Asset Pricing I: Pricing Models and Derivatives

Lecture Notes 2023

Markus Brunnermeier
with edits by Caio Almeida, Hao Jin, and Moritz Lenel
Copyright © 2023 Markus Brunnermeier

These lecture notes are currently updated and maintained by Hao Jin ([Link][at][Link]).
Please contact Hao for comments and corrections.

This version: Nov 21, 2023. When this sentence is gone, this will be the complete lecture
notes.
 Contents

1 One Period Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.1 States and Securities 7
1.2 General Security Structure & Prices 10
1.3 LOOP & No Arbitrage 11
1.4 Exercises 12

2 Forwards and Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.1 Derivatives 13
2.1.1 Forward Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Back to Security Structures 16

2.3 Forwards Revisited 17
2.3.1 Prepaid Forwards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Forwards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.4 Options Revisited 19

2.4.1 Option Price Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4.2 Time to Expiration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.3 Strike Price . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.5 Exercises 21
3 State Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1 Pricing in One Period Model 23
3.1.1 State Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.2 The Fundamental Theorem of Finance . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.3 State Prices and Incomplete Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Asset Pricing Formulas 26
3.2.1 State Price Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.2 Stochastic Discount Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.3 Equivalent Martingale Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.4 State-Price Beta Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.5 Four Asset Pricing Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Recovering State Prices from Option Prices 29
3.4 Appendix 32
3.5 Exercises 33

4 Empirical Multi-Factor Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.1 Exercise 35

5 The Multi-Period Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Model Setup 37
5.2 Dynamic Trading and Market Completeness 40
5.3 The Multi-Period Stochastic Discount Factor 43
5.4 Martingales 45
5.5 Exercises 48

6 Multi-Period Model: Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

6.1 One-Period Binomial Tree 50
6.2 Two-Period Binomial Tree 52
6.3 The Relation with the Black-Scholes Model 53
6.4 Delta-Hedging 53
6.5 The Volatility Smile 54
6.6 Collateralized Debt Obligations 54
6.7 Exercises 55

7 Risk Preferences and Expected Utility Theory . . . . . . . . . . . . . . 56

7.1 State-by-State Dominance 56
7.2 Stochastic Dominance 57
7.3 von Neumann Morgenstern Expected Utility Theory 60
7.4 Risk Aversion, Concavity, Certainty Equivalent 64
7.5 Measures of Risk Aversion 66
7.6 Risk Aversion and Portfolio Allocation 68
7.7 Savings 71
7.8 Mean-Variance Preferences 72
7.9 Exercises 74

8 Mean-Variance Analysis and CAPM . . . . . . . . . . . . . . . . . . . . . . 77

8.1 Preliminaries 77
8.2 Simple CAPM with quadratic utility functions 78
8.3 The Traditional Derivation of CAPM 80
8.3.1 Mean-Variance Frontier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.3.2 Adding Risk-free Asset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.3.3 Two Fund Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.3.4 Equilibrium leads to CAPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.4 Exercises 89
 1. One Period Model

1.1 States and Securities

We begin our analysis of asset pricing by considering a simple setting in which there are two dates:
today (t = 0) and some future date (t = 1). We know everything about today’s state of the world
s = 0, but we don’t know which one of s = 1, . . . , S states will materialize in the future.

t =1
s=1

s=2
t =0
...

s=S

Figure 1.1: One period model with S states.

Under this setup, we can represent a security j ∈ {1, ... , J} as a vector

 j
x
j
 .1 
x =  .. 

,
j
xS

where xsj denotes the payoff of security j in state s.

We then define a security structure by a matrix X:

 
x11 x12 · · · x1J−1 x1J
 1 2 J−1 J 

 x2 x 2 · · · x2 x 2
 . .. .. .. 
  h i
X =  .. ..  = x1 x2 · · · xJ−1 xJ l
. . . .
 
x1 2 J−1 J 
 S−1 xS−1 · · · xS−1 xS−1 
xS1 xS2 · · · xSJ−1 xSJ

where column j collects the payoffs of security j across states.

8 Chapter 1. One Period Model

Arrow-Debreu security structure

An important example of a security structure is given by securities that are standard basis vectors,
which are called Arrow-Debreu securities. These securities pay one unit in a single state, and
pay zero in all other states. Figure 1.2 depicts an example with two possible states in the next
′
period (S = 2). The Arrow-Debreu security e1 = (1, 0) pays one unit in state 1 and nothing in state
2.
c2

Payoff cannot be replicated!

c1

′
Figure 1.2: Two states (S = 2) and a single Arrow-Debreu security e1 = (1, 0) .

Note that if this is the only security available in the market, we can replicate any security in the
horizontal axis by buying a quantity α ∈ R of e1 , but we cannot replicate any of the securities
outside of the horizontal axis (e.g. the red security). When the available securities are not enough
to replicate all securities in RS we say that markets are incomplete.

If we introduce another asset e2 = (0, 1)′ , all securities in R2 become replicable. In this case, we
say that markets are complete. Note that adding another asset x3 ∈ R2 does not benefit us in any
way, as we were already able to span the whole set R2 with the previous two. The security structure
generated by e1 and e2 is
" #
1 0
X= .
0 1

In general, given S possible states in t = 1 and J = S securities we call Arrow-Debreu security

structure the matrix
 
1 0 ··· 0
0 1 · · · 0
 
AD
X = . . .

.
.. 
.
. .. . . .
0 0 ··· 1

Note that 1) all payoffs (securities) are linearly independent vectors in RS , 2) markets are complete
by construction (since also J = S).
1.1 States and Securities 9

General security structure

Consider now a general security structure in R2 . For example, suppose there is only a risk-less
bond that pays 1 in each state of the world: b = (1, 1)′ .

c2

c1

′
Figure 1.3: Two states (S = 2) and one risk-free security b = (1, 1) .

Under this security structure, all riskless securities on the 45 degree line are replicable but none
of the securities outside of it are replicable (e.g. the red security). Adding a risky security, say
c = (2, 1)′ , allows us to span the whole R2 set. For instance, suppose we wish to replicate the
security d = (1, 2)′ . We can buy 3 securities b (in blue below) and sell short one unit of security c
(in orange), thus obtaining security d (in green):

c2

c1

Figure 1.4: Replicating d = (1, 2)′ with 3 units of bond and −1 (short) unit of the risky asset.

Finally, notice that in the market structure considered here

" #
h i 1 2
X= b c =
1 1

b and c are linearly independent vectors in R2 , and therefore the payoffs that can be obtained with
their linear combinations coincide with those that can be obtained with linear combinations of the
 10 Chapter 1. One Period Model

Arrow-Debreu securities e1 and e2 considered before: therefore, markets are complete.

Before going further, it is worthwhile to mention three important remarks.

First, the state space itself is an important modeling choice. The relevance of this can hardly
be overstated, because market completeness depends directly on how we specify the state space.
Suppose for example that the security structure looks like:
 
1 0 9 4 2
2 1 5 2 2
 
 
X =
3 1 8 4 9

4 2 1 2 2
 

4 2 1 2 2

Clearly markets are incomplete, but if the state s = 5 is a state of the world equivalent to s = 4 in all
aspects with the only exception that “at noon my cat is on the third floor”, then state s = 5 probably
isn’t a relevant one and for modeling purposes should be disregarded. And after eliminating state
s = 5 markets are complete (in the relevant states s = 1, 2, 3, 4). Another example where the state
space choice is important is in derivatives pricing, where we assume that the state space corresponds
to the space of possible values of the underlying asset.

Second, although we said we would not deal with frictions until the third part of the course, market
incompleteness is a market friction: an incomplete market is equivalent to a market in which there
are infinite transaction costs to trade assets whose payoffs are linearly independent from the existing
assets, for which instead transaction costs are zero. From this point of view, we are dealing with a
very “black and white” setup: either we have no frictions at all (with market completeness) or an
extreme friction (with market incompleteness). Later in this course we will explore the “grey” area
in between.

Finally, trade limitations are another kind of market friction we will encounter in this part of the
course. Suppose there are only two states of the world and the security structure includes only a
risk-less bond, which we can neither buy in large amounts nor short-sell (assume there is a law that
prohibits these practices):

Therefore, we can only achieve payoffs on the 45 degree line between the two purple dots.

1.2 General Security Structure & Prices

We now have all the tools to formalize our analysis. We can define:

• A Portfolio as a vector h ∈ RJ , whose entries represent a quantity for each asset in the
security structure.
J
• The Portfolio Payoff is given by ∑ h j x j = Xh
j=1
• The Asset Span is ⟨X⟩ = z ∈ RS : ∃h ∈ RJ such that z = Xh

 1.3 LOOP & No Arbitrage 11

c2

size constraint

short-sale constraint
c1

Figure 1.5: Examples of trade limitations.

Note that it is always the case that ⟨X⟩ ⊆ RS , and we have market completeness if and only if
⟨X⟩ = RS , that is, if and only if rank(X) = S. In other words, market completeness refers to a
market structure in which there are at least S linearly independent assets. We say that security j
is redundant if there exists h ∈ RJ such that x j = Xh and h j = 0: when markets are complete and
J > S there are J − S redundant assets. When rank(X) < S markets are incomplete, and if there
are J < S linearly independent assets then S − J linearly independent assets are needed to achieve
market completeness.

Let p ∈ RJ be the vector of prices for each asset. Then the cost of portfolio h is given by

J
p·h ≡ ∑ p jh j
j=1

xj
and if p j ̸= 0, the (gross) return vector for asset j is given by R j = pj .

1.3 LOOP & No Arbitrage

Let us begin with some notation: for x, y ∈ Rn we write

• y ≥ x if for each i = 1, ... , n yi ≥ xi

• y > x if y ≥ x and y ̸= x
• y ≫ x if for each i = 1, ... , n yi > xi

A fundamental concept in Asset Pricing is that of no arbitrage.

Definition 1.3.1 — No Arbitrage. In our setup it has three forms: given any two portfolios
h, k ∈ RJ and a security structure X ∈ RS×J ,

1. Law of One Price: if Xh = Xk then p · h = p · k

2. No Strong Arbitrage: if Xh ≥ 0 then p · h ≥ 0
3. No Arbitrage: There exists no strong arbitrage, plus, if Xh > 0 then p · h > 0
 12 Chapter 1. One Period Model

These definitions are related through the following lemmas:

Lemma 1.3.1 Law of One Price (LOOP) implies that every portfolio with zero payoff has zero
price.

Proof. Grouping terms we can equivalently write X(h − k) = 0 ⇒ p · (h − k) = 0, and therefore

portfolio w ≡ h − k, which has zero payoff by construction, also has zero price. ■

Lemma 1.3.2 No Arbitrage (NA) implies No Strong Arbitrage (NSA).

Proof. Trivial. ■

Lemma 1.3.3 NSA implies LOOP.

Proof. We prove the contrapositive “if LOOP does not hold, then NSA does not hold either”.
If the LOOP does not hold, then Xh = Xk and p · h ̸= p · k. Assume p · h < p · k, then we have
X(h − k) = 0 ≥ 0 and p · (h − k) < 0, which is a violation of NSA for the portfolio w ≡ h − k.
Similarly, if p · h > p · k then X(k − h) = 0 ≥ 0 and p · (k − h) < 0, again a violation of NSA for
the portfolio q ≡ k − h. ■

1.4 Exercises

1. Determine whether the following statements are true or false. Provide a proof or a counter-
example.
a) Law of one price and complete markets imply no strong arbitrage.
b) Law of one price and complete markets imply no arbitrage.
c) No strong arbitrage and complete markets imply no arbitrage.

2. Suppose there exist three states of the world s = 1, 2, 3 and two assets x1 , x2 .
a) Suppose x1 = (2, 1, 0) and x2 = (0, 1, 0). Describe the asset span. Are markets com-
plete?
b) Suppose p1 = 4 and p2 = 3. What type of no-arbitrage requirements does this market
satisfy?
c) What are the restrictions on p1 and p2 such that this market satisfies LOOP, NSA and
NA? Write each restriction separately.
d) Repeat a), b) and c) for x1 = (1, 1, 0), x2 = (0, 2, 0) and x3 = (0, 1, 1). Hint: it is a
little bit confusing as the price for the third asset is not given. So here is what you are
expected to do for d): 1) Given the payoff vectors for three assets, derive the conditions
for LOOP, NA, NSA to hold respectively. The conditions might be in terms of the three
prices (p1 , p2 , p3 ). 2) After that, plug in (p1 = 4, p2 = 3) in b) to derive the condition
for p3 .
 2. Forwards and Options

2.1 Derivatives
While securities represent property rights (hence contracts), derivatives derive their value from
an underlying security. The most popular derivatives are Swaps, Futures and Options. A natural
question to ask is: since a derivative’s value is a function of the underlying security, are derivatives
always redundant assets? We will see that this is not the case in general, and the answer depends on
the fact that functional dependency does not imply linear dependency.

2.1.1 Forward Contracts

Forward contracts are binding agreements to buy or sell a given security at a specified price, quantity,
time and delivery logistics. Futures contracts are the same from a payoff point of view, but differ
from forwards because they are traded on exchanges that require collateral posting and hence are
more liquid.

600
Long forward
Short forward
400

200
Payoff

−200

−400

−600
2,600 2,800 3,000 3,200 3,400
Level of S&P 500

Figure 2.1: S&P 500 6-months forward contract, agreed price $3060.

The strike level is normally set so that the value of the contract at initiation is zero. Compared to
an outright long position in the underlying asset, it requires no cash outlay at initiation and gives
the same synthetic exposure to the underlying (that is, if the underlying ends up higher by 10 at
expiration, the forward will pay 10 as well). Assuming that a bond pays $1 at maturity, the relation
between a unit of the underlying and the corresponding forward contract is given by

Forward Contract Value = Value of Underlying − Strike × Bond

To see this, suppose the forward contract on a stock S is agreed at time 0 to buy an asset at a future
date T at a specified price K, so that the payoff is (ST − K). Now we ask: what is the fair price of
 14 Chapter 2. Forwards and Options

this contract today? If we can replicate this payoff with a combination of stocks and bonds whose
price is known today, we know that the price of the forward is going to be equal to the price of that
combination of stocks and bonds.

Suppose today we buy the stock at price S0 , and sell K bonds maturing at T at a price of e−rT each.
What do we get at time T ? The stock will be worth ST and each bond will be worth 1, so K bonds
will be worth K. Since we sold them, we will have to pay K. In total, we have ST − K. Therefore we
replicated the payoff of the forward contract, and hence the price of the forward contract today will
be equal to the value of the portfolio of long 1 stock and short K bonds today, that is, S0 − Ke−rT ,
that is, the value of the underlying asset minus the strike price times the bond price.

The forward contract can settle in one of two ways: 1) in cash, i.e. parties exchange the difference
between the underlying at maturity and the strike price in dollar value, which is less costly and
more practical than 2) by physical delivery, i.e. parties exchange the underlying at the agreed price.
Credit risk can be an issue for over-the-counter forwards, for which credit checks and bank letters
of credit may be required other than collateral postings, while for exchange-based transaction
counter-party risk is reduced since the clearing house guarantees the transactions.

2.1.2 Options
A call option is a contract that gives the right (not the obligation) to buy an asset at a specified price
on a future date. From the point of view of the buyer, it preserves the upside potential from owning
an asset without the downside risk. The seller has an obligation to sell if the buyer chooses to buy.
The strike price is the price at which the parties agree to exchange the underlying. When the buyer
chooses to buy the asset, we say he exercises the option. The expiration date is the date by which
the buyer has to decide whether to exercise his right. Classified by exercise style, there are three
main classes of options: 1) European options can only be exercised at expiration date 2) American
options can be exercised anytime between inception and expiry and 3) Bermudan options can be
exercised during some specified dates before or on expiration date.

For European options, because the buyer only exercises when the spot price at maturity is higher
than the strike price (otherwise he would rationally buy the security in the market for less). Therefore
the payoff at expiration is given by

max {S − K, 0} ≡ (S − K)+

and the profit is given by the payoff minus the future value of the option price, called option
premium.

■ Example 2.1 Suppose you buy a 6-month call option on S&P 500, with strike price $3,000 and
premium $93.81. Assuming a 2% 6-month risk-free rate, if the index value in 6 month is $3,100
then the profit is given by $3, 100 − $3, 000 − $93.81 × (1 + 2%) = $4.32, but if it turns out to be
$2,900 then the loss is just $93.81 × (1 + 2%) = $95.68. ■
2.1 Derivatives 15

200 Purchased call

100

$Payoff
0

−100

−200
2,800 2,900 3,000 3,100 3,200
Level of S&P 500

Figure 2.2: Payoff function of a call option.

Purchased call
100 Long forward

0
$Profit

−100

−200

−300
2,800 2,900 3,000 3,100 3,200
Level of S&P 500

Figure 2.3: Profit function of a call option.

Similarly, a put option gives the buyer the right but not the obligation to sell an asset at a determined
price on a future date. The seller of a put is obligated to buy if called upon to do so by the buyer.
Similarly to a call option, for a buyer the payoff is max (K − S, 0) ≡ (K − S)+ and the profit is
given by the payoff minus the future value of the option price. It is worthwhile to remark that while
a call option increases in value when the underlying rises in value, the opposite is true for a put
option.

We say that an option is in-the-money (out-of-the-money) if it would have a positive (negative)

payoff if exercised immediately. If the spot equals the strike we say the option is at-the-money.
There are two interesting examples of derivatives different from options which turn out to be valued
just like options:

1. Homeowner insurance: because insurance pays only in the case of a damage to the house,
it can be thought of as a put option on the value of the house.

2. Equity-Linked Certificates of Deposit: this contract pays the invested amount plus 70% of
the gain in the S&P 500 index. For instance, suppose we invest $10,000 when the S&P 500
 16 Chapter 2. Forwards and Options
 n o
S&P500Final
is at 1700, then the payoff would be 10, 000 × 1 + 0.7 max 1700 − 1, 0 .

Below is a short recap of the profit functions of the derivatives discussed so far:

Figure 2.4: Profit functions of forward and options.

2.2 Back to Security Structures

We may now address the question: are derivatives useful to make markets complete? To make
things concrete, let us consider a specific example in which the possible stock value at time t = 1
are equal to the index of the state of the world: s = (1, 2, . . . , S). We can introduce S − 1 call options
with payoff (s − k)+ for k = 1, . . . , S − 1: we obtain the securities

c1 = (0, 1, 2, . . . , S − 2, S − 1)′

c2 = (0, 0, 1, . . . , S − 3, S − 2)′
..
.

cS−1 = (0, 0, 0, . . . , 0, 1)′

Which together with the stock give rise to the security structure
 
1 ··· 0
0
2 1 · · · 0
 
X = .
 .. . . .. 

