ & ⇒
I x · γ& =0

A stronger version of this is the following.

Definition 8.4. F6γ & ≡ Eγ (&|f ) = 0 for all f ∈ F The definition of strong orthgonality is going
to be useful as it allows us to make mean-preserving spreads used in the proofs of the following
theorems.
0* + 1
Definition 8.5. The factor space of any GEI APT model economy u h , eh h∈H
,A is the
minimal subspace F ⊂ R satisfying
S

1. e ∈ F and

2. . ∈ F⊥ ∩ Span(A)
8.5. APT IN GEI 133

which implies that

Eγ (&|f ) = 0, ∀f ∈ F

i.e.,
F⊥ ∩ Span(A) = F⊥⊥ ∩ Span(A)

The factor space is then the minimal subset of RS which contains the aggregate endowment and such
that all the vectors contained in the span of Awhich are orthogonal to it are strongly orthogonal
to it.

8.5.3 APT Theorem

In general model (in which we do not make particular assumptions about utility functions and the
distribution of variables), what holds is the following.
0* + 1
Theorem 8.6. (APT Theorem). Let uh , eh h∈H , A be a GEI APT economy with factor span
5 0 1 6
h
h
F and equilibrium q, u , θ . Then, assuming xh ≫ 0 it holds that xh ∈ F for all h∈ H
h∈H
and µ∗ ∈ FIn particular, if & ⊥γ F then q(&) = 0

Proof. Note that

xh = eh + Aθ h ∈ Span(A)

with
h
xh = f + .h

where
h
f ∈F
G
H ∈ )
F γ ∩ Span(A)
&h = )
I = F γ ∩ Span(A)
However,
2 2 h 2
h xh = h f + h &h
⇔&=&+0
2
since h .h 6γ F ∋ &. Thus,
E F
) ) . /
π &h =0= q &h
h∈H h∈H
134 CHAPTER 8. APT
! "
implying that there exists h such that q .h ≥ 0.
h
! "
However, xh is a mean-preserving spread of f and by strict concavity of uh , uh xh <
C D
h h h
uh xh0 , f unless &h = 0. But if xh is affordable, so is f = xh −&h Hence, .h = 0 and xh = f ∈ F.
But then
) h
xh = e − e − f ∈ F
h∈H\h

and we can repeat the same argument for H\h. Thus, all xh ∈ F for all h ∈ H It follows that since
. /
uh is smooth, any . with . ∈ Span(A) and x ⊥⊥γ & must be priced with q .h = 0. To see why,
note that for all λ ∈ R
. / . /
uh xh0 , xh + λ. ≤ uh xh

µh · γ& = 0 ⇒ q(&) = 0

⇒ µ∗ · γ& = 0∀& ∈ F⊥γ ∩ Span(A)

µ∗ ∈ F

This result also sheds light onto an interesting problem which might arise in such models. One
might conceive of a theorem stating that given any economy and any µ∗ , it is always possible to
find preferences for the individuals such that µ∗ turns out to be the right pricing functional. In
this case, there would be no observable restrictions on µ∗ . This is clearly not the case, as we have
just demonstrated that, restricting attention to v.N-M preferences, there is a factor space and the
pricing functional has to be in that space. If we make more particular assumptions about utilities
and distribution, we get, as an interesting result, the following.
0* + 1
Corollary 8.7. Let u h , eh h∈H
,A ≡ E be a GEI economy with an infinite state space S.
h
Suppose that L = 1 and that e and a’ are normally distributed random variables for all h ∈ H, j ∈
J. Suppose, in addition, that each uh is v.N − M with respect to objective probabilities γ and that
eh ∈ Span(A) for all h ∈ H. In short, suppose E is AP T with normally distributed endowments
! " ! "
∗ ∗
and assets. Then xh ∈ Span 1 , e , µ∗ ∈ Span 1 , e for all h.

This corollary clearly shows how we obtain the same results as in GEI CAPM under quite
different assumptions.
8.5. APT IN GEI 135
! "
∗ ∗
Proof. Let 1 ≡ ProjγA 1. Let F = Span 1 , e and &6F. Then, if & ∈ Span(A),

0 = a∗ ·γ & = 1 ·γ & ⇒ Eγ (&) = 0

and
0 = e ·γ . ⇒ Covγ (e, &) = 0 ⇒ e ·γ &

given that the two variables are normally distributed. Then e6γ .. This shows, in particular, that
F is a factor space and that ! "
∗
µ∗ ∈ Span 1 , e
! "
∗
xh ∈ Span 1 , e

for all h.

Remark 8.8. If we assume q(e) > 0, we get as before

Eγ z − δ Covγ (e, z)
q(z) =
1+r

for any z ∈ Span(A).

136 CHAPTER 8. APT
Bibliography

David Cass. The structure of investor preferences and asset returns, and separability in portfolio
allocation: A contribution to the pure theory of mutual funds.

John C Cox, Stephen A Ross, and Mark Rubinstein. Option pricing: A simplified approach. Journal
of Financial Economics, 7(3):229–263, 1979.

Andrei Nikolaevich Kolmogorov. Grundbegriffe der wahrscheinlichkeitsrechnung. 1933.

Stephen A. Ross. The arbitrage theory of capital asset pricing. Journal of Economic Theory, 13:
341–360, 1976.

137