 .. . . .
S S −1 ··· 1
 2.3 Forwards Revisited 17

This is an upper triangular S × S matrix whose determinant (the product of diagonal entries) is one.
Therefore X is full rank and markets are complete.

2.3 Forwards Revisited

Consider the following payment and payoff timing combinations:

1. Outright purchase: ordinary transaction

2. Fully leveraged purchase: you borrow the money to execute the purchase
3. Prepaid forward: you pay today to receive shares in the future
4. Forward contract: you agree to a price now, which you pay when you receive the shares in
the future

Description Pay at time Receive security at time Payment

Outright purchase 0 0 S0 at 0
Fully leveraged purchase T 0 S0 erT at T
Prepaid forward contract 0 T ?
Forward contract T T ?

2.3.1 Prepaid Forwards

Suppose we wish to price a prepaid forward for a stock with no dividends. Clearly the timing of
p
delivery is irrelevant, and the price of the prepaid forward F0,T for a stock delivered at t = T is just
equal to the current stock price S0 at t = 0. This reasoning is called pricing by analogy.

Another way of getting to the same result is pricing by arbitrage. Suppose that at t = 0 we observe
p
that F0,T > S0 , then we could buy the stock today at S0 , sell the forward at F0p and pocket the
p
difference F0,T − S0 > 0. In t = T our stock is worth ST , and we owe ST from the forward contract
we sold in t = 0: as a result, in t = T we receive 0. This would constitute an arbitrage. A similar
p
argument can be used for the case when F0,T < S0 . Therefore, absence of arbitrage requires that
p
F0,T = S0 .

When there are dividends the two pricing arguments do not hold any longer: the holder of the stock,
unlike that of the forward, will now receive dividends in the period [0, T ]. As a consequence, it
p p
must be that F0,T < S0 since F0,T = S0 − PV (Dividend Payments in [0, T ]). In particular:

• With discrete dividends Dti for t1 , ... ,tn ∈ [0, T ] and assuming reinvestment at the risk-free
n n
p
rate, we have F0,T = S0 − ∑ PV0 (Dti ) = S0 − ∑ Dti e−rti
i=1 i=1
p
• With continuous dividends with annualized dividend yield δ , we have F0,T = S0 e−δ T

Note that this only applied to deterministic dividends: when dividends are stochastic the securities
structure turns from a matrix into a cube and we can no longer use our one period model.

2.3.2 Forwards
Obviously the forward price is just the future value of the prepaid forward:
18 Chapter 2. Forwards and Options

• With no dividends, F0,T = S0 erT

n n
• With discrete dividends, F0,T = S0 erT − ∑ FVT (Dti ) = S0 erT − ∑ Dti er(T −ti )
i=1 i=1
• With continuous dividends, F0,T = S0 e(r−δ )T

Indexes are an example of assets with continuous dividends. We call forward premium the quantity
F0,T
S0 , which can be used to infer the current stock price from the forward price. The annualized
1 F0,T
forward premium is π = T ln S0 .

We can also use a no-arbitrage argument to price a forward: assuming continuous dividends with
rate δ , we can buy e−δ T units of the stock worth S0 for total price of S0 e−δ T by borrowing the full
amount. At t = 0 there is no cash outlay; however, at t = T the portfolio is worth ST − S0 e(r−δ )T .
Since the long forward payoff is ST − F0,T , this implies that to exclude arbitrage it must be the case
that F0,T = S0 e(r−δ )T .

Transaction Time 0 Time T (expiration)

Buy e−δ T units of the stock −S0 e−δ T +ST
Borrow S0 e−δ T +S0 e−δ T −S0 e(r−δ )T
Total 0 ST − S0 e(r−δ )T

It follows that

Forward = Stock − Zero-Coupon Bond

Long forward
3000$ bond
3,000
Forward + bond

2,000
Payoff

1,000

2,600 2,800 3,000 3,200 3,400

Level of S&P 500

Figure 2.5: Forward vs. outright purchase.

An interesting application of this fact is the so-called cash and carry arbitrage: a market maker can
make a risk-less profit by (for instance) selling short a forward contract and going long a synthetic
forward: the payoff from this strategy is F0,T − S0 e(r−δ )T .

A natural question at this point is: is the forward price a market prediction of the future price? The
answer is no: the formula F0,T = S0 e(r−δ )T shows clearly that the forward price provides no more
information than r, δ and S0 .
 2.4 Options Revisited 19

2.4 Options Revisited

Consider two European options, one call and one put, with the same strike K and time to expiry T .
The put-call parity relation requires that

C(K, T ) − P(K, T ) = PV0 (Forward Price − Strike) = e−rT (F0,T − K)

Note that if F0,T = K the long call short put portfolio above is equivalent to a synthetic forward,
and in fact will have zero price.1 With a dividend stream {Dti }ni=1 we can rewrite the above relation
as

C(K, T ) − P(K, T ) = S0 − PV0 {Dti }ni=1 − e−rT K

While for an index (with continuous dividends) we have

C(K, T ) − P(K, T ) = S0 e−δ T − e−rT K

2.4.1 Option Price Boundaries

Because an American option can be exercised at any time, while a European option can only be
exercised at maturity, it must be the case that

CA (K, T ) ≥ CE (K, T )

PA (K, T ) ≥ PE (K, T )

In general, the American call option price cannot exceed the stock price (otherwise you would
never buy the option but just buy the stock). Moreover, the European call option cannot be lower
than 1) the price implied by put-call parity by setting to zero the put price, or 2) zero, whichever is
highest. That is,2

S0 > CA (K, T ) ≥ CE (K, T ) > e−rT (F0,T − K)+

Similarly, the American put option price cannot exceed the strike price (otherwise you would never
buy the option but just buy the bond). Moreover, the European put option cannot be lower than 1)
the price implied by put-call parity by setting to zero the call price, or 2) zero, whichever is highest.
That is,

K > PA (K, T ) ≥ PE (K, T ) > e−rT (K − F0,T )+

1An alternative definition of “at-the-money” option is to say that an option is at the money when the forward price
equals the strike price. Under this definition, a “long call short put” portfolio replicates a forward when the two options
are at the money.
2 Since max e−rT F

−rT F

+
0,T − S0 , 0 = e 0,T − S0 and CE (K, T ) > 0.
 20 Chapter 2. Forwards and Options

Rationally, we never exercise an American call option on a stock with no dividends: this is
because3

CA (K, T ) ≥ CE (K, T ) = S0 − Ke−rT + PE (K, T ) =

= S0 − K + K(1 − e−rT ) + PE (K, T ) > S0 − K

| {z }
>0

That is, for a holder of an American option it is always best to sell the option rather than exercise it
early. By LOOP, the price of an American call option on a stock with no dividends is the same as
that of an European option. Note that this is not true for a dividend-paying stock, as well as for an
American put on a non-dividend-paying stock.

2.4.2 Time to Expiration

An American option (both put and call) with more time to expiration is at least as valuable as
an American option with less time to expiration. This is because the longer option can easily be
converted into the shorter option by exercising it early. European call options on dividend-paying
stock and European puts may be less valuable than an otherwise identical option with less time to
expiration. 4 A European call option on a non-dividend paying stock will be more valuable than an
otherwise identical option with less time to expiration.

2.4.3 Strike Price

Let K1 < K2 , then we know that C(K1 ) ≥ C(K2 ) and P(K1 ) ≤ P(K2 ) (since their payoff is more
likely to be positive at t = T ). A less obvious fact is that C(K1 ) −C(K2 ) ≤ K2 − K1 , because the
maximum payoff for a collar (long option with low strike, short option with high strike) with strikes
K1 < K2 is K2 − K1 , and the price of the collar cannot exceed its maximum payoff. In the same way,
for put options we have P(K2 ) − P(K1 ) ≤ K2 − K1 . Finally, the option price is convex with respect
to its strike: for K1 < K2 < K3 ,

C(K2 ) −C(K1 ) C(K3 ) −C(K2 )

≤
K2 − K1 K3 − K2

To conclude, it is worthwhile to note that many of these results on option price bounds can be
derived within our (very simple) one period model: they are incredibly robust. Once we adopt a
more specific setting (like the famous Black-Scholes model) we can use more sophisticated tools
such as needs dynamic replication, and the results become deeper but at the same time more hinging
on the specific model used.

3 Remember that with no dividends F0,T = S0 erT .

4Why? This is a popular quant interview question.
 2.5 Exercises 21

2.5 Exercises

1. Similar to slides 12-13 in Lecture slides Part 2 Forwards and Options, please use put options
instead of call options to complete the market given the stock’s payoff as in slide 13. Please
specify the number of put options used and the strike prices. Please write down the payoff
matrix for the assets (with rows as states and columns as assets) and explain why this payoff
matrix indicates a complete market.

2. a) Suppose there are 3 call options traded on a stock with strike prices equal to 40, 50, and
60 and with prices C(40) = 8, C(50) = 6, and C(60) = 2. Show that prices allow for
arbitrage and provide an arbitrage portfolio with initial price equal to zero. What kind
of strategy is this?
b) We have 3 call options on a single stock with strike prices K1 = 100, K2 = 110, and
K3 = 120. We know that C(K = 100) = 10 and C(K = 110) = 8.

i. Can you infer the price of the call with strike 120? Explain your answer.
ii. If C(K = 120) = 6, provide a portfolio strategy with total price <= 0 and positive
payoff at maturity.
iii. For standardization, assume we could ONLY long/short EXACTLY ONE call
with strike K = 120. Please specify the position, amount and strike price for all
instruments used in your strategy.

3. a) For European options with the same strike price K and time to expiration T , let C(K, T )
be the price for such call and P(K, T ) the price for such put at time 0. Construct an
arbitrage strategy when we have C(K, T ) > P(K, T ) + e−rT (F0,T − K). Here F0,T is
the forward contract price on the same underlying stock that expires at T , and r is the
risk-free interest rate if we borrow/lend money. For simplicity and standardization,
assume that:
• For the European call, we can ONLY long/short ONE unit.
• Let the arbitrage strategy’s payoff at time T be ZERO under all circumstances.
Please specify the amount and position for all instruments (European call, etc.) in your
strategy.
b) Suppose a stock index is currently trading at $300, the dividend yield on the index is 3%
per annum, and the risk-free interest rate is 8% per annum. What is the lower bound for
the price of a 6-month European call option on the index when the strike price is $290?
c) Now assume a stock currently sells for $32. A 6-month call option with a strike of
$30 has a premium of $4.29, and a 6 month put with the same strike has a premium
of $2.64. Assume a 4% continuously compounded risk-free rate. What is the present
value of dividends payable over the next 6 months?
22 Chapter 2. Forwards and Options

† 4. Using the Laplace expansion of the determinant of an S × S matrix, prove that

 
1 0 ··· 0
2 1· · · 0
 
X =
 .. ..
. . .. 

. . . .
S S −1 ··· 1

has determinant equal to 1.

† 5. Suppose there exist only a risk-free asset x1 = (1, 1, . . . , 1)′ and a risky asset x2 ̸= x1 and S
states of the world. Let p1 and p2 be the prices of these two assets. A forward contract on
the stock is an agreement to pay an amount F at a future date t = T in exchange for the
payment xsj when the state s ∈ {1, 2, . . . , S} realizes, with no cash flow exchange at time t = 0.
Assuming arbitrage opportunities are ruled out, find the fair value of F.
 3. State Prices

3.1 Pricing in One Period Model

So far we used prices of existing assets to directly derive the price (or price bounds) on other assets.
Now we will go along an indirect route: first we will derive the price of each individual state - called
state price - and then we will use this theoretical tool to derive the price of other assets.

For a given price vector p ∈ RJ and z ∈ ⟨X⟩ define the set v as

v(z) ≡ {p · h : z = Xh}

If LOOP holds then v is a linear functional, that is, a function mapping ⟨X⟩ onto R such that:

1. v is single-valued, because for any z if there exists h ̸= k : z = Xh = Xk then by LOOP

p · h = p · k, and therefore v is a singleton. This means that for any h ∈ RJ such that Xh ∈ ⟨X⟩
we can write we can write v (Xh) = p · h.
2. v is linear on ⟨X⟩,1 since for any α, β ∈ R, z1 = Xh ∈ ⟨X⟩ and z2 = Xk ∈ ⟨X⟩ we have
αz1 + β z2 ∈ ⟨X⟩ and

αv(z1 ) + β v(z2 ) = αv(Xh) + β v(Xk) = α p · h + β p · k = p · (αh + β k) =

= v (X(αh + β k)) = v (αXh + β Xk) = v (αz1 + β z2 )

3. v(0) = 0, since Xh = 0 always has the solution h = 0 and because v is single-valued (by
LOOP) it must be that p · h = p · 0 = 0 is the only price of the payoff z = 0.

The converse is also true: if there exists a linear functional v defined in ⟨X⟩, then LOOP holds.

3.1.1 State Prices

We define state price and linear functional here.
Definition 3.1.1 — State Price. A vector of state prices is a vector q ∈ RS such that p = X ′ q.
That is for j = 1, ... J we have p j = x j · q.

Definition 3.1.2 — Linear Functional. A linear functional V : RS → R is a valuation function if

1. V (z) = v(z) for every z ∈ ⟨X⟩

/ ⟨X⟩, V (z) = q · z for q ∈ RS with qs = V (es ): V extends v from ⟨X⟩ to RS .
2. For every z ∈

Recall that es ∈ RS is the standard basis: a vector with the sth entry equal to 1 and all other entries
1 That is, for all z1 , z2 ∈ ⟨X⟩ and α, β ∈ R such that αz1 + β z2 ∈ ⟨X⟩ it holds that v(αz1 + β z2 ) = αv(z1 ) + β v(z2 ).
24 Chapter 3. State Prices

equal to zero. The next proposition addresses the relationship between v, V and q.

Proposition 3.1.1 if LOOP holds and q is a vector of state prices, then V (z) = q · z for all z ∈ ⟨X⟩.

To see this we only need to show that also for z ∈ ⟨X⟩ we have V (z) = q · z (= v(z)). Suppose that q
is a vector of state prices and LOOP holds, then for z ∈ ⟨X⟩ we have
!
J J J S
xj ·q hj = ∑ ∑ xsj · qs


v (z) = p · h = ∑ p jh j = ∑ hj =
j=1 j=1 j=1 s=1

!
S J S
= ∑ ∑ xsj · h j qs = ∑ zs qs = q · z
s=1 j=1 s=1

Moreover the converse is also true, and therefore the valuation function V (z) = q · z is a linear
functional for all z ∈ RS if and only if q is a vector of state prices and LOOP holds.

■ Example 3.1 Below is a graphical example of state prices. Given the securities structure
!
1 2
X=
1 1

we know that p1 = q1 + q2 and p2 = 2q1 + q2 . As a consequence, we can value the security

x3 = (1, 2)′ = 3x1 − x2 simply as

3p1 − p2 = q1 + 2q2

by LOOP. ■

c2

c1

Figure 3.1: Replicating x3 with 3 units of bond and −1 (short) unit of the risky asset.
 3.1 Pricing in One Period Model 25

3.1.2 The Fundamental Theorem of Finance

Proposition 3.1.2 Security prices exclude arbitrage if and only if there exists a valuation functional
with q ≫ 0.

Proposition 3.1.3 Let X be a S×J matrix, and p ∈ RJ . There is no h ∈ RJ satisfying h · p ≤ 0,

Xh ≥ 0 and at least one strict inequality if and only if there exists a vector q ∈ RS with q ≫ 0 and
p = X ′ q.

It is hard to overstate the importance of this theorem: the absence of arbitrage is equivalent to the
existence of a vector of positive state prices.

3.1.3 State Prices and Incomplete Markets

We have established an astonishing equivalence between arbitrage (the absence of arbitrage) and
state prices (the existence of positive state prices). A natural follow-up question is whether there
is a similar equivalence between state prices and market completeness. Suppose for instance that
there are two states of the world and only one bond x1 = (1, 1)′ with price p1 . What are the state
prices consistent with this incomplete market structure? We know that any p1 = q1 + q2 would
work, hence the state prices consistent with no arbitrage are all q ∈ R2 such that q1 ∈ (0, p1 ) and
q2 = p1 − q1 .

z2
⟨X⟩
(1, 1)
p(1, 1)

q̂ q∗

q z1
p(1, 1)

Figure 3.2: Pricing kernel is the unique projection of any state price vector q on ⟨X⟩.

In the picture above, the orange line is the set of q’s consistent with prices and the security structure.
Of all q ∈ R2 consistent with no arbitrage however, there is a very special one that also belongs to
⟨X⟩: it is the unique projection of any state price vector q on ⟨X⟩. In our example, this would be

′
achieved for q1 = p21 since then q = p21 , p21 = p21 × (1, 1)′ ∈ ⟨X⟩. We call this state price pricing
kernel and denote it as q∗ .

Now we are ready to state the important relation between state prices and market complete-
ness:

Proposition 3.1.4 Under no arbitrage, markets are complete if and only if there exists a unique
valuation functional.
 26 Chapter 3. State Prices

Proof. If markets are complete and there is no arbitrage, then for a given Xand p the system
X ′ q = p has a unique solution q ≫ 0 (positivity follows from the assumption of no arbitrage).

We prove the if direction by contrapositive statement. If markets are not complete, then there exists
a vector v ∈ RS such that v ̸= 0 and X ′ v = 0. Under no arbitrage, there must exist some positive
q ≫ 0 and some α ∈ R such that q + αv ≫ 0 and X ′ (q + αv) = p (provided α is small enough).
Since X ′ (q + αv) = p this is also a valid state price vector. Hence, there are an infinite number of
strictly positive state prices. ■

3.2 Asset Pricing Formulas

We now present four asset pricing formulas. They are effectively equivalent, but each one can be
interpreted in a particular way and derived from the one period model.

3.2.1 State Price Model

S
This is just the pricing formula seen above: p j = ∑ qs xsj
s=1

3.2.2 Stochastic Discount Factor

S
Analogously to the state price model, we can write p j = ∑ πs πqss xsj where π is the physical proba-
s=1
qs
bility distribution of states. Defining the random variable Stochastic Discount Factor m as ms ≡ πs ,
S
we get p j = ∑ πs ms xsj = E m · x j (probability weighted inner product between m and x j ).
 
s=1

Note that p j = E m · x j = E [m] E x j + Cov m, x j , and since for a risk-free bond xsb = 1 for
     

all s, we have pb = E [m] = R1f where R f is the gross risk-free return. Therefore, for any asset j,
E[x j ]
p j = R f +Cov m, x j . Typically, Cov m, x j < 0.
   

j 1
Defining R j ≡ xp j , we get E m · R j = 1. Since for a risk-free bond R f = E[m]
 
, we can write
 j f


E m· R −R = 0, or

E m · R j − R f = E [m] E R j − R f +Cov m, R j = 0

  

That is,

Cov m, R j


E Rj −Rf = −
 
E [m]

which implies that the excess return for a generic asset j is determined solely by the covariance
with the stochastic discount factor. This also means that an investor is only compensated (with a
higher return) for holding systematic risk, not idiosyncratic risk.
 q∗ 
1

∗
 π.1  ∗
Consider the stochastic discount factor obtained from the pricing kernel m ≡  .
 . . Note that m
q∗S
πS
3.2 Asset Pricing Formulas 27

is the projection of any stochastic discount factor m on ⟨X⟩, that is,

m∗ = proj (m | ⟨X⟩) ∈ ⟨X⟩

which means that there exists a vector h∗ ∈ RJ such that m∗ = Xh∗ . Therefore for any asset j we
can write

p j = E m∗ · x j
 

So for all assets we have

p = E X ′ m∗ = E X ′ Xh∗ = E X ′ X h∗
     

E [X ′ X] is a second order moment: assuming it is invertible we can write


−1
h∗ = E X ′ X

p

And therefore plugging this in m∗ = Xh∗ we obtain


−1
m∗ = X E X ′ X

p

This is similar to the language of linear regressions. When we run a regression of p on

y = Xb + ε

We find the linear combination of X that is “closest” to y by minimizing the variance of the residual
ε. We do this by forcing the residual to be “orthogonal” to X, E [Xε] = 0. The projection of y
onto X is defined as the fitted value Xb = X (E [X ′ X])−1 E [X ′ y]. This idea is often illustrated by a
residual vector ε that is perpendicular to a plane defined by the variable X. Thus, when the inner
product is defined by a second moment, the operation “project y onto X” is a regression.

Finally, note that we can represent our previous discussion for state prices by shrinking the axes by
a factor π:
z2
⟨X⟩
p(1, 1)π2

(1, 1)

m∗
m
z1
p(1, 1)π1

Figure 3.3: m∗ is the projection of any stochastic discount factor m on ⟨X⟩.

 28 Chapter 3. State Prices

3.2.3 Equivalent Martingale Measure

S S
Starting again from p j = ∑ qs xsj , for a riskless bond we have pb = ∑ qs = 1
1+r f
, where r f is the
s=1 s=1
S S
1 qs
xsj = 1 j 1
EQ x j , where
 
risk-free net return. Thus we can write p j = 1+r f ∑ S 1+r f ∑ π̂s xs = 1+r f
s=1 ∑ qs s=1
s=1
qs
π̂s ≡ S .2 Compared to the stochastic discount factor approach, we simply used a different
∑ qs
s=1
probability measure to discount future states. We will see that the significance of this probability
measure is market-determined and its importance is paramount in options pricing theory.

3.2.4 State-Price Beta Model  q∗ 

1

∗
 π.1 
Consider the stochastic discount factor obtained from the pricing kernel m ≡  .
 . , and define its
q∗S
πS
m∗
return as R∗ = pm∗ ≡ αm∗ for α > 0 3 . Then we can write

Cov R∗ , R j


E Rj −Rf = −
 
E [R∗ ]

Cov(R∗ ,R j )
Defining β j ≡ Var(R∗ ) we can write for the asset j:

Var (R∗ )
E R j − R f = −β j
 
E [R∗ ]

While for security x∗

Var (R∗ )
E [R∗ ] − R f = −
E [R∗ ]

Therefore, for security j we have

E R j − R f = β j E [R∗ ] − R f
 

which, if we assume a linear model for R j and R∗ can be specified and tested empirically as

Rkj − R f = β j R∗k − R f + εk

with Cov (R∗ , ε) = E [ε] = 0.

3.2.5 Four Asset Pricing Formulas

In summary, the four equivalent pricing relations are:
S
1. State Price Model: p j = ∑ qs xsj
s=1
2 The Q notation comes from the literature about risk-neutral valuation.
3 Since m∗ ∈ ⟨X⟩, ∃h∗ : m∗ = Xh∗ . Furthermore ∃α : p(αh∗ ) = 1.
 3.3 Recovering State Prices from Option Prices 29

2. Stochastic Discount Factor: p j = E mx j

 

1
EQ x j
 
3. Equivalent Martingale Measure: p j = 1+r f

4. State-Price Beta Model: E R j − R f = β j E [R∗ ] − R f

 

As a last remark, note that whenever markets are incomplete, the multiplicity of state price vectors q
translates directly into the multiplicity of stochastic discount factors m and of equivalent martingale
measures π̂.

3.3 Recovering State Prices from Option Prices

Let’s assume for a moment that ST , the value of the asset at expiration, can take on a continuum of
values: at time t = T the stock price can have any value ST ∈ R+ . In this section we will use the
law of one price to derive the price of an Arrow-Debreu security for a continuum of states, which is
a function q : R+ → R called state price density. In this context, a state price density is the price of
an asset that pays one dollar in a particular state x ∈ R+ and zero in all others (here each value of
the price for the underlying asset at time t = T corresponds to a different state). Assuming further
that there exist call options offered in the market that cover a continuum of strike prices (one for
each possible outcome at expiration), the result of section 3.2 extends to this case: markets are
complete.

Fix a strike K > 0, and a (small) number ε > 0. Consider the following portfolio:

• Buy x call options with strike K − ε

• Sell 2x call options with strike K
• Buy x call options with K + ε

State Payoff
ST ∈ [0, K − ε) 0
ST ∈ [K − ε, K) x (ST − (K − ε)) = x (ST − K) + xε
ST ∈ [K, K + ε) x (ST − (K − ε)) − 2x (ST − K) = −x (ST − K) + xε
ST ∈ [K + ε, +∞) x (ST − (K − ε)) − 2x (ST − K) + x (ST − (K + ε)) = 0

The payoff of this portfolio (symmetric butterfly) looks like the following:

Portfolio Payoff

xε

K −ε K K +ε ST

Figure 3.4

This portfolio corresponds the payoff of an Arrow-Debreu security for state ST = K if two conditions
are satisfied:
30 Chapter 3. State Prices

1. The area under the payoff diagram (the sum of all possible payoffs) must be equal to one.
2. The option portfolio must yield a payoff of zero for all values of the underlying asset different
from K.
(xε)×(2ε) 1
The area under the payoff diagram is equal to 2 = xε 2 : to achieve (1), we set x = ε2
; to
achieve (2), we take the limit for ε → 0. Now that we have replicated the payoff of an Arrow-Debreu
security, we can use the law of one price to determine the state prices: for a fixed ε > 0 the price of
the portfolio is

C (K + ε, T ) − 2C (K, T ) +C (K − ε, T )
V (K, ε) =
ε2

and therefore

C (K + ε, T ) − 2C (K, T ) +C (K − ε, T ) ∂ 2C (K, T )
q (K) = limV (K, ε) = lim =
ε→0 ε→0 ε2 ∂ K2

∂ 2C(K,T )
We established an important fact: because q(K) > 0 if and only if ∂ K2
> 0, by the fundamental
∂ 2C(K,T )
theorem of finance there is no arbitrage if and only if for all K > 0 we have ∂ K2
> 0, that is, if
option prices are convex.
∂ 2C(K,T )
There is another important relationship related to state prices and ∂ K2
. We can write the option
price using the equivalent martingale measure representation as follows:

C (K, T ) = e−rT EQ (ST − K)+

 

Taking the first derivative with respect to K we get

∂C (K, T )  
= −e−rT EQ [I {ST − K ≥ 0}] = −e−rT PQ {ST ≥ K} = −e−rT 1 − FSQT (K)
∂K

where FSQT is the cumulative distribution function for the random variable ST under the equivalent
martingale measure Q.4 Differentiating again with respect to K we get

∂ 2C (K, T )
= e−rT fSQT (K)
∂ K2

Together with our previous result,

∂ 2C (K, T )
2
= q(K) = e−rT fSQT (K)
∂K

From the findamental theorem of finance, it follows that there exists a (positive) equivalent martin-
gale measure density function fSQT if and only if no arbitrage holds.
4We define I {ST − K ≥ 0} as a function that is equal to one if ST − K ≥ 0 and zero otherwise. Note that we can take
the partial derivative of C(K, T ) with respect to K since EQ (ST − K)+ = R fSQT (x) (x − K)+ dx = K∞ fSQT (x) (x − K) dx
  R R

is smooth in K.
3.3 Recovering State Prices from Option Prices 31

Note that this implies that we can evaluate any future payoff h(ST ) in 3 ways:

∂ 2C (K, T )
Z Z Z
−rT −rT
p=e Q
E [h (ST )] = e fSQT (x) h (x) dx = (x) h (x) dx = q(x)h (x) dx
R R ∂ K2 R

where the last term is the continuous-states equivalent of the formula p = X ′ q.

This means the following -apparently unrelated- conditions are equivalent:

• Absence of arbitrage
• Existence of a positive equivalent martingale measure density function
• Convexity of options

Note that in the market we only directly observe the option prices for some discrete and finite set
of strikes: but if all the option prices C(K, T ) for the strikes K ∈ {0, ∆, 2∆, . . . , N∆} are observable
∂ 2C(K,T )
(for N large and ∆ > 0 small enough) we can approximate ∂ K2
in the following way

∂ 2C (K, T ) C (K + ∆, T ) − 2C (K, T ) +C (K − ∆, T )
≈
∂ K2 ∆2

and therefore we can also calculate the empirical market-implied probability distribution of ST ,
fSQT , and the empirical state price density q.

Going back to our finite-state one-period model, if all the option prices C(s, T ) for the strikes
s ∈ {0, 1, 2, . . . , S} are observable, we have

π̂s
△2C (s, T ) ≡ C (s − 1, T ) − 2C (s, T ) +C (s + 1, T ) = qs =
(1 + r f )
 32 Chapter 3. State Prices

3.4 Appendix
The following is an alternative derivation from option prices of the state price density function. We
can construct the following portfolio: for some ε > 0 and δ > 0 and a fixed ŜT

• Buy one call with strike ŜT − δ2 − ε

• Sell one call with ŜT − δ2
• Sell one call with ŜT + δ2
• Buy one call with ŜT + δ2 + ε

Payoff

ε
ŜT − δ2 ŜT + δ2

ŜT − δ2 − ε ŜT + δ2 + ε ST

Figure 3.5
h i
1
If we buy ε units of this portfolio, when ST ∈ ŜT − δ2 , ŜT + δ2 the total payoff is equal to 1. The
total value of this portfolio is
        
1 δ δ δ δ
C K = ŜT − − ε, T −C K = ŜT − , T −C K = ŜT + , T +C K = ŜT + + ε, T
ε 2 2 2 2

Letting ε → 0 this boils down to

   
∂C K = ŜT − δ2 , T ∂C K = ŜT + δ2 , T
− +
∂K ∂K

Finally, dividing by δ and letting δ → 0, we obtain the continuum-states version of a vector of state
∂ 2C(K=ŜT ,T )
prices, the state price density function ∂ K2
.

Suppose we want to evaluate a one-year “wedding cake option” of the type


$1 if ST ∈ [1700, 1750]
Payoff =
$0 otherwise

We now have the technology to price this. Its value will be equal to the integral of the state price
density over the interval [1700, 1750], that is, for T = 1,
Z 1750 2
∂ C(K, T ) ∂C (K = 1750, 1) ∂C (K = 1700, 1)
dK = −
1700 ∂ K2 ∂K ∂K

Finally, note that this is equivalent to a portfolio comprising a long position in a binary 1700 call
and a short position in a 1750 binary call.
 3.5 Exercises 33

3.5 Exercises

1. Prove that if LOOP holds, v(z) is a linear functional.

Remember from lecture slides that we define v(z) for each z ∈ ⟨X⟩ as v(z) := {p · h : z = Xh}
(Hint: To prove that f is a linear functional you need to prove three properties i) f is single
valued, ii) f (0) = 0, iii) f is linear on the subspace ⟨X⟩)

2. Suppose there are S possible states of the world in t = T and each has a (physical) probability
of occurrence ηs > 0 with for ∑Ss=1 ηs = 1. Consider the vector µ ∈ RS with s = 1, ..., S,
µs = ηqss where q is the known state-price vector. Write E(y) = ∑Ss=1 ηs ys for any y ∈ RS .

a) Consider an asset with payoff x = (x1 , ..., xS ). Show that the price of this asset must be
E(z), where zs = µs xs . Interpret this result.
xs
b) Let the rate of return for an asset with price p > 0 in state s be rs = p and let ws = rs µs .
Show that E(w) = 1. Is there some function of the excess return of the asset r − r f such
1
that E[ f (r − r f )] = 0? (Here we define r f such that, rf
= ∑Ss=1 qs ).

3. Consider a market where there are 3 assets with payoff x1 = (1, 0, 1, 2)′ , x2 = (0, 1, 1, 3)′ , x3 =
(0, 1, 4, 0)′ . Assume p1 = 0.8, p2 = 1.1 and p3 = 1.4, find the state price vector q =
(q1 , q2 , q3 , q4 )’.
a) Is q unique? If not, please express the q with free variables starting from q4 . (i.e., if
there is one free variable, express q1 , q2 , q3 with q4 ; if there are two, express q1 , q2 with
q3 , q4 , ...).
b) Now assume q4 = 0.25 in this sub-question only. What is q now? Does LOOP hold? Is
there an arbitrage opportunity? Please explain.
c) Now assume the risk-free interest rate is 25% for this sub-question only. (And the
risk-free bond is available in the market as well.) What is q now? Is there an arbitrage
opportunity? Is the market complete? Please explain. And please use one-period
discrete discounting here (instead of involving continuous exponential terms).
d) What are the ranges of the free variables from your answer in part 1, if there is no
arbitrage? What is the price range for asset X = (1, 0, 0, 1)′ under NA?

4. Suppose there exist 3 states of the world s = 1, 2, 3 and 2 assets x1 = (2, 1, 0) and x2 = (0, 1, 0).
a) Suppose p1 = 1 and p2 = 0.3. What state prices are consistent with these prices?
b) Solve for the unique pricing kernel q∗ .
c) Use the pricing kernel to value a third asset x3 = (0, 1, 1).
d) Now suppose p3 = 0.6. Solve for the state price vector. Does this market permit
arbitrage?
e) Solve for the stochastic discount factor assuming the physical probability is such that
1
π1 = π2 = π3 = 3
f) Solve for the distribution (π̂) under the equivalent martingale measure.
34 Chapter 3. State Prices

5. Suppose there are three assets and three states. The payoff matrix is X = [X1 , X2 , X3 ] ∈ R3×3 ,
where the columns represent assets and rows represent states. Let the assets have linear
independent payoffs. Furthermore, the price vector for the three assets is p ∈ R3 .
a) Please express the state price vector q∗ with X and p, so that it belongs to ⟨X⟩.

Now for the rest of the sub-questions, let X1 = (1, 0, 1)′ , X2 = (0, 1, 1)′ , X3 = (0, 0, 1)′ ,
and p = (2, 1.5, 1)′ .

b) Please calculate the q∗ .

c) Suppose π = (π1 , π2 , π3 )′ = (0.25, 0.25, 0.5) is the physical probability distribution of
states. What is the stochastic discount factor m∗ ?
d) Now please find the R∗ , which is in the span of m∗ (that is, R∗ = αm∗ ). (Hint: p =
E[mX].)

6. Suppose that the function that determines the price of a call option with its strike is of the
form C(K) = aK 2 + bK + c.
a) Without any computation find three constraints on the parameters for no arbitrage.
(Hint: zero/first/second-order properties for the call price.)
b) Please first write down the state price density function, then price the following options
in terms of a, b and c:
i. A wedding cake option, where the option which gives the payoff 1 if the ST is
between 2200 and 2600, and 0 otherwise.
ii. Price the double barrier option that gives 0 if ST is below 2300 or above 2500 and
gives ST2 if ST is between 2300 and 2500.
iii. Price the put option on the S&P500 with strike K = 2300 and maturity 6 months.
Explain why this price could be off given our assumptions.
(Comment: This question is for illustration only so that you could get familiar with
using state price density to price certain instruments. The assumptions for C(K) in this
question are actually not quite appropriate)

† 7. Suppose a stock index is currently trading at $25, and there are 5 possible states of the world
in t = T such that ST ∈ {15, 20, 25, 30, 35}.
a) Given a zero risk-free interest rate, describe a valid equivalent martingale measure.
b) Under this measure, price call options at K = 15, 20, 25, 30, 35.
c) Use this information to recover state prices.

† 8. Show in detail how to retrieve state prices using put options both in a continuous and discrete
states setup.
 4. Empirical Multi-Factor Models

This chapter has not been typed. See the lecture slides for the coverage of this material.

4.1 Exercise
On Canvas you can find two files, 25_Portfolios_5x5.csv and F_F_Research_Data_Factors.csv.
These files are downloaded from Kenneth French’s website that provides a large set data series
useful for empirical, multi-factor asset pricing. As always feel free to collaborate on this question
but try to write your own code.

1. Open the two files in a text editor to read the descriptions and get an idea of the file format.
Have a look at Kenneth French’s website,
[Link]
and locate the two files that we are working with.

2. We will work with the value weighted monthly portfolio returns and the monthly factor series
from July 1926 to July 2022. Use your favorite language/software to read in the data.

3. We start from the hypothesis that mt∗ = a + b(RtMkt − Rtf ). Estimate the regression

Rti − Rtf = α̂i + β̂i (RtMkt − Rtf ) + εti

for each portfolio i. Is mt∗ in the payoff space? What does β̂i measure statistically? Based on
our hypothesis, what value should α̂i take approximately?

4. Estimate the average excess return of each portfolio i, R̄i − R̄ f . Regress those average returns
on the previous regression coefficients β̂i , so run the following regression:

R̄i − R̄ f = αi + λ β̂i + εi .

Is your estimate of λ significantly different from zero? What is the R2 of this regression?
Should we reject our initial hypothesis that mt∗ = a + b(RtMkt − Rtf )?

5. Split the sample into three different episodes, 1926 − 1964, 1965 − 2004, and 1983 − 2022.
Does the model work better in any of these sub-samples?

6. Repeat the two-stage procedure for the three factor model, based on the new hypothesis that

mt∗ = a + b1 (RtMkt − Rtf ) + b2 RSMB + b3 RtHML .

36 Chapter 4. Empirical Multi-Factor Models

Are the three λ coefficients significant? How does the model fare across sub-samples? Do
the λ ’s remain stable across samples?
 5. The Multi-Period Model

In the first four chapters, we assume that there is only one period (between t = 0 and t = 1). A
natural question to ask is then, how can we deal with a multi-period setup in which several outcomes
are possible after each outcome realizes? One way to tackle this problem is to bring down the
problem to a set of many “smaller” one-period model problems:

s2,1

A1,1

s2,2

A0

s2,3

A1,2

s2,4

t =0 t =1 t =2

Figure 5.1: The structure of security payoffs—two periods—two states at each node

With this approach however, we need a more efficient way to define and work with the information
available at each time step. A natural way to approach this is using the concepts of σ -algebra and
filtration.

5.1 Model Setup

We begin by defining the relevant concepts:
Definition 5.1.1 — σ −Algebra. Given a state space (or sample space) Ω, a σ -algebra (or
σ -field) F is a non-empty collection of subsets of Ω iff it has the following properties: 1) 0/ ∈ F ,
2) F is closed under complementation: A ∈ F ⇒ AC ∈ F , 3) F is closed under countable
unions: A1 , A2 , A3 , · · · ∈ F ⇒ A1 ∪ A2 ∪ A3 ∪ · · · ∈ F .

Definition 5.1.2 — Measurability. A random variable Y is measurable with respect to the

σ -algebra F if {ω ∈ Ω | Y (ω) ≤ y} ∈ F for all y ∈ R.
38 Chapter 5. The Multi-Period Model
Definition 5.1.3 — Stochastic Process. A stochastic process is a collection of random variables
T
{Yt }t=0 .

T
Definition 5.1.4 — Filtration. A filtration is a collection of algebras FT = {Fs }s=t such that if
u ≤ v then Fu ⊆ Fv .
T
Definition 5.1.5 — Adapted Process. A stochastic process {Yt }t=0 is adapted to the filtration
FT = {Fs }Ts=t if Ys is measurable with respect to Fs for s = t, . . . , T .

Figure 5.2: Filtration, events, states.

Adaptation is the requirement that people cannot see in the future. Moving from a static to a
dynamic setting, we need to adapt our notation:

• Rather than asset holdings, we will talk about dynamic strategies

• Instead of asset payoff vectors x we will talk about the “next period payoff” xt+1 + pt+1 of a
strategy
• Instead of the asset span, we will talk about the subset of marketed dynamic strategies
• Market completeness can be interpreted in two ways: static completeness (Debreu complete-
ness) and dynamic completeness (Arrow completeness)
• We defined no arbitrage with respect to asset holdings, but now it will be defined on dynamic
strategies
• States s ∈ {s1 , s2 , . . . sS } generalize to events At,i and final states st,i
• State prices qs become event prices qt,i
• The risk-free rate Rtf varies over time
• the discount factor ρt is time-dependent and discounts from t to 0
q
• The risk-neutral probability is π Q (At,i ) = ρt,it
h  i
• The pricing kernel now reads Mt ptj = Et Mt+1 pt+1 j j
+ xt+1 f
, and Mt = Rt+1 Et [Mt+1 ]
5.1 Model Setup 39

We will use trees to illustrate dynamics. In Figure 5.3, the tree has 7 events. Every event has a
positive probability attached to it, and we impose the consistency requirement that the probability
of any event equals the sum of the events that follow from its node: for example, denoting πt,s the
probability of At,s , we have π1,1 = π2,1 + π2,2 . Clearly, for each t we have ∑πt,s = 1.
s

A2,1 = s2,1

A1,1

s2,2

A0

s2,3

A1,2

s2,4

t =0 t =1 t =2

Figure 5.3: The structure of security payoffs—two periods—two states at each node

There are two ways to reduce this seemingly more complicated model to the one-period model.
The first is to boil down every event to a final state, and apply what we know from the one-period
model. Note that if we simplify the tree in this way it follows that to achieve completeness we will
need six independent assets, one for each final state.

s2,1 A1,1

A1,1 A1,2

s2,2 s2,1

A0 A0

s2,3 s2,2

A1,2 s2,3

s2,4 s2,4

t =0 t =1 t =2 t =0 t =2

Figure 5.4

The second way is to reduce each branch of the tree to a one-period model. In this case, assets can
 40 Chapter 5. The Multi-Period Model

be re-traded conditionally on the the occurrence of event A1,1 or A1,2 . Completion can be achieved
with two categories of assets: short-lived assets pay off only in the period following the one in
which they are traded, while long-lived assets pay off over many periods.

s2,1 s2,1

A1,1 A1,1

s2,2 s2,2

A0 A0

s2,3 s2,3

A1,2 A1,2

s2,4 s2,4

t =0 t =1 t =2 t =0 t =1 t =2

Figure 5.5

5.2 Dynamic Trading and Market Completeness

Suppose now that asset can be traded in each period. This allows for dynamic completion, as well
as for the existence of bubbles and Ponzi schemes in an infinite horizon setting. We will study
completeness with both short- and long-lived assets.

A short-lived asset is an asset that pays out only in the period immediately after the asset is issued.
Without uncertainty and with T one-period assets (that is, T assets that can be bought in period
t − 1 at price pt and pays off in period t for t = 1, . . . , T ), completeness requires that we are able
to transfer wealth between any two periods. For example, if zero coupon bonds are traded in
each period and T = 2, at time t = 0 we can buy a quantity p2 of bonds that pay off at time t = 1
at a total cost of p1 p2 , and then in period t = 1 we can invest the proceeds from the bond we
bought to buy one unit of the bond that pays at time t = 2: in this way we replicated a bond that is
bought at t = 0 and pays $1 at time t = 2, and therefore markets are complete. In general, for T
periods the cost of rolling over this kind of short-lived bond strategy is just pt · pt+1 · . . . · pT −1 , and
completeness requires that we are able to transfer wealth between any two periods t and t ′ (not just
consecutive).

With uncertainty the reasoning is very similar. Suppose the event tree is the one in Figure 5.6 and
we want to replicate the payoff of an asset that is bought at t = 0 and pays $1 if event A2,2 = s2,2
realizes. Let q2,2 denote the state price corresponding to A2,2 traded in t = 1. Then, working
backwards, in event A1,1 we can buy a one-period asset that pays $1 in event A2,2 for a price q2,2 ; to
be able to dispose of an amount q2,2 if event A1,1 realizes, in event A0 we need to buy a quantity q2,2
5.2 Dynamic Trading and Market Completeness 41

of the asset that pays $1 if event A1,1 realizes, which costs q1,1 per unit. Thus the price of the payoff
of an asset that is bought at t = 0 and pays $1 if event A2,2 realizes is q1,1 · q2,2 , and we can repeat
the argument for all other events to see that this market achieves dynamic completeness.

s2,1

A1,1

s2,2

A0

s2,3

A1,2

s2,4

t =0 t =1 t =2

Figure 5.6

With long-lived assets, consider a T -period model without uncertainty. There is a single asset
paying $1 in t = T , which is tradable in each period for a price pt : this means that sequentially,
there are T prices for this asset. The payoff can be transferred from period s to period t > s by
purchasing the asset at time s and selling it in period t. For example, suppose there are two assets
that pay off like in Figure 5.7.

Asset 1 Asset 2

s2,1 1 0
p11,1
A1,1
p21,1
s2,2 0 1

A0

s2,3 1 0
p11,2
A1,2
p21,2
s2,4 0 1

t =0 t =1 t =2

Figure 5.7
42 Chapter 5. The Multi-Period Model
j
The assets are long-lived, since they can be re-traded at t = 1 for the price pt,s . Note that we have
two assets, but each assets can be traded in three events; moreover, the price of each asset can be
endogenously determined at time t = 0. Although one may be tempted to think that because we
only have 2 assets markets cannot be complete, in fact dynamic trading provides a way to transfer
wealth in all events:

Asset 1 Asset 2

s2,1 1 0
p11,1
A1,1
p21,1
s2,2 0 1
p10
A0
p20
s2,3 1 0
p11,2
A1,2
p21,2
s2,4 0 1

t =0 t =1 t =2

Figure 5.8

Call trading strategy [ j, At,i ] the cash flow of asset j that is purchased in event At,i and sold one
period later. There are six such trading strategies: [1, A0 ], [1, A1,1 ], [1, A1,2 ], [2, A0 ], [2, A1,1 ], [2, A1,2 ].
For example, trading strategy [1, A1,1 ] costs p11,1 and pays out $1 in the first final state and zero in
all other events; and trading strategy [1, A0 ] costs p10 and pays out p11,1 in event 1, p11,2 in event 2,
and zero in all the final states.

These six trading strategies give rise to a 6 × 6 payoff matrix:

Strategy [1, A0 ] [2, A0 ] [1, A1,1 ] [2, A1,1 ] [1, A1,2 ] [2, A1,2 ]
Event A0 −p10 −p20 0 0 0 0
Event A1,1 p11,1 p21,1 −p11,1 −p21,1 0 0
Event A1,2 p11,2 p21,2 0 0 −p11,2 −p21,2
State s2,1 0 0 1 0 0 0
State s2,2 0 0 0 1 0 0
State s2,3 0 0 0 0 1 0
State s2,4 0 0 0 0 0 1

The payoff matrix is nonsingular (and hence markets are complete) if and only if the subma-
trix
 5.3 The Multi-Period Stochastic Discount Factor 43

p11,1 p21,1
p11,2 p21,2

is nonsingular (that is, it has rank equal to 2). The components of this submatrix are the prices
of the two assets conditional on period-1 events. There are cases in which these are colinear in
equilibrium: for instance when the per capita endowment is the same in event A1,1 and A1,2 , in state
s2,1 and s2,3 , and in state s2,2 and s2,4 , and if the probability of reaching state s1,1 after event A1,1 is
the same as the probability of reaching state s2,3 after event A1,2 then the submatrix is singular (but,
one might argue, then events A1,1 and A1,2 are effectively equivalent and could be collapsed into a
single event). Moreover, a random square matrix is regular: this means that generically, the market
is dynamically complete (that is, it is not complete only for information trees and asset structures
which have zero probability of showing up in any application).

We call branching number the maximum number of branches that fan out from any event in the
uncertainty tree: it turns out that this is also the number of assets necessary to achieve dynamic
completeness. This generalizes to the continuum of events and continuous time case, in which a
small number of assets is sufficient to achieve completeness because the need for many assets to
achieve completeness that arises from a large number of possible events is offset by the ability to
trade continuously, thus generating a large number of trading strategies. The classic example is the
Black-Scholes formula: Cox, Ross and Rubinstein showed that is is possible to build a binomial
tree model of Black-Scholes, with constant interest rates and in which the stock can only go up or
down, in which the market is dynamically complete with just two assets: the stock and the risk-free
bond. Thus an option on the stock can be replicated with dynamic delta-hedging.

5.3 The Multi-Period Stochastic Discount Factor

The idea of no arbitrage extends naturally to the multi-period setting: an arbitrage is a strategy
that has either no cost today but some positive payoff along the tree or a negative cost today and
no negative payoff along the tree. Moreover, it is equivalent to having no static arbitrage in each
branch of the tree.

Recall that no arbitrage requires that there exists a positive stochastic discount factor mt+1 for each
subperiod t such that

pt = Et [mt+1 (pt+1 + xt+1 )]

Define the multi-period stochastic discount factor as1

Mt+1 = m1 · m2 · . . . · mt+1

Multiplying each side by m1 · m2 · . . . · mt (which is measurable until the time t filtration) we

1We assume m0 = 1 from now on.

44 Chapter 5. The Multi-Period Model

get

Mt pt = Et [Mt+1 (pt+1 + xt+1 )]

∞
Assuming for this asset there is a stream of cash flows {xt }t=1 where each xt is a random payoff,
we have for time t = 0

p0 = E0 [M1 (p1 + x1 )]

similarly

M1 p1 = E1 [M2 (p2 + x2 )]

plugging in the equation for p0 we have

p0 = E0 [E1 [M2 (p2 + x2 )] + M1 x1 ] = E0 [E1 [M2 p2 + M2 x2 + M1 x1 ]]

The Law of Iterated Expectations (LIE) states that, because of the property of filtrations that if u ≤ v
then Fu ⊆ Fv , for any random variable X the time-u expectation of the random variable “time-v
expectation of X” is equal to the time-u expectation of X. The intuition is as follows: consider your
current guess of what the weather will be on Sunday, and compare it to the guess that you will have
on Saturday about the weather on Sunday. LIE states that your current guess on your guess for the
Sunday weather is the same as your current guess for the Sunday weather. Therefore we get

p0 = E0 [M2 p2 + M2 x2 + M1 x1 ]

Iterating k steps forward we have

k
p0 = E0 [Mk pk ] + ∑ E0 [Mt xt ]
t=1

Taking the limit for k → ∞ (and assuming lim E0 [Mk pk ] = 0) we have

k→∞

∞
p0 = ∑ E0 [Mt xt ]
t=1

∗ = proj (m ∗
In terms of projections, recall that mt+1 t+1 | ⟨Xt+1 ⟩), that is, there exists some ht such
∗ =X
that mt+1 ∗
t+1 ht and

h ′ i h ′ i
∗
pt = Et Xt+1 mt+1 = Et Xt+1 Xt+1 ht∗

So that
 h ′ i−1
ht∗ = Et Xt+1 Xt+1 pt
 5.4 Martingales 45

And therefore
 h ′ i−1
∗
mt+1 = Xt+1 Et Xt+1 Xt+1 pt

Clearly we also have Mt∗ ≡ m∗1 · m∗2 · . . . · mt∗ ∈ ⟨Xt ⟩. Further, we can express mt+1
∗ in terms of the
covariance of returns Σt :

∗
 ∗    ∗  ′
mt+1 = Et mt+1 + pt − Et mt+1 Et [Xt+1 ] Σt−1 [Xt+1 − Et [Xt+1 ]]
 e 
And also in terms of covariance of excess returns Ψt ≡ Covt Rt+1 :

∗ 1 1  e ′ −1  e  e 
mt+1 = − Et Rt+1 Ψt Rt+1 − Et Rt+1
Rtf Rtf

Compare it to its continuous-time analogous2

′
dmt∗

D
f
= −r dt − µ + − r f
Ψt−1 dZt
mt p

Recall that if the one-period stochastic discount factor mt is not time-varying (that is, the distribution
of mt is i.i.d for each t), then the expectations hypothesis holds and the investment opportunity set
does not vary. The corresponding Rt∗ of a single-factor state-price beta model is then relatively easy
to estimate, since over time we collect more and more realizations of Rt∗ . However, if mt (or the
corresponding Rt∗ ) is time-varying, then we can assume that it depends on some state variable and
need a multi-factor model to account for it. Indeed, suppose that Rt∗ = R∗ (zt ), where zt is some
state variable. For example, suppose that zt can be either 1 or 2 with equal probability. Then we
could take all the periods in which zt = 1 and back out R∗ (1), and find R∗ (2) in a similar way. Is it
possible to do so in reality? The answer is no, because we have hedges across state variables.

5.4 Martingales
Let {Xt } be a stochastic process and {xt } a sequence of its realizations. We say that {Xt } is a
martingale if E [Xt+1 | Xt = xt , Xt−1 = xt−1 , . . . , X1 = x1 ] = xt . Paul Samuelson argued in 1965 that
asset prices have to be martingales in equilibrium:

1
pt = EtQ [pt+1 + dt+1 ]
1 + rtf

This is trivial assuming no dividends, no discounting and risk-neutral agents. Let’s examine these
cases one by one.

With discounting, in order for the price process to follow a martingale we simply need:

Et [δ pt+1 ] = pt
2Where Zt is a Brownian Motion.
46 Chapter 5. The Multi-Period Model

With dividend payments, things are more complicated because pt depends on the dividend of the
asset in period t + 1 but pt+1 does not (it is the ex-dividend price). However, for example, the value
of a fund that keeps reinvesting the dividends follows a martingale (LeRoy 1989): suppose the fund
owns nothing but one unit of asset j. Assuming agents are risk-neutral, the value of the fund at
time 0 is
" #
h i
t j j
f0 = p0 = E ∑ t
δ x = δ E p 1 + x1
t≥1

After receiving a (state contingent) dividend x1j , the fund buys more of asset j at price p1 , thus
j
x1
owning 1 + p1 units of the asset. The expected discounted value of the fund is then
" !#
xj h  i
E [ f1 ] = E δ p1 1+ 1 = E δ p1 + x1j = p0 = f0
p1

So the discounted expected value of the fund is a martingale.

A similar statement is true of the agent is risk-averse. The difference is that we have to discount with
the risk-free interest rate instead of the agent’s discount factor, and use risk-neutral probabilities
(by now you should have an idea about why it is also called “equivalent martingale”) instead of
objective probabilities. Just like in the one-period model, we define the risk-neutral probability of
event A as

πt,s Mt,s
π Q (At,s ) =
ρt
 −1
t
f
where ρt = ∏ Rs is the risk-free discount factor between 0 and t. The initial value of the
s=0
fund is
" # " # " !#
t
f0 = p0 = E ∑Mt xtj =E M1 x1j + ∑ Mt xtj = E m1 x1j + ∑∏ mt xtj =
t≥1 t≥2 t≥2s=2

h  i h  i
= E m1 x1j + p1 = E M1 x1j + p1

Under the risk-neutral measure, this can be rewritten as

h i
f0 = ρ1 EQ x1j + p1 = ρ1 EQ [ f1 ]

Therefore, the properly discounted (ρ instead of δ ) and properly expected (π Q instead of π) value
of the fund is indeed a martingale.

We can now introduce a fifth asset pricing formula. Let Ps (t, T ) be the time-s price of a (zero-
coupon) bond to be purchased at time t and with maturity T . We know the pricing relation
5.4 Martingales 47

is

Pt (t, T ) = Et [mt+1 Pt+1 (t + 1, T )]

Dividing side by side the pricing relation for a generic asset j and rearranging we get
   
j j " j j
#
ptj Pt+1 (t + 1, T ) mt+1 xt+1 + pt+1 xt+1 + pt+1
F T
= Et   = Et
Pt (t, T ) Pt+1 (t + 1, T ) Et [mt+1 Pt+1 (t + 1, T )] Pt+1 (t + 1, T )

Where the expectation is taken with respect to the risk-forward measure FT for which πsFT =
S
πs Pt+1 (t+1,T )mt+1
Et [mt+1 Pt+1 (t+1,T )] (note that πsFT ≥ 0 and ∑ πsFT = 1 guarantee that π FT is a probability distribution).
s=1
This asset pricing formula was speficically devised to price bond options, and boils down to the
risk-neutral measure if t + 1 = T like in the one-period model case.
 48 Chapter 5. The Multi-Period Model

5.5 Exercises

1. Consider the following stochastic processes for a risk-free bond:

1.10 6.60

1.00 pu1

1.10 p0 3.30
1.00
1.20 5.00

1.00 pd1

1.20 3.50
Note that no asset pays dividends. The price of the one-period risk-free bond is normalized
1
to 1 in each period. Suppose that the risk-neutral probability of an “up” tick is 3 and that of a
2
“down” is 3, for every point in time.
a) Is the market statically complete? Is the market dynamically complete?
b) What is the event price of the event {up, up} and of the event of only one up, i.e.
{down, up} ∪ {up, down}?
Now suppose we change the process in the following manner:
1.10 6.60

1.00 pu1

1.10 p0 3.30
1.00
1.00 5.00

1.00 pd1

1.00 3.50
c) Is the market statically complete? Is the market dynamically complete?

2. Consider a security market model with three dates,t = 0, 1, 2 and five states of the world,
s = 1, ..., 5. Investors have no information at time 0 and full information at time 2. At the
intermediate date, their information partition consists of the two sets A = 1, 2, 3 and B = 4, 5.
There are two assets with the following price and dividend process:

• p10 = 3, p11 (A) = 32 , p11 (B) = 3;

• d11 (A) = 23 , d11 (B) = 2, d21 (s) = s;
• p20 = 49 , p21 (A) = 3, p21 (B) = 65 ;
• d12 (A) = 1, d12 (B) = 45 , d22 (s) = 6 − s

a) Is this price-dividend system arbitrage-free?

b) Is the contingent claim with the safe payoff stream yt = 1(t = 1, 2) attainable? If it is,
calculate its arbitrage price and a replicating portfolio strategy.
5.5 Exercises 49

3. Consider a security market with three dates, t = 0, 1, 2, and five states of the world, s = 1, . . . , 5.
The information structure is described by the following sequence of partitions of the state
space:

P0 = {1, ..., 5}
P1 = {A, B,C}
P2 = {{1}, {2}, {3}, {4}, {5}}

where A = {1, 2}, B = {3, 4},C = {5}. There are four securities with terminal payoffs given
by the matrix
 
2 4 5 1
1 1 2 1
 
 
D=
3 2 4 1

4 5 7 1
 

5 6 10 1

with the j-th column corresponding to the j-th security. The prices of these securities at date
t = 0, 1 are given by
 
3.6 4 6.6 1
 
 1.5 2.5 3.5 1
 
3.25 2.75 4.75 1
 
5 6 10 1

where the j-th column lists the price of the j-th security at date 0, event A, event B, and event
C, respectively. The securities do not pay dividends prior to the terminal date.
a) Verify that this securities market permits no arbitrage.
b) Is the market dynamically complete?
c) Compute the initial no-arbitrage price of the following three securities:
i. A call option on security 1 with an exercise price of 3.5.
ii. A down-and-under call option on security 1 with an exercise price of 3.5. This is a
call option with an extra provision - if the price of security 1 ever drops below the
exercise price, then option becomes worthless.
iii. A convertible security that in each state pays at the terminal date the largest payout
of the securities 1-4 in that state.
 6. Multi-Period Model: Options

6.1 One-Period Binomial Tree

Consider a European call option with strike K and maturity T . Its payoff is (ST − K)+ , with no
cash flows between t = 0 and t = T . Unfortunately we are unable to statically replicate this payoff
using just the stock a risk-free bond: we need to dynamically hedge: that is, we need to engage in a
strategy where the required stock (and bond) position changes for each period, until maturity. Since
the replication strategy depends on specified random process of stock price, we need to impose a
specific model for the evolution of the stock. In a discrete time setting, the canonical model is the
Cox-Ross-Rubinstein binomial model. We assume that:

• The stock pays no dividends.

• The length of a period is h and over each period we assume that the stock can either go up to
St+h = u × St , or down to St+h = d × St , so for each period t the distribution is binomial.
f
• The gross risk-free rate between periods is constant R f = er ·h , and by no arbitrage we require
d < R f < u.

We start with a one-period binomial tree (for h = 1):

uS
S
dS

Figure 6.1: A single-period binomial tree.

Suppose we buy (“go long” in trading jargon) ∆ stocks and B bonds. The payoff from portfolio
C0 = ∆S0 + B is

∆uS + B · R f if the stock goes up
f 0
C1 = ∆S1 + B · R =
∆dS + B · R f if the stock goes down
0

Call Cu the option payoff in the “up” state and Cd the option payoff in the “down” state. The
replicating strategy must satisfy

Cu = ∆uS0 + B · R f
Cd = ∆dS0 + B · R f
6.1 One-Period Binomial Tree 51

Solving for ∆ and B we get

Cu −Cd uCd − dCu

∆= and B=
S0 (u − d) R f (u − d)

∆ can be interpreted as the sensitivity of call price to a change in the stock price, or equivalently,
how much of the stock we should hold to hedge the option: for instance, to hedge a long call
position we need to sell ∆ units of the stock. Substituting ∆ and B in C0 we get

Rf −d u−Rf
 
Cu −Cd uCd − dCu 1
C0 = S0 + f = Cu + Cd
S0 (u − d) R (u − d) R f u−d u−d

The Risk Neutral Measure

R f −d u−R f
Define π Q ≡ u−d and note that u−d = 1 − π Q:

1 h Q 
Q
 i 1
C0 = f
π Cu + 1 − π Cd = f EQ [C1 ]
R R

Therefore the option price is the discounted payoff of the option under the equivalent martingale
(risk-neutral) measure Q. Note that Q is also the probability measure that would justify the current
stock price in a risk-neutral world:

1 h Q 
Q
 i
S0 = π uS0 + 1 − π dS0
Rf

Note how we never even mentioned the physical probability measure π, since we are working with
relative asset pricing. To price a call or any other derivative we do not need to know the actual
probability P, but Q. We will be taking expectations with respect to Q, so we can put π Q directly
on the branches.

Figure 6.2: Dynamic replication.

 52 Chapter 6. Multi-Period Model: Options

6.2 Two-Period Binomial Tree

A two-period binomial tree is just a concatenation of single-period binomial trees:

u2 S
uS
S udS
dS
d2S

Figure 6.3: A two-period binomial tree.

To price the option at time t = 0 we just apply the same procedure backwards, starting from the
two sub-trees in the final period. To summarize,

1. We compute the risk-neutral probability π Q from the stock price

2. We plug these probabilities in the formula for C at the final two nodes, and once we find the
option prices for each of the two possible states in time t = 1 we repeat the procedure for the
initial node
3. We find ∆ and B at each time to find the replicating strategy.

The general procedure for a two-period tree is illustrated below (in the figure below, p = π Q ):

Figure 6.4: Pricing procedure for a two-period binomial tree.

Suppose we observe an option price on the market of $36, while the cost of the dynamic replication
strategy is $34.08: then we can take advantage of this arbitrage opportunity by selling the option
and buying the synthetic portfolio. Today we pocket $1.92, and at maturity our cash outflows equal
our inflows.
 6.3 The Relation with the Black-Scholes Model 53

6.3 The Relation with the Black-Scholes Model

In this section, we take a glimpse at the Black-Scholes model for options pricing, according to
which a European call option on a stock worth S0 with strike K and time-to-maturity T today is
worth (assuming zero interest rates)
2 2
! !
ln SK0 + σ2 T ln SK0 − σ2 T
S0 N √ − Ke−rT N √
σ T σ T

where σ is the annualized standard deviation, or volatility, of stock returns.

The Black-Scholes model can be viewed as the limit of a binomial tree,

√ in which the number
√ of
periods n (and therefore of states) goes to infinity. To see this, take u = eσ T /n
, d = 1/u = e−σ T /n

and R f = e r f ·T /n where T is the time to expiration and σ is the annualized standard deviation of the
stock log-returns. The general binomial formula for a European call on non-dividend paying stock
n periods from expiration is:
" #
n
1 n!  j  n− j
+
C0 = f n ∑ j! (n − j)! π Q
1 − π Q
u j d n− j S0 − K
(R ) j=0

Plugging in u, d and R f and taking the limit as n → ∞ we get

 2
    
f − σ2 T
 
ln SK0 + r f + σ2 T ln S0
+ r 2
S0 N  √  − Ke−r f T N  K √ 
σ T σ T

2
 
ln
S0
K + r f + σ2 T √
Let d1 = √
σ T
and d2 = d1 − σ T , we can interpret the trading strategy under the Black-
f
Scholes formula as ∆ = N (d1 ) ∈ [0, 1] and B = −Ke−r T N (d2 ). We can also find the price of a put
option: by put-call parity, we have

fT f
P0 = C0 − S0 − Ke−r = Ke−r T N (−d2 ) − S0 N (−d1 )

Note that the put option has ∆ = −N (−d1 ) ∈ [−1, 0]

We call intrinsic value of the option the quantity (St − K)+ and time value of the option the quantity
Ct − (St − K)+ .

6.4 Delta-Hedging
A stock has a delta equal to one, while a portfolio of J assets - where each asset j has notional
J
amount N j - has a delta △ p = ∑ ∆ j N j . We say that the portfolio is delta neutral if ∆ p = 0. Delta-
j=1
hedging a portfolio is only a “perfect” hedge (yielding the same final payoff as the corresponding
option) with continuous trading: computing the delta is a linear approximation to the option value,
but the option price is convex and therefore the second- and higher-order derivatives matter. Delta-
hedging when trading is not continuous is effective only for small price changes. Delta-Gamma
 54 Chapter 6. Multi-Period Model: Options

hedging reduces this convexity risk, and involves trading with other options.

6.5 The Volatility Smile

The Black-Scholes model assumes that the volatility σ is constant over time and through strikes.
This turns out to be a bad assumption: suppose we take option prices as they are observed in the
market, and then back out the volatility implied by these prices for many values of the time to
maturity T and strike K. It turns out that the volatility implied by options on the market exhibits
a volatility smile, while the Black-Scholes model implies a constant parameter σ in the (T, K)
space. Interestingly, this effect was almost absent before the 1987 “Black Monday” stock market
crash, and became very evident afterwards. This means that the Black-Scholes model underprices
out-of-the money puts (and thus in-the-money calls) and overprices out-of-the-money calls (and
thus in-the-money-puts).

One way to get around this problem is to use a different model for the volatility of the stock called
“stochastic volatility”. Other issues in the Black-Scholes include stochastic interest rates, bid-ask
spreads, other transaction costs, etc.

6.6 Collateralized Debt Obligations

Collateralized Debt Obligations (CDOs) are derivatives which repackage cash-flows from a set of
assets. Payoffs are divided in tranches: in the event of under-performance of the underlying assets,
the Senior tranche is paid out first, the Mezzanine second, and the Junior tranche last. Option theory
is very useful in pricing CDOs: the tranches can be priced using analogues from option pricing
formulas, and it is possible to estimate the implied default correlations between the underlying
assets that correctly price the tranches.
 6.7 Exercises 55

6.7 Exercises

1. Consider a binomial tree, and the stock trades at S0 = 80. At each period the stock can go up
with probability p in which case St+1 = 1.2St or it can go down with probability 1 − p and
St+1 = 0.8St .The risk-free rate for each period is 10% and we are using discrete discounting
of R f = 1 + r f
a) One-Period Model:
i. What is the probability of stock price going up under the risk-neutral measure?
ii. Suppose we long ∆ shares of stock and B dollars in risk-free bond to replicate the
payoff for a European call option with strike K = 80. Please calculate ∆ and B.
iii. What’s the interpretation for ∆? Suppose we long 100 European calls and we
want to hedge the risk that the portfolio/total assets we have changes when the
underlying stock price changes over time. How to use stocks to hedge such risk?
Please indicate the long/short position and the amount of stocks.
iv. For ∆, what’s its corresponding term in Black-Scholes “Greeks?” (Please write in
the form of partial derivatives).
b) Multi-Period Model (3-Period)
i. What is the price of the prepaid forward (at t=0) delivering the stock at t = 3 (you
pay at t = 0).
ii. What is the price at t = 0 of the European call on this stock with K = 100 and
maturity t = 3?
iii. What is the price at t = 0 of an American put on this stock with strike K = 95 and
maturity t = 3? (Hint: early exercise at each node is possible.)

† 2. Consider the binomial model with u = 1.2, d = 0.9, R = 1.1 and initial asset price S = 100.
a) Calculate the price of a call option with exercise price K = 100 and n = 4 periods left
until expiry.
b) Calculate the prices of the European and American put options with the same exercise
price and time to expiry as in (a).
c) Calculate the price of an option which, at the end of the third period, gives its holder
the right to purchase the underlying asset at the minimum price realized over the life of
the option.
† 3. A stock price is currently $30. During each 2-month period for the next 6 month it will
increase by 8% or reduce by 10%. The annual risk-free interest rate is 5% (continuous
compounding). Use a three-step tree to calculate the value of a derivative (called turbo
option) that pays off ST × (30 − ST )+ , where ST is the stock price in 6 months.
 7. Risk Preferences and Expected Utility Theory

So far in this course we dealt with relative asset pricing: we derived asset prices through information
available on other, pre-existing assets. In this chapter we will study absolute asset pricing, a task
for which we have to specify the agents’ preferences towards risk. We will see that risk involves
knowledge of the probabilities of uncertain events, and agents’ attitude towards risk determines
how much they would be willing to pay for an asset - that is the core of absolute asset pricing.
A concept related to risk that we will not see in detail is that of uncertainty, which relates to the
agents’ preferences over events whose probability distribution is unknown.

It is important to stress that preferences over risk refer to and are defined over “final payoff gambles”,
that is, after combining any random payoffs, such as the payoff of securities, investments, insurance,
and even non-random endowments.

We will start this chapter trying to answer the following question: is there a criterion that allows us
to compare random payoffs?

7.1 State-by-State Dominance

A first attempt to answer this question is given by the principle of state-by-state dominance. Suppose
we have to choose one of the following three investments:

t =0 t =1
1
Cost Probabilities π1 = π2 = 2
s=1 s=2
Investment 1 -1000 1050 1200
Investment 2 -1000 500 1600
Investment 3 -1000 1050 1600

It is hard to tell which one is better between investment 1 and 2: the two possible states they have
the same probability and in state 1 investment 1 pays more, while in state 2 investment 2 pays more.
However, investment 3 pays at least as much as investment 1 and 2 in any state of the world! Clearly
from this perspective investment 3 is more desirable, and we say that it state-by-state dominates
investments 1 and 2.

Definition 7.1.1 — State-by-State Dominance. Given two random variables X and Y defined
over the probability space (Ω, F , P), we say that Y state-by-state dominates X if ∀ω ∈ Ω
X (ω) ≤ Y (ω).

We have already developed the notation needed to express state-by-state dominance: recall that for
 7.2 Stochastic Dominance 57

x, y ∈ Rn we write

1. y ≥ x if for each i = 1, ... , n yi ≥ xi

2. y > x if y ≥ x and y ̸= x
3. y ≫ x if for each i = 1, ... , n yi > xi

Assuming that x and y are scaled to have the same price, under any of the three conditions above
we would say that y state-by-state dominates x (you will notice that in the list above we see an
“increasing degree” of dominance).

A problem with state-by-state dominance is that it is an incomplete ranking. In fact, it is as

incomplete as RS . Another popular criterion which has been widely used is the mean-variance
dominance: assume that we like more expected return and less volatility. Consider the following
example:

t =0 t =1
1
Cost Probabilities π1 = π2 = 2 Moments
s=1 s=2 E [R] σ (R)
Investment 1 -1000 +5% +20% +12.5% 7.5%
Investment 2 -1000 -50% +60% +5% 55%
Investment 3 -1000 +5% +60% +32.5% 27.5%

Investment 1 has a higher expected return and lower volatility than investment 2. However, it is
not the case that investment 1 state-by-state dominates investment 2! Moreover, while investment
3 state-by-state dominates both investment 1 and 2, it only mean-variance dominates investment
2. Evidently, mean-variance dominance and state-by-state dominance are somehow unrelated
orderings. However, while mean-variance dominance is an incomplete ranking, it does simplify the
problem of ranking investments to two dimensions only (it is as incomplete as R2 ). It looks like,
potentially, a way of getting round the incompleteness problem is to find a criterion that amounts
to some number, for instance, in R, an ordered set. For example, suppose we rank investments
according to which Sharpe Ratio is highest:
Definition 7.1.2 — Sharpe Ratio.

E [R] − r f
Sharpe Ratio =
σ (R)

In the example above, assuming r f = 3%, we get 1.27, 0.04 and 1.07 for investment 1, 2 and 3
respectively, and therefore we would select investment 1 (which, remember, we would eliminate
based on both state-by-state dominance and mean-variance dominance!)

7.2 Stochastic Dominance

Before going on, it is worthwhile to remark that we can write the “payoff per state” representation
of random payoffs:
58 Chapter 7. Risk Preferences and Expected Utility Theory

State s1 s2 s3 s4 s5
Probability π1 π2 π3 π4 π5
Payoff X 10 10 20 20 20
Payoff Y 10 20 20 20 30

in terms of “probability lotteries”:

Payoff 10 20 30
Probability X p10 = π1 + π2 p20 = π3 + π4 + π5 p30 = 0
Probability Y q10 = π1 p20 = π2 + π3 + π4 p30 = π5

This allows us to define preferences px ≻ py over probability distributions over states (rather than
on payoffs x ≻ y ∈ RS ) and will turn out to be useful when thinking of final payoffs. Note that
we can always go from the“payoff per state” to the “probability lotteries” representation, but the
reverse is not true.

Another criterion to rank investment is the stochastic dominance. It is related to state-by-state

dominance, and is still an incomplete ordering of investments. Suppose there are two investment
opportunities 1 and 2:

Event e1 e2 e3
Payoff 50 100 200
Probability 1 40% 60% 0%
Probability 2 40% 40% 20%

In this case, every investment pays the same amount in each event, but the probabilities of the
events happening vary. To reconcile this with our previous “states of the world” discussion, it is
enough to note that the above example is such that investment 1 pays off 100 in the states s2 and s3
so that e2 = {s2 , s3 }, while investment 2 pays off 100 in state 2 and 200 in state 3 (and obviously
e1 = s1 ).

F(x)
F1
1

0.8
F2
0.6

0.4

0.2

0 50 100 150 200x

Figure 7.1: CDF of investment 1 and 2.

7.2 Stochastic Dominance 59
Definition 7.2.1 — First Order Stochastic Dominance. Let FA and FB represent, respectively,
the cumulative distribution functions of two random variables (investments payoff) defined in
the interval [a, b]. We say that FA first-order stochastically dominates (FSD) FB if for all x in the
interval [a, b], FA (x) ≤ FB (x).

Visually, FSD occurs whenever FA “stays below” FB for the whole domain. State-by-State Domi-
nance implies First Order Stochastic dominance.

1 cdf(x)

0.8

0.6

0.4

0.2
A
0 B
x
−10 −8 −6 −4 −2 0 2 4 6

Figure 7.2: First Order Stochastic Dominance

Next, consider the following example:

Payoff 1 4 5 6 8 12
Probability 3 0% 25% 50% 0% 0% 25%
Probability 4 33% 0% 0% 33% 33% 0%

It is easy to verify that FSD does not hold for any investment. However, investment 3 is somewhat
special because its CDF stays below investment 4’s “most of the time”.

F(x)
F4
1

0.8
F3
0.6

0.4

0.2

0 2 4 6 8 10 12x

Figure 7.3: CDF of investment 3 and 4.

 60 Chapter 7. Risk Preferences and Expected Utility Theory
Definition 7.2.2 — Second Order Stochastic Dominance. Let FA and FB represent, re-
spectively, the cumulative distribution functions of two random variables (investments payoff)
defined in the interval [a, b]. We say that FA second-order stochastically dominates (SSD) FB if
Rx
∀x ∈ [a, b] a [FB (t) − FA (t)] dt ≥ 0.

Clearly, in our example, investment 3 SSD investment 4.

Proposition 7.2.1 First Order Stochastic Dominance implies Second Order Stochastic Dominance.

Another related concept is that of mean-preserving spread:

Definition 7.2.3 — Mean-Preserving Spread. We say that the random variable XA is a mean-
preserving spread of the random variable XB if XA = XB + ε, where the random variable ε is
independent of XA and XB , and has zero mean and positive variance.

For normally distributed XA , XB and ε the picture looks like:

0.4 f (x)

0.3

0.2

0.1
fA
0 fB
x
−6 −4 −2 0 2 4 6

Figure 7.4: Mean-Preserving Spread (for normal distribution)

The concept of mean preserving spread and that of second-order stochastic dominance are connected
by the following proposition:

Proposition 7.2.2 Let FA and FB be the CDFs of two random variables XA and XB defined on the
same space with identical means. Then FA SSD FB if and only if XA is a mean-preserving spread of
XB .

7.3 von Neumann Morgenstern Expected Utility Theory

So far we mostly encountered criteria to rank uncertain payoffs which are incomplete, in the sense
that we can encounter two investments that we are unable to compare and decide which one is better.
We saw that this is linked to the fact that the criteria examined require to compare finite-dimensional
vectors (in the case of state-by-state dominance and mean-variance dominance) or even functions,
7.3 von Neumann Morgenstern Expected Utility Theory 61

which can be considered as infinite-dimensional vectors (in the case of first and second-order
stochastic dominance). In this section we will elaborate another method which aims at simplifying
the problem by assigning to each investment a number u ∈ R, so that every two investments are
comparable. To do this by avoiding arbitrary choices of this function that maps investments onto
numbers, we will assume the existence of regular preferences over investments.

Let’s start with an example. Suppose you can enter the following bet: we flip a fair coin once. If
tails comes up, I’ll give you $1. If heads comes up, we’ll flip again, and if tails comes up I’ll pay
you $2, while if heads comes up we’ll flip again. Every time we flip again, the prize for a tails up
doubles. The question is: what is the fair price for this bet?

1 n


Since for the n-th flip the probability of a tails is 2 and the payoff is 2n−1 , and the number of
flips ranges from 1 to infinity, it is straightforward to see that the expected payoff is
 n
∞ ∞
1 1
Expected Payoff = ∑ × 2n−1 = ∑ = ∞
n=1 2 n=1 2

However this is the correct expected value, it does not make sense to pay all the money in the
world for this coin-flipping game (investment), since it pays you less that $64 in 99% of cases.
Clearly this criterion is inappropriate to value uncertain payoffs. This problem is was named the
St. Petersburg paradox and was first proposed by Daniel Bernoulli, who proposed the following
solution: because when you are rich an additional dollar is less valuable to you than when you are
poor, we should weight less the largest payoffs. He proposed to discount large payoffs using the
natural logarithm: introduce the parameter p for the denominator of the probability, in the previous
example equal to 2. Then we can write:
 n
∞
1
× ln 2n−1


Bernoulli Value = ∑
n=1 p
∞
= ln 2 ∑ (n − 1)p−n
n=1
∞
= − ln 2 ∑ (1 − n)p−n
n=1
∞
d h (1−n) i
= − ln 2 ∑ p
n=1 d p
" #
∞
d (1−n)
= − ln 2 ∑p
d p n=1
" #
∞
d 1
= − ln 2 ∑ pn
d p n=0
 
d 1 1
= − ln 2 = ln 2
dp 1− p (1 − p)2

Which for p = 2 is just ln 2. If the value is ln 2 and the function used to discount large payoffs is
the natural logarithm, we deduce that the price we are willing to pay for the coin-flipping game is
62 Chapter 7. Risk Preferences and Expected Utility Theory

just x such that ln x = ln 2, that is, $2. Let’s dig deeper into this idea of specifying a function that
maps uncertain payoffs (investments) to a single number (utility).

Representation of Preferences
Suppose all we care about is consumption in each state. Then, the choice set would consist
of vectors in RS+1 where the first element represents consumption at t = 0, and all other el-
ements represent consumption in the possible S states at t = T . To reconcile this definition
with our one-period model, suppose an investor has an initial wealth W (constant across states
and time) and is considering buying a security j with price p j and payoff x j . The vector
W − p j , (W − p j ) × I + x j = (c0 , c1 , ... cS )T ∈ RS+1 (where I is a S × 1 vector where each el-

ement is equal to one) would be the corresponding consumption profile. Moreover, given two
consumption profiles c, c′ ∈ RS+1 , if an investor chose c over c′ we say that that investor prefers c
to c′ and write c ≻ c′ .

Theorem 7.3.1 — Representation Theorem. Suppose the preference ordering is i) complete ii)
transitive iii) continuous (and satisfies some further regularity conditions that we do not discuss
here). Then the preference ordering can be represented by a utility function, that is, c ≻ c′ if and
only if there exists a utility function U such that U(c) > U(c′ ).

A related but different approach was developed by John von Neumann and Oskar Morgenstern. In
this setup, we define the (finite) set of outcomes X (in the previous setup for instance, this was the
uncertain consumption profile at time t = T ) and we define a lottery as a probability measure over
X, that is a function p : X → [0, 1] such that 1) for x ∈ X : p(x) ≥ 0, 2) ∑ p(x) = 1. The set of all
x∈X
possible lotteries is
( )
P≡ p : X → [0, 1] | ∀x ∈ X : p(x) ≥ 0, ∑ p(x) = 1
x∈X

A preference relation expresses a person’s feelings between pairs of objects in P. We denote the
preference relation by ⪰, which means “at least as good as”. Two properties of a preference relation
are defined below.
Definition 7.3.1

1. A preference relation on P is complete if for every pair p and q from P , either p ⪰ q or

q ⪰ p (or both).
2. A preference relation on P is transitive if p ⪰ q and q ⪰ r implies that p ⪰ r.

Assume there exists a preference relation ≻ over P and consider the following three axioms:
Definition 7.3.2 — von Neumann - Morgenstern Axioms.

1. (Regularity) Agents have complete and transitive preferences over P.

7.3 von Neumann Morgenstern Expected Utility Theory 63

2. (Independence) For any three lotteries p, q, r ∈ P and α ∈ (0, 1], p ⪰ q ⇔ α p+(1−α)r ⪰

αq + (1 − α)r
3. (Continuity) For any three lotteries p, q, r ∈ P such that p ≻ q ≻ r there exist α, β ∈ (0, 1)
such that α p + (1 − α)r ≻ q ≻ β p + (1 − β )r

Then the following theorem holds:

Theorem 7.3.2 — Expected Utility Representation. If a preference relation ≻ over P satisfies

the Regularity, Independence and Continuity axioms, then there exists a function u : X → R such
that

p⪰q⇔ ∑ p(x)u(x) ≥ ∑ q(x)u(x)

x∈X x∈X

That is, preferences over lotteries correspond to the expected utility of the lotteries. Below is a
graphical representation of an expected utility function over a lottery defined on a two-dimensional
state space:

x2

45◦

U(x1 , x2 ) = k

x1
0 z πx1 + (1 − π)x2 = c

Figure 7.5

The 45° line represents the set of lotteries that pay off the same amount in both states. The
negatively sloped straight line is the set of lotteries in the (x1 , x2 ) plane such that the expectation
of the lottery is equal to a value c > 0, that is, if π is the probability of the state 1 occurring,
π c
c = πx1 + (1 − π) x2 , or x2 = 1−π x1 + 1−π . The convex curve represents the indifference curve for
the agent: to give up a certain consumption amount in state 2, he is willing to take some amount in
state 1 to keep his utility level constant. The curve is the locus of lotteries in the (x1 , x2 ) plane such
that k = πu (x1 ) + (1 − π)u (x2 ), k ∈ R.

Going back to the axioms, consider the Independence axiom: it basically states that if we “dilute”
 64 Chapter 7. Risk Preferences and Expected Utility Theory

p and q with some third lottery r in the same way, but we prefer p to q, then our preference ranking
should not change. It can be visualized using trees:

π p π q
p⪰q ⇔ ⪰
r r

Figure 7.6

However, this axiom rarely holds in experiments. For instance, suppose you are given the choice
between lottery p that pays $10 with probability 10% (and zero otherwise) and lottery q that pays
$15 with probability 9% (and zero otherwise). Because the prize is 50% higher in lottery q than
in lottery p and the probability of winning a positive amount is only 10% lower in lottery q than
lottery p, most people in experiments choose lottery q. However, between a lottery u that pays $10
with 100% probability and a lottery v that pays $15 with 90% probability, most people choose the
sure bet implied by lottery u. However, it turns out that lotteries u and v can be mixed with a third
lottery r that pays zero with probability 100%, using a parameter α = 10%, to obtain lotteries p
and q again(see picture below): the preference relation between p and q is now the opposite of
the initial preference, and therefore this constitutes an empirical violation of the independence
axiom.1

10% 10 9% 15
≺
0 0

Figure 7.7

100% 10 90% 15
10% 10%
0 ≻ 0

0 0

Figure 7.8

7.4 Risk Aversion, Concavity, Certainty Equivalent

Consider a lottery p with possible outcomes x0 < x1 . We define the certainty equivalent as the
certain payoff which gives the same expected utility as the uncertain lottery p. Clearly, if E [u(x)]
is the expected utility of lottery p, u−1 (E [u(x)]) will be the certainty equivalent of p. Now
suppose that the agent is risk-averse: clearly, for him the certainty equivalent will be lower than
the expected value of the lottery (he will be willing to “give up” some money off the expected
value of the lottery to have a certain payoff). As it turns out, this property is equivalent to the von
1 Top left: lottery p. Top right: lottery q. Bottom left, in black: lottery u. Bottom right, in black: lottery v. In red:

with α = 10% and r = 0 we can transform u and v into p and q.

7.4 Risk Aversion, Concavity, Certainty Equivalent 65

Neumann-Morgenstern function u being concave.

u−1 (E [u(x)]) < E [x] ⇔ u′′ (x) < 0

u(x)

u(E[x])

E[u(x)])

x
x0 CE E[x] x1

Figure 7.9: Concavity and certainty equivalence.

And the difference E [x] − u−1 (E [u(x)]) is called risk premium. This fact derives directly from
Jensen’s Inequality:

Theorem 7.4.1 — Jensen’s Inequality. Let g be concave over [a, b] and let x be a random
variable such that P [x ∈ [a, b]] = 1. If the expectations E [x] and E [g (x)] exist, then E [g (x)] ≤
g (E [x]). Furthermore, if g(·) is strictly concave then the inequality is strict.

The following theorems establish a relationship between First and Second Order Stochastic Domi-
nance and Expected Utility:

Theorem 7.4.2 Let FA and FB be two CDFs for the random payoffs x ∈ [a, b]. Then FA FSD FB
if and only if EA [U (x)] ≥ EB [U (x)] for all nondecreasing utility functions U(·).

Proof. (⇐) We prove by contrapositive. Suppose FA does not FSD FB . Then, we can find at least
one x∗ ∈ [a, b] such that FA (x∗ ) > FB (x∗ ). Define nondecreasing utility function u(x) = 1{x > x∗ }.
Clearly,
Z b Z b
EA [u(x)] = fA (x)dx = 1 − FA (x∗ ) < 1 − FB (x∗ ) = fB (x)dx = EB [u(x)]
x∗ x∗

(⇒) First, using integration by parts, we can show that

Z b Z b
u(x) f (x)dx = u(b) − u′ (x)F(x)dx
a a
 66 Chapter 7. Risk Preferences and Expected Utility Theory

Then we can show

Z b Z b Z b
u(x) fA (x)dx − u(x) fB (x) = u′ (x)(FB (x) − FA (x))dx
a a a

which is nonnegative if FA FSD FB . ■

Theorem 7.4.3 Let FA and FB be two CDFs for the random payoffs x ∈ [a, b]. Then FA SSD FB
if and only if EA [U (x)] ≥ EB [U (x)] for all nondecreasing and concave utility functions U(·).

We conclude this subsetion with a parallel between (non-expected) utility representation and
expected utility representation. In the former case, suppose U(c) = U (c0 , c1 , ... cS ) represents a
complete, transitive and continuous preference ordering between consumption profiles. Then also
V (c) = f (U(c)), for f strictly increasing, represents the same preference ordering. In the latter case,
suppose that E [u(c)] represents a preference ordering satisfying the von Neumann-Morgenstern
axioms. Then for a, b ∈ R the affine function v(c) = a + bu(c) represents the same preference
ordering.

7.5 Measures of Risk Aversion

We define the following measures of risk aversion defined over the wealth Y of the agent:

Definition 7.5.1
(Y ) ′′
1. Absolute Risk Aversion: RA (Y ) = − uu′ (Y )
(Y ) ′′
2. Relative Risk Aversion: RR (Y ) = −Y · uu′ (Y ) = Y · RA
1
3. Risk Tolerance: RT (Y ) = RA (Y )

There is an interesting link between expected utility and these measures of risk aversion: consider
the following lottery, where Y is the initial wealth of the agent:

π Y +h

1−π Y −h

What is the change in π needed for an agent to be indifferent when h rises? We can Taylor-expand
to the second degree the expected utility in a neighbourhood of Y to obtain:

1 1
π(Y, h) = + hRA (Y ) + o(h)
2 4

Similarly, consider the lottery

7.5 Measures of Risk Aversion 67

π Y (1 + θ )

1−π Y (1 − θ )

Again, the change in π needed for an agent to be indifferent when h rises is obtained by Taylor-
expanding to the second degree the expected utility in a neighbourhood of Y to obtain:

1 1
π(Y, h) = + θ RR (Y ) + o(θ )
2 4

Similarly, suppose we add a zero-mean small risk (lottery) X to the initial wealth Y0 . Since
E [u(Y0 + X)] = u Y CE , we can use a second order Taylor expansion to show that (Hint: let risk

premium Π ≡ Y0 −Y CE , note that u(Y0 − Π) = E[u(Y0 + X)]. Taylor-series expand both sides around
Y0 .)

Var(X)
Y0 −Y CE ≈ RA (Y0 )
2

where Y0 −Y CE is the risk premium, and it is approximately linear in the variance of the additive
risk, with slope equal to half the coefficient of absolute risk aversion.

Similarly, suppose we have a multiplicative risk. Suppose that E [u(Y0 (1 + θ ))] = u (kY0 ), where
1 + θ is a positive random variable with unit mean and k is is the certainty equivalent growth rate.
This time we get

Var(θ )
1 − k ≈ RR (Y0 )
2

Therefore the coefficient of absolute risk aversion is relevant for additive risk, while the relative
risk aversion is relevant for multiplicative risk.

We now introduce two specific functional forms for von Neumann-Morgenstern function U:
Definition 7.5.2

• Constant Absolute Risk Aversion (CARA) utility function: U(x) = −e−ρY , for some
ρ ∈ R+ 
 Y 1−γ i f γ ̸= 1
1−γ
• Constant Relative Risk Aversion (CRRA) utility function: U(x) =
lnY if γ = 1

For example, consider the CRRA utility and a lottery that adds $50,000 or $100,000 to the initial
wealth Y . The certainty equivalent is defined as the number CE ∈ R such that
 68 Chapter 7. Risk Preferences and Expected Utility Theory

(Y +CE)1−γ 1 (Y + 50, 000)1−γ 1 (Y + 100, 000)1−γ

= × + ×
1−γ 2 1−γ 2 1−γ

For Y = 0 we get

γ =0 CE = 75, 000 (risk neutrality)

γ =1 CE = 70, 711
γ =2 CE = 66, 246
γ =5 CE = 58, 566
γ = 10 CE = 53, 991
γ = 20 CE = 51, 858
γ = 30 CE = 51, 209

While for Y = 100, 000 and γ = 5 the certainty equivalent is already 66,530 (close to γ = 2 for
Y = 0!)

7.6 Risk Aversion and Portfolio Allocation

Suppose that all assets are consumed in t = T , so that there are no savings in t = T . Moreover,
assume there is a riskless asset with net return r f and a risky asset with a random net return r. We
want to maximize the expected utility by choosing the risky asset allocation parameter a ∈ R:

max E U Y0 1 + r f + a r − r f




a∈R

for some initial wealth Y0 . The problem has first order conditions (FOC)

E U ′ Y0 1 + r f + a r − r f r−rf = 0






We can characterize the solution to the problem with the following theorem:

Theorem 7.6.1 Assume U ′ > 0, U ′′ < 0 and let â denote the solution to the problem above.
Then

â > 0 ⇔E [r] > r f

â = 0 ⇔E [r] = r f
â < 0 ⇔E [r] < r f

Proof. define W (a) ≡ E U Y0 1 + r f + a r − r f , then the FOC can br written as W ′ (a) = 0.






By risk aversion (that is, since U ′ > 0 and U ′′ < 0) we have

h
2 i
W ′′ (a) = E U ′′ Y0 1 + r f + a r − r f r−rf




< 0,
7.6 Risk Aversion and Portfolio Allocation 69

so W ′ (a) is everywhere decreasing in a. This implies that â > 0 if and only if W ′ (0) > 0, and since
U ′ > 0 it follows that â > 0 if and only if E [r] > r f . The other assertions follow similarly. ■

How do measures of risk aversion depend on the initial wealth Y0 ? This is the subject of Arrow’s
theorems (1971). The first theorem is about the absolute measure of risk aversion

Theorem 7.6.2 — Arrow (1971) Part 1. Let â = â (Y0 ) be the solution to the problem above.
Then

∂ RA ∂ â
(DARA) <0⇒ >0
∂Y0 ∂Y0
∂ RA ∂ â
(CARA) =0⇒ =0
∂Y0 ∂Y0
∂ RA ∂ â
(IARA) >0⇒ <0
∂Y0 ∂Y0

Proof. No proof. ■

The second theorem is related to the relative risk-aversion measure.

Theorem 7.6.3 — Arrow (1971) Part 2. If, for all wealth levels Y :

∂ RR d â/â
(DRRA) <0⇒ >1
∂Y dY /Y
∂ RR d â/â
(CRRA) =0⇒ =1
∂Y dY /Y
∂ RR d â/â
(IRRA) >0⇒ <1
∂Y dY /Y

Proof. No proof. ■

d â/â
Note that the quantity dY /Y is the wealth wealth elasticity of investment in the risky asset. If
d â/â
dY /Y > 1, then as wealth increases, the optimal proportion invested in the risky assets also increases.
Below we also provide an example for CRRA utility function.

In the special case U(Y ) = lnY the FOC is

r−rf
 
E =0
Y0 (1 + r f ) + a (r − r f )

and assuming that r can take on two possible values r1 and r2 with probabilities π and 1 − π,
respectively with r1 < r f < r2 , it is possible to show that

1 + r f E [r] − r f



a
= >0
Y0 (r2 − r f ) (r f − r1 )
70 Chapter 7. Risk Preferences and Expected Utility Theory

That is, a constant fraction of the initial wealth is invested in the risky asset.

How does all this generalize to the J assets case? The answer is provided by the following
theorem:
Theorem 7.6.4 — Cass and Stiglitz. Let the vector
 
â1 (Y0 )
 . 
 .. 
 
âJ (Y0 )

denote the amount optimally invested in the J risky assets if the initial wealth is Y0 . Then
   
â1 (Y0 ) â1
 .  .
 ..  =  ..  · f (Y0 )
   
âJ (Y0 ) âJ

if and only if either i) U ′ (Y0 ) = (θY0 + κ)∆ or ii) U ′ (Y0 ) = ξ e−vY0 .a b

a Restrictionson parameters will be imposed to ensure that U ′′ (Y0 ) < 0.

bWecan recover the utility functions corresponding to these marginal utilities by intagrating i) and ii). It is easy to
show that i) include CRRA and ii) corresponds to CARA. functions:

In other words, in the case of constant absolute or constant relative risk-aversion preferences, it
is sufficient to offer a mutual fund: the optimal proportion to invest in each asset is given by the
vector [â1 , . . . , âJ ]T , and agents would then invest an amount depending on their initial wealth level
according to the function f (Y0 ).

To conclude this subsection, we generalize the class of utility functions by starting from the absolute
risk aversion (and then backing out the resulting utility function).

We call linear risk tolerance (or hyperbolic risk aversion) any utility function such that

u′′ (c) 1
RA = − =
u′ (c) A + Bc

Clearly, for B = 0 and A ̸= 0 we have a CARA utility function. If B ̸= 0 we obtain a generalized

power function

1 B−1
u(c) = (A + B · c) B
B−1

for which if B → 1 we obtain the log utility u(c) = ln (A + Bc), for B = −1 we obtain the quadratic
B−1
utility − 21 (A − c)2 , and for A = 0 we obtain the CRRA utility function u(c) = 1
B−1 (Bc) B .
 7.7 Savings 71

7.7 Savings
Within our one period model, consider an asset structure composed of one risk-free asset with
constant gross return R f . The agent has to make a decision on how much to consume in each period
t = 0 and t = T . If R f increases, we will see two effects: one is the incentive to save more at time
t = 0, since by doing so we can now consume more in t = T - this is called substitution effect - and
the incentive to consume more (save less) at time t = 0, since the agent is effectively richer when
the gross riskless return increases - this is called income effect. In the log utility case, these two
effects cancel each other; in other cases (γ ̸= 1) one effect can prevail over the other.

Suppose now there is a risk-free assets but the endowment in the future period is random. How does
the behavior of the agent change when we increase his exposure to risk? An old idea (going back
to J. M. Keynes) is that people save more when they face greater uncertainty. This phenomenon
is called precautionary savings and it generally arises under two circumstances, one linked to the
shape of the utility function and one to the presence of borrowing constraints.

Let’s jump forward a little and suppose that in a multi-period setting agents maximize the expected
utility
" #
∞
E0 ∑ β t u (ct )
t=0

subject to the intertemporal budget constraint

ct+1 = et+1 + (1 + r) (et − ct )

The standard Euler equation is

u′ (ct ) = β (1 + r) Et u′ (ct+1 )
 

And if u′′′ > 0, Jensen’s inequality implies

1 Et [u′ (ct+1 )] u′ (Et [ct+1 ])

= >
β (1 + r) u′ (ct ) u′ (ct )

which shows that the marginal rate of intertemporal substitution is higher in the presence of
uncertainty in ct+1 . The difference between the two marginal rates, with and without uncertainty,
is attributed to precautionary savings. to see this, suppose the variance of et+1 increases (in a
mean preserving fashion). Since numerator Et [u′ (ct+1 )] is increasing in the variance of ct+1 , in
order for the above equality to hold ct must decrease, that is, savings increase due to precautionary
savings.

Suppose now there are no risk-free assets and one risky asset with random gross return R.

Theorem 7.7.1 — Rothschild and Stiglitz (1971). Let RA and RB be two return distributions
 72 Chapter 7. Risk Preferences and Expected Utility Theory

with identical means such that RB = RA + ε where ε is a white noise, and let sA and sB be the
savings out of Y0 corresponding to the return distributions of RA and RB respectively. Then

• If R′R (Y ) ≤ 0 and RR (Y ) > 1, then sA < sB

• If R′R (Y ) ≥ 0 and RR (Y ) < 1, then sA > sB

Definition 7.7.1 Define absolute prudence as the quantity

u′′′ (w)
PA (w) = −
u′′ (w)

and relative prudence as

u′′′ (w)
PR (w) = −w ·
u′′ (w)

Note that this depends directly on the curvature of u′ (u′′′ > 0 implies that u′ is convex) and does
not follow directly from risk aversion. It turns out that precautionary savings occur if P(w) > 0.
Moreover, we have the following result:

Theorem 7.7.2 Let RA and RB be two return distributions such that RA SSD RB , and let sA and
sB (respectively) be the savings out Y0 . Then

sA ≥ sB ⇔ cP(c) ≤ 2
sA < sB ⇔ cP(c) > 2

Finally, agents might save precautionarily also because they are concerned that they will face
borrowing constraints in some state in the future.

7.8 Mean-Variance Preferences

A different approach is to specify the utility function over the mean-variance space of investment
returns. This is a much simpler approach than the von Neumann-Morgenstern, so it is natural to ask
under which conditions the two approaches are equivalent. We discuss three cases here.

The first case in which mean variance and vN-M are equivalent is when agents have quadratic
utility:

u(c) = ac − bc2

Then the expected utility is

E [u(c)] = aE [c] − bE c2 = aE [c] − bE [c]2 − bVar (c)

 

Therefore, the expected utility is a function of the mean E [c] and variance Var (c) only.
7.8 Mean-Variance Preferences 73

This is also true in another case: suppose all lotteries have normally distributed outcomes (thus
the state space now is infinite). Then, because any linear combination of normally distributed
random variables is also normal, the distribution of any lottery (or linear combination thereof) will
be completely described by its first two moments, mean and variance. Therefore also expected
utility can be expressed as a function of these two numbers only.

The third cases has been covered in section 7.5, where we showed that if the risk (lottery) X is
zero-mean and small then

Var(X)
Y0 −Y CE ≈ RA (Y0 ) .
2
 74 Chapter 7. Risk Preferences and Expected Utility Theory

7.9 Exercises

1. Using the definition in the lecture notes,

a) Show that state-by-state dominance implies first order stochastic dominance.
b) Provide a simple example of discrete, finite-state space random variables (that is, a
table of payoffs, states and probabilities) that can be ranked by first order stochastic
dominance, but not by state-by-state dominance.

2. Consider the random variables X ∼ N(0, 1) and Y ∼ N(1, 1).

a) Suppose Y = X + 1. What is the state-by-state ordering of X and Y (if any)? What is
the first-order stochastic dominance ordering of X and Y (if any)?
b) Suppose X and Y are independent random variables. What is the state-by-state ordering
of X and Y (if any)? What is the first-order stochastic dominance ordering of X and Y
(if any)?
c) What conclusion can we draw about state-by-state and first order stochastic dominance
in relation to the probability distribution of the random variables?

3. Suppose an agent has an income of $10. He has the possibility to buy for $2 a lottery ticket
which pays the winner $19 (and zero otherwise). Suppose the agent has a von-Neumann
Morgenstern utility u(x) = ln(x) and believes that the probability of winning π is 31 .
a) Write the agent’s expected utility.
b) Should the agent buy the ticket?
c) Suppose the agent buys the ticket, but before a winner is selected he decides to sell it.
What is the minimum certain amount that the agent would accept in exchange for the
lottery ticket?

4. To resolve the [Link] Paradox, Bernoulli proposed to give less weight to large payouts
in the computation of the expected value of the game, to account for the fact that people
value money less and less the richer they become (though this certainly does not apply to
MFins). This property is shared by any increasing and concave function: can you guess why
Bernoulli (a pretty smart guy) proposed exactly u(x) = ln(x)? (Hint: consider the increase in
utility u given by a small increase in the payoff x.)

5. Suppose an agent is given the opportunity to enter, for free, a lottery that pays him $50,000
1
with probability 2 and $100,000 with probability 12 . Suppose the agent has CRRA utility:

c1−γ
u(c) =
1−γ

and γ = 7. Find the certainty equivalent of this lottery for the initial wealth Y equal to $0,
$100,000 and $300,000. What happens to the certainty equivalent when the initial wealth
increases? Explain this result.
7.9 Exercises 75

6. Consider a general von Neumann-Morgenstern utility function u(x) and a lottery which adds
an amount h to your personal wealth c in case of victory (which occurs with probability π)
and subtracts the same amount in case of loss:

c+h
π
c
1−π
c−h

Show that the following relation holds:

1 1
π(c, h) = + hRA (c) + o(h)
2 4

Similarly, consider a lottery which increases your personal wealth c by θ % in case of victory
(which occurs with probability π) and decreases your personal wealth by the same percentage
amount in case of loss:

c(1 + θ )
π
c
1−π
c(1 − θ )

Show that the following relation holds:

1 1
π(c, θ ) = + θ RR (c) + o(θ )
2 4

−νx
7. Use the CARA function U(x) = − e ν to solve the risk aversion and portfolio allocation
problem in Lecture slides 3, Slide 47 and show that it confirms Arrow’s theorem.

8. As in Lecture slides 3, Slide 51, show that a constant fraction of wealth is invested in risky
asset for an agent using CRRA utility function with r2 > r f > r1 . Show that it works also
with any distribution of r.

9. Consider the following one-period portfolio problem faced by an agent with vN-M utility
U(·):
max E[U(Y0 − s) + δU(sR̃)], s.t. Y0 ≥ s ≥ 0
s

where s is the amount he saves to consume at the final period (1). Observing the variables
appearing in the theorem proposed by Rothschild and Stiglitz (1971) (Lecture 3, Slide 66),
without directly using the theorem, please solve the portfolio problem above for RA and for
RB = RA + ε (i.e. RA s.s.d RB ) to show that:
a) When the agent has mean variance preferences U(x) = x − θ2 x2 , then sA > sB .
76 Chapter 7. Risk Preferences and Expected Utility Theory

b) When the agent has log-utility U(x) = ln(x), then sA = sB .

10. Assume that an investor with utility u(ct , ct+1 ) = u(ct ) + E[u(ct+1 )] can freely buy or sell
as much of the payoff xt+1 as he wishes, at a price pt . Assume that he has an exogenous
endowment of (et , et+1 ):
a) How much will he buy or sell?
b) What is the relation between the price pt and the payoff xt+1 ?
Notice that et+1 and xt+1 are random variables. For question i), further assume that et+1
and xt+1 follow binomial distributions, with probability π of being in the up state u (et+1 =
eu , xt+1 = xu ) and (1 − π) of being in the down state d (et+1 = ed , xt+1 = xd ). Obtain a
solution when the utility function is u(c) = ln(c). (Obtaining a simplified equation containing
the unknown amount to buy or sell would be fine, as long as you show the equation can be
solved for an explicit solution).

11. Assume CARA preferences U(x) = E[−e−cx ]. An agent values terminal wealth W and begins
with wealth W0 . He can invest in a risk-free asset with net return R f and in two risky assets
with independent net returns Ri ∼ N(ui , vi ), i = 1, 2. Suppose u1 > u2 > R f and 0 < v1 < v2 .
a) Write down the wealth W in terms of W0 , R f , R1 , R2 , and the fraction of wealth invested
in the two risky assets x1 and x2 .
b) Calculate U(W ).
c) Solve for the optimal x1∗ and x2∗ . Show that x1∗ > x2∗ > 0.
d) Explain why the agent invests a positive amount in asset 2 even though it has a lower
mean and higher variance than asset 1.

(A+Bx)1−1/B
12. Consider the following HARA u(x) = B−1 with A > 0 and B ∈ R. Using an idea that
positive affine transformation does not affect the order of preference expressed by this utility,
show that, after the right affine transformation, the limit of u(x) when B goes to zero is a
CARA utility function, i.e. an exponential concave utility function.
 8. Mean-Variance Analysis and CAPM

8.1 Preliminaries
In last chapter we introduced the representation of preferences over a consumption profile c =
(c0 , c1 , . . . , cS ) ∈ RS+1
+ . Suppose that, at birth, agent i is given an endowment e = (e0 , e1 , . . . , eS ) ∈
S+1
R+ and has a utility function U i : RS+1 i
+ → R. We say that U is:

• Quasiconcave, if the sets C = {c : U(c) ≥ v} are convex for each v ∈ R

• Concave, if for any c, c′ ∈ RS+1 and α ∈ [0, 1] we have U i (αc + (1 − α)c′ ) ≥ αU i (c) + (1 −
α)U i (c′ )

∂U i
Concavity implies quasiconcavity. If U i is also differentiable, a standard requirement is that ∂ cs >0
for each s = 0, 1, . . . S. We are now ready to formulate the portfolio consumption problem:

max U i (c0 , c1 , . . . cS )
c,h

subject to the constraints

0 ≤ c0 ≤ e0 − p · h

J
0 ≤ cs ≤ es + ∑ xsj h j , s = 1, . . . , S
j=1

If we define cT ≡ (c1 , . . . , cS ) ∈ RS and eT ≡ (e1 , . . . , eS ) ∈ RS the latter constraint can be rewritten

more compactly as follows:

0 ≤ cT ≤ eT + Xh

The Lagrangian for this problem is

L (c; λ , µ) = U i (c0 , c1 , . . . cS ) − λ [c0 − e0 + p · h] − µ [cT − eT − Xh]

where λ ∈ R, µ ∈ RS are the dual variables. Suppose c∗ solves the problem above, then there exist
λ ∈ R and µ ∈ RS such that:

∂U i ∗
(c ) = λ
∂ c0
∂U i ∗
(c ) = µs
∂ cs
λ p = µX
 78 Chapter 8. Mean-Variance Analysis and CAPM

The third FOC implies that

S
µs
pj = ∑ λ xsj
s=1

Plugging in the first and second FOCs we have

S
∂U i /∂ cs j
pj = ∑ i xs
s=1 ∂U /∂ c0

Assume that utility is time separable:

U i (c) = ui (c0 ) + δ ũi (cT )

and also that the utility over uncertain states ũi is a von Neumann-Morgenstern utility:

U i (c) = ui (c0 ) + δ E ui (cT )

 

Then we can rewrite our pricing relation as

S
∂ ui /∂ cs
pj = ∑ πs δ ∂ ui /∂ c0 xsj
s=1

We can read our findings as follows:

i /∂ cs
• δ ∂∂ uui /∂ c0
is the stochastic discount factor ms
i /∂ cs
• πs δ ∂∂ uui /∂ c0
= πs ms is the state price qs

Thus, we are back to our equivalent asset pricing formulas

pj = xj ·q = E m·xj
 

∂U i /∂ cs
Finally, note that the vector of marginal rates of substitutions ∂U i /∂ c0
is the state price vector (note
that this is positive by our assumption on U i ).

8.2 Simple CAPM with quadratic utility functions

In section 3.2 we discussed the State-Price Beta model, which states that for an asset j

E R j − R f = β j E [R∗ ] − R f
 

Cov(R∗ ,R j )
where β j = Var(R∗ ) and R∗ is the return of an asset whose payoff is equal to the discount factor
m∗ derived from the pricing kernel q∗ ∈ ⟨X⟩. This is a very general statement about returns, but it
does not help us identify what assets in reality have a return of R∗ .
8.2 Simple CAPM with quadratic utility functions 79

Suppose all agents have the same quadratic utility u (c0 , c1 ) = v (c0 ) − (c1 − α)2 . Recall that the
S
∂1 u
expected utility is E [U (c)] = ∑ πs u (c0 , c1,s ) and therefore m = ∂0 v where ∂i (·) ≡ ∂ (·)/∂ ci : in the
s=0
quadratic utility case we have
h i
∂1 u = −2 (c1 − α) · · · −2 (cS − α)

So the excess return can be written as

Cov m, R j


Cov ∂1 u, R j


 j f f
E R −R = − = −R ×
E [m] ∂0 v
j


2Cov c1 , R j


f Cov −2 (c1 − α) , R f
= −R × =R ×
∂0 v ∂0 v

This holds for any asset x j ∈ ⟨X⟩, and can be generalized to any portfolio h:

2Cov c1 , Rh
h i

h f f
E R −R = R ×
∂0 v

Now consider the market portfolio. The above equation must hold for it too, and therefore dividing
side by side we get

E Rh − R f
 
Cov c1 , Rh


=
E [Rmkt ] − R f Cov (c1 , Rmkt )

where Rmkt is the return of the market portfolio. If moreover agents are homogeneous and live in
an exchange economy, then we know that c1 corresponds to the aggregate endowment and this is
perfectly correlated with Rmkt :

E Rh − R f
 
Cov Rmkt , Rh


=
E [Rmkt ] − R f Var (Rmkt )

Cov(Rmkt ,Rh )
Define βh = Var(Rmkt )
we can write

h i  h i 
E Rh = R f + βh E Rmkt − R f

which we call the Security Market Line (SML). Note that in order for the above result to hold, R∗
must be a linear function of Rmkt of the type

a + bRmkt
R∗ =
a + bR f

with b < 0 since we know that the stochastic discount factor m is high (hence high R∗ ) in states
in which the economy is doing poorly, so in which the market portfolio xmkt is low (hence low
Rmkt ).
 80 Chapter 8. Mean-Variance Analysis and CAPM

8.3 The Traditional Derivation of CAPM

We define the mean return of the portfolio h as
" #
J J
µh ≡ E [rh ] = E ∑ w jr j = ∑ w j µ j = w′h µ
j=1 j=1

hj pj
where rh = Rh − 1 is the net return1 and w j = , h j is the amount of asset j ∈ 1, 2, . . . , J (as
∑Jj=1 h j p j
usual), so that the price of Rh is still 1. The variance of the portfolio is given by

σh2 ≡ Var (rh ) = w′h Vwh

p
and σh ≡ Var (rh ) is the standard deviation of portfolio h, where V is the covariance matrix of
the assets in portfolio h.

Definition 8.3.1 We say that portfolio A mean-variance dominates portfolio B if µA ≥ µB and

σA < σB , or µA > µB and σA ≤ σB .

Note that in the (σh , µh ) space, for a given value of µh and σh we can immediately find the subset
of portfolios which are not dominated by the given µh and σh :
Definition 8.3.2 — Efficient Frontier. For given µ and σ , the Efficient Frontier is the locus of
all non-dominated payoffs in the (σh , µh ) space.

It follows directly from the definition of efficient frontier that no rational investor with mean-
variance preferences would choose to hold a portfolio outside of the efficient frontier.

In the J = 2 case, for portfolio h we have w2 = 1−w1 , so the mean return is µh = w1 µ1 +(1 − w2 ) µ2
and the variance is σh2 = w21 σ12 + (1 − w1 )2 σ22 + 2w1 (1 − w1 ) σ1 σ2 ρ1,2 . Note that in the 2 assets
example, for ρ1,2 = 1 we get

±σh − σ2
w1 =
σ1 − σ2

So that plugging back in the expressions for µh we get

µ2 − µ1
µh = µ1 + (±σh − σ1 )
σ2 − σ1

which looks like Figure 8.1.

For ρ1,2 = −1 we get

±σh + σ2
w1 =
σ1 + σ2
1 Note that we can always write the excess return in gross or net terms: E R j − R f = E r j − r f .
   
8.3 The Traditional Derivation of CAPM 81

µ2

µh

µ1

σ
σ1 σh σ2

Figure 8.1: The efficient frontier: two perfectly correlated risky assets.

And therefore

σ2 σ1 µ2 − µ1
µh = µ1 + µ2 ± σh
σ1 + σ2 σ1 + σ2 σ1 + σ2

which looks like Figure 8.2.

µ2 −µ1
µ2 slope: σ1 +σ2

σ2 σ1
σ1 +σ2 µ1 + σ1 +σ2 µ2

µ1 slope: − σµ21 −µ
+σ2
1

σ
σ1 σ2

Figure 8.2: The efficient frontier: two perfectly negatively correlated risky assets.

While for ρ1,2 ∈ (−1, 1) we have Figure 8.3

 82 Chapter 8. Mean-Variance Analysis and CAPM

µ2

µ1

σ1 σ2 σ

Figure 8.3: The efficient frontier: two risky assets ρ = −0.5, 0, 0.5 (blue, black, orange).

8.3.1 Mean-Variance Frontier

The same concept generalizes to the J assets case: a frontier portfolio has minimum variance among
all feasible portfolios with the same expected portfolio return, so the problem is

1
max w′ Vw
w 2

subject to λ : w′ µ = µh
γ : w′ 1 = 1

where 1 = (1, . . . , 1)′ and µh ∈ R is a given (fixed) expected return. The idea is that we fix
an expected return µh and then look for the minimum variance possible given that expected
return.

The FOCs are:

∂L
= Vw − λ µ − γ1 = 0
∂w
∂L
= µh − w′ µ = 0
∂λ
∂L
= 1 − w′ 1 = 0
∂γ

where λ and γ are the Lagrange multipliers for the two constraints. Pre-multiplying the first FOC
by µ′ V−1 we get

µ′ w = λ µ′ V−1 µ + γ µ′ V−1 1 ≡ λ B + γA



8.3 The Traditional Derivation of CAPM 83

And since µ′ w = w′ µ = µh by the second FOC,

µh = λ B + γA

While pre-multiplying the first FOC by 1′ V−1 we get

1 = 1′ w = λ 1′ V−1 µ + γ 1′ V−1 1 ≡ λ A + γC

Solving for λ and γ we get

Cµh − A
λ=
D
B − Aµh
γ=
D

where D ≡ BC − A2 .

Therefore, pre-multiplying the first FOC by V−1 and plugging in λ and γ we have

Cµh − A −1 B − Aµh −1
w∗ = V µ+ V 1 (8.1)
D D
BV−1 1 − AV−1 µ CV−1 µ − AV−1 1
= + µh = g + hµh (8.2)
| D
{z } | D
{z }
≡g ≡h

Suppose now that for some portfolio with return r we set µh = E [r]. Similarly to the 2 assets
problem, the solution portfolio weights are linear in the (expected) portfolio returns:

w∗ = g + hE [r] (8.3)

Note that for E [r] = 0 and E [r] = 1 we get g and g + h respectively, which are therefore frontier
portfolios as well. We established the following facts:

Proposition 8.3.1 The entire set of frontier portfolios can be generated by g and g + h: any portfolio
in the frontier is a linear combination of these two portfolios.

Proof. Let p be an arbitrary frontier portfolio with expected return equal to E [r p ]. Consider
portfolio weights wg = 1 − E [r p ] and wg+h = E [r p ]. Then

(1 − E [r p ])g + E [r p ](g + h) = g + hE [r p ] = w p

Proposition 8.3.2 Any linear combination of frontier portfolios is also a frontier portfolio: the
portfolio frontier can be described as linear combinations of any two frontier portfolios (not just g
84 Chapter 8. Mean-Variance Analysis and CAPM

and g + h).

Proof. Exercise. ■

Substitution of the solution in 8.2 into the variance expression w′ Vw gives

A 2 1
 
2 C
σ (E [r]) = E [r] − + (8.4)
D C C

This is the equation of the minimum variance frontier.

Definition 8.3.3 The portfolio with the smallest possible variance is called the global minimum
variance portfolio or mvp.

We observe the following results

1. The expected return of the minimum variance portfolio is CA ;

2. the variance of the minimum variance portfolio is C1 ;
3. 8.4 is the equation of a parabola with vertex at C1 , CA in the expected return/variance space

and of a hyperbola in the expected return/standard deviation space. Figure 8.4 and 8.5
visualize the set of frontier portfolios in these spaces.

Figure 8.4: The set of frontier portfolios: mean-variance space.

Given two portfolios A and B in the frontier, we can see that points above B and below A correspond
to portfolios in which we short-sell assets. See Figure 8.6.

Based on the observations above, we have an equivalent definition for efficient frontier.
8.3 The Traditional Derivation of CAPM 85

Figure 8.5: The set of frontier portfolios: mean-standard deviation space

Figure 8.6: Short selling.

Definition 8.3.4 — Efficient Frontier. Efficient portfolios are those frontier portfolios for which
the expected return exceeds A/C, the expected return of the minimum variance portfolio.

Proposition 8.3.3 Any convex combination of frontier portfolios is also a frontier portfolio.
 86 Chapter 8. Mean-Variance Analysis and CAPM

8.3.2 Adding Risk-free Asset

How does all this change if we have a risk-free asset (i.e., an asset whose variance is zero)? We can
adapt the problem considered above as follows:

1
max w′ Vw
w 2

s.t. w′ µ + (1 − w′ 1)r f = µh

where r f is the net risk-free rate. 2 Note that in this problem weights automatically sum to one since
w′ 1 + (1 − w′ 1) = 1, so we don’t have the second constraint in the problem without the risk-free
asset. First order conditions yield

w = λ V−1 (µ − r f 1)

So premultiplying both sides by (µ − r f 1)′ yields

µh − r f
λ=
(µ − r f 1)′ V−1 (µ − r f 1)

since (µ − r f 1)′ w = w′ µ − w′ 1r f = µh − r f from the constraint in the problem above. Plugging

this back in the expression of w we finally get

V−1 (µ − r f 1) µh − r f
w= ′ −1 (µh − r f ) = V−1 (µ − r f 1)
(µ − r f 1) V (µ − r f 1) H2
q
where the scalar H = B − 2Ar f +Cr2f is the Sharpe ratio. We have two important results at this
point:

Result 1: For any two frontier portfolios p and q, we have E [rq ] − r f = βq,p (E [r p ] − r f ).

To see this, note that

E [r p ] − r f (E [rq ] − r f ) (E [r p ] − r f )
Cov (rq , r p ) = w′q Vw p = w′q (µ − r f 1) 2
=
H H2
2
(E[r p ]−r f )
and Var (r p ) = H2
, and dividing side by side yields the result. Importantly, this holds for
any two frontier portfolio p, thus in particular it also holds for the market portfolio.

Result 2: The frontier is linear in the (σ (r), E [r]) space.

2
(E[r p ]−r f )
This follows immediately from Var (r p ) = H2
by taking the square root and rearranging: we
get

E [r p ] = r f + Hσ (r p ) . (8.5)

2 That is, r f = R f − 1.
 8.3 The Traditional Derivation of CAPM 87

Therefore the efficient frontier with a risk-free asset is linear, and it is called the Capital Market
Line (CML).

E(r)
Mean Variance
Efficient Porfolio
20
L
CM
15

10

5
rf

0
0 10 20 30 40 50 60 σ

Figure 8.7: Capital market line.

8.3.3 Two Fund Separation

What is the result of individual optimization on the aggregate supply and demand of assets? That is,
what can we say about the equilibrium state given what we have seen so far? We approach this
problem in two steps: first we solve for the efficient frontier of the J risky assets, then we solve for
the tangent point with the agents’ indifference curves in the (σ (r), E [r]) space. The advantage of
this approach is that we have the same portfolio of J risky assets for different agents with differing
risk aversion, and this makes it easier to apply equilibrium arguments. Represented in the graph
below are the optimal portfolios of two investors with different degrees of risk aversion (the black
and red indifference curves).

E(r)

20
L
CM
15

10

5
rf

0
0 10 20 30 40 50 60 σ

Figure 8.8: Two fund separation.

In this setup, mean-variance preferences represented by a utility function U µ, σ 2 simply satisfy

2 = µ − ρ σ 2 . As we have already seen,
∂U ∂U


∂ µ > 0 and ∂ σ 2 < 0: a simple example is U µ, σ 2
 88 Chapter 8. Mean-Variance Analysis and CAPM

these preferences are equivalent to those of a von Neumann-Morgenstern quadratic utility: if

u (X) = a + bX + cX 2 then E [u (X)] = a + bµ + cσ 2 + cµ 2 = U µ, σ 2 . Again as already seen,

if asset returns are gaussian then also any portfolio is gaussian and therfore preferences can only
depend on the first two moments. Moreover if agents have CARA utility function we know that the
certainty equivalent for a lottery with mean µ and variance σ 2 is going to be µ − ρ2 σ 2 where ρ is
the absolute risk aversion of agents.

8.3.4 Equilibrium leads to CAPM

The theory examined so far only tells us about the demand of assets: prices are taken as given, and
the composition of the optimal (risky) portfolio is the same for all investors. Setting the aggregate
demand for assets equal to the supply of assets, that is, the market portfolio. Through CAPM we
can find assets’ equilibrium prices and agents’ risk premium. As we have seen, the market portfolio
is efficient (since it lies on the efficient frontier); moreover, all individual optimal portfolios are
located on the half line originating at the point (0, r f ). As introduced above, this half-line is the
CML and can now be rewritten as

E [rmkt ] − r f
E [rh ] = r f + σh (8.6)
σ (rmkt )

Note that the slope is the Sharpe ratio of the market portfolio.

Similarly, since

h E[r ]−r
f
σh H E [rh ] − r f Cov(rh , rmkt )
= = = = βh,mkt
σ (rmkt ) σ (rm kt) E [rmkt ] − r f Var(rmkt )

we can write

E [rh ] = r f + βh,mkt (E [rmkt ] − r f ) (8.7)

which defines the Security Market Line (SML)in the (β , E [r]) space:

E(r) ine
Market Porfolio
r ket L
Ma
urity
Sec
r2

rM

r1

rf

0 0.5 1 1.5 2 β

Figure 8.9: Security market line.

 8.4 Exercises 89

8.4 Exercises

1. Consider a two-period economy. There is an investor with CRRA utility and no discounting,
so that she maximizes expected utility

1−γ
" #
1−γ
c0 c1
+E ,
1−γ 1−γ

where c0 is consumption in period 0 and c1 is consumption in period 1. There are J different

assets. Each asset j has a stochastic gross return R j that is realized in period 1.
a) Write c0 and c1 in terms of initial wealth w, gross returns R j for j ∈ {1, . . . , J} and
the fractions of wealth invested in each asset j, which we denote α j for j ∈ {1, . . . , J}.
Note that it is here not the case that ∑ j α j = 1. What is the interpretation of w ∑ j α j ?
b) Using the previous notation write down the maximization problem that solves for the
optimal fraction of wealth α j invested in each asset. Derive the FOC for α j . Do the
α j ’s depend on wealth w? If yes/not, are you surprised by that?
Assume from here on that the economy has a total endowment e0 of the consumption good in
period 0, which cannot be stored and therefore has to be consumed in period 0. The payoffs
of all assets in the economy sum up to a random amount e1 in period 1, so that e1 is the total
amount of consumption goods in the economy in period 1. There are no other sources of
income in period 1.
Assume that the investor above is the representative investor of the economy, which just
means that she will consume the total endowment in period 0 and the total payoff of all assets
in period 1.
c) Write the FOC for α j from the previous part in terms of e0 and e1 . Use this equation to
derive the SDF m of the economy.
d) Assume that the endowment in period 0 is e0 = 10. There are two different states of the
world in period 1. In state 1, the total payoff of all assets is e1,1 = 10. In state 2, the
total payoff is e1,2 = 20. Both states have equal probability (50%). What is the risk-free
rate in the economy when γ = 1? What is the risk-free rate when γ = 2?
e) Calculate the slope of the capital market line in this economy for γ = 1 and γ = 2.
Interpret your finding.
f) Now assume that e1,1 = 5 in state 1 and e1,2 = 10 in state 2. The endowment in the
initial period remains the same (e0 = 10). How does the risk-free rate change for the
two cases γ = 1 and γ = 2? Why do you think that is?

2. Consider an economy in which there are 3 risky assets (A, B and C) and one riskless asset.
Asset A has an expected net return of 15% and the variance of its return is 0.20. Asset B has
an expected net return of 20% and a variance of 0.40. Asset C has an expected net return
of 55% and its variance is 0.60. The covariance between each pair of asset returns is zero.
There are 3000 shares of A available in the economy and the current price of A is 20. There
90 Chapter 8. Mean-Variance Analysis and CAPM

are 1500 shares of B available and its current price is 40. Finally, there are 10,000 shares of
C outstanding and the current price of C shares is 18. Does there exist a riskless rate of return
R f so that given the current prices of the risky asset, the Sharpe-Lintner pricing equation
E R j = R f + β j,m (E Rmkt − R f ) holds? If so, what is that rate? If not, why is it impossible
   

to find such a rate?

3. A manufacturing firm has just been founded and needs to build a factory. It has a choice
between 2 types of factory (factory 1 and factory 2) and the cost of building each is K1 = 120
and K2 = 50, respectively. Only one factory can be built at this time. The table below lists
the properties of the payoffs (X1 and X2 ) from the different projects, the gross market rate of
return (Rmkt ) and the gross risk-free rate (R f ). You may assume that all the assumptions of
the CAPM hold and that the management seeks to maximize the market value of the firm at
all times. To simplify calculations, the net risk-free rate of return is set to zero. You may also
assume that any investments the firm makes have a negligible effect on the market portfolio.

E Rmkt = 1.1
 
E [x1 ] = 300 E [x2 ] = 150
Var[x1 ] = 10, 000 Var[x2 ] = 2, 500 Var[Rmkt ] = 0.01
Corr[X1 , X2 ] = 0.4 Corr[X1 , Rmkt ] = 0.8 Corr[X2 , Rmkt ] = 0.2

a) In an meeting, an executive argues as follows: Since the expected (net) return on factory
2 is higher (200% vs. 150%), that one should be built. Another executive disagrees
and argues that naturally one should always build factory 1, since the expected profit
is larger than if factory 2 were built (180 vs. 100). Which (if any) of the two decision
procedures suggested will maximize the firm value? Why?
b) In the end, the executives ask you to decide which factory to build. What is your
decision? Compute the firm’s total Value V . If 100 shares are outstanding, what is the
stock price P? (Assume the CAPM holds)
c) Suppose factory 2 has been built (do not assume this was the correct answer to part
b) and a year has passed. It is time to expand, and the firm plans to build factory 1.
However, it has no cash to fund the investment and need to raise K1 = 120 in the capital
market. The firm decides to issue shares for a total value of 120. How many shares
should it issue and at what price?
d) In a shareholder meeting, there is a heated argument about whether or not to go ahead
with the building of the factory and the stock issue. An angry shareholder claims that
although expected returns are impressive, factory 1 is an extremely risky investment,
not least because most of its risk is not diversifiable. She argues that exposing the firm
to this risk will reduce the value of her shares. A concerned executive responds by
pointing out that since the values of the 2 factories are negatively correlated, the project
does not really increase the riskiness of the firm’s portfolio of assets that much and is
therefore a good investment. What, in your opinion, is missing from the analysis by the
shareholder and the executive. Are their arguments valid?
8.4 Exercises 91

4. Assume that the market return is in the mean variance frontier. Suppose that there are 3
equally likely states. One security is risk-free and has a gross return R f = 0.9. A second
security has a gross return vector of (0,1,1). The market portfolio has a payoff of (1,2,2).
a) Use the CAPM to price a security that pays off (0,0,1).
b) Find the pricing kernel.
c) Calculate the price of the same security (that pays off (0,0,1)) using the pricing kernel.
Bibliography

[1] Arrow, Kenneth J. (Kenneth Joseph), 1921-2017 Amsterdam : North-Holland ; New

York : American Elsevier, c1974, 1976 printing.

[2] Kerry E. Back, Asset Pricing and Portfolio Choice Theory, Oxford University Press,
2010.

[3] Markus K. Brunnermeier, Lecture Notes for the course FIN 501: Asset Pricing, 2013.

[4] John Y. Campbell, Andrew W. Lo, and A. Craig McKinay, The Econometrics of
Financial Markets, Princeton University Press, 1996.

[5] John Y. Campbell and Robert J. Shiller, Valuation Ratios and the Long-Run Stock
Market Outlook, The Journal of Portfolio Management, 1998.

[6] John H. Cochrane, Asset Pricing revised edition, Princeton University Press, 2002.

[7] Robert L. McDonald, Derivatives Markets, Pearson series in finance, 2009.

[8] Stephen Morris, Lecture Notes for the course ECO 317: The Economics of Uncer-
tainty, 2011.