Asset Allocation and Investment Strategies

Lecture 1

Andrea Buraschi
Road Map

1. Introduction
2. Some housekeeping and questions
3. Overview of the course
4. Portfolio theory
I Consumption based asset pricing
Andrea Buraschi: Introduction

Chair in Finance
PhD at University of Chicago
Specialize in Financial Economics (Asset Pricing, Fixed Income, Derivatives)
Previously taught at University of Chicago, Columbia and LBS
The Course

I Lecture notes
I Reading list
I “Asset Pricing” by John Cochrane, 2005, Princeton University Press.
I “Asset Management: A Systematic Approach to Factor Investing”,
by Andrew Ang
I “E¢ ciently Ine¢ cient: How Smart Money Invests and Market Prices
Are Determined”, by Lasse Heje Pedersen
I Several articles
I Case studies
Course Requirements

I Come to class, engage with the material, and ask questions.

I Assignments
I Problem sets + Case studies
I Final exam.
I Grading
I Problem sets plus Case Studies 40%
I Final exam (closed book) 60%
The Stochastic Discount Factor
An introduction

Question: Suppose you buy a stock today, t, which yields an stochastic payo¤
tomorrow Xt +1 (possibly, the payo¤ tomorrow is the stock price plus dividend,
Xt +1 = pt +1 + dt +1 ). You know the possible outcomes and the associated
probability distribution. How much are you willing to pay for the stock?

Answer:
I Method 1: Asset prices must be consistent with how much utility you
derive from consumption.
I Method 2: Asset prices must be consistent among themselves (there are
no arbitrage opportunities: "Black and Scholes").
The Stochastic Discount Factor
An introduction

Method 1: Asset prices depend on your preferences.

I Notice that the CAPM (E (Ri ) = Rf + βi [E (Rm ) Rf ]) is a speci…c
model that makes assumptions about preferences (quadratic).
I time preferences: generally, investors are impatient, valuing consumption
today more than consumption tomorrow; we’ll assume a discount rate δ to
specify the rate of time preferences.
I risk preferences: generally, investors prefer a consumption stream that is
steady over time and across states of nature; the concavity of the the
utility function characterizes the aversion to risk.
The Stochastic Discount Factor
An introduction

I The General Framework: the following is the general framework that can
help us understand most asset pricing models. The majority of them will
be special cases.
I The main idea is to think agents as interteporal optimizers. A simple 2
period case captures most of the intuition: preferences are represented by
a utility function de…ned over current and future levels of consumption

U (Ct , Ct +1 ) = u (Ct ) + δEt [u (Ct +1 )]

where Ct represents consumption at date t,and δ represents the

(subjective) discount factor.

Ct +1 is stochastic, as the investor does not know her wealth tomorrow.

The Stochastic Discount Factor
Investors optimization problem

The investor chooses how much, θ , of a stock that pays Xi ,t +1 to hold in order
to maximize her utility U (Ct , Ct +1 ), given her endowment stream Kt , Kt +1

max u (Ct ) + Et [δu (Ct +1 )] s.t.

θ

date t budget constraint

Ct = Kt pt θ
and date (t + 1) budget constraint

C t +1 = K t +1 + X t +1 θ
The Stochastic Discount Factor
Investors optimization problem

The investor continues to buy (or sell) the stock until

I the marginal loss in utility from acquiring another unit of the asset at time
t: pt u 0 (Ct )

equals

I the marginal gain from consuming the extra payo¤ at time (t + 1):
E t [ δ u 0 (C t +1 )X t +1 ]
The Stochastic Discount Factor
The basic pricing equation

The optimization problem yields the basic pricing equation

u 0 (C t +1 )
pt = E t δ X t +1
u 0 (C t )

We de…ne the stochastic discount factor mt +1 as the rate at which the investor
is willing to substitute consumption at time (t + 1) for consumption at time t

u 0 (C t +1 )
m t +1 δ
u 0 (C t )

Then the basic pricing equation can be expressed as

p t = E t [ m t +1 X t +1 ]
The Stochastic Discount Factor
The basic pricing equation

I We can use the basic pricing equation

p t = E t [ m t +1 X t +1 ]
to price any asset with a stochastic payo¤ Xt +1 (stocks, bonds, options,
etc), for any utility function that may capture investors preferences.
I This result is the core of the "Fundamental Theorem of Asset Pricing".
The price of any tradable asset can be represented as the (conditional)
expected value of the product of a stochastic discount factor mt and the
future payo¤ Xt .
I The process mt is linked to the change of measure that you studied in
other courses:
pt = EtP [mt +1 Xt +1 ]
Z Z
= mt +1 Xt +1 dPt = Xt +1 (mt +1 dPt )
Z
= Xt +1 (dQt ) = EtQ [Xt +1 ]

where mt +1 dPt = dQt , or

dQt
m t +1 =
dPt
The Stochastic Discount Factor
The risk-free rate

I There is more: the expected value of the stochastic discount factor gives
you enough information to infer the interest rate.
I The price of a zero coupon bond with face value 1 and maturity T is

p t = E t [ m t +T 1]

thus the term structure of bond prices is simply given by the sequence of
expected values of mt +T for di¤erent T
I If the risk-free rate is the rate you get if you invest $1 today and yo get
$1(1 + r ) next period, then:

1 = Et [mt +1 (1 + r )]

or
1
1+r =
E t (m t +1 )
The Stochastic Discount Factor
The risk-free rate

I A model is nothing else than a speci…cation of the stochastic properties of

mt
I Special case: consider that investors preferences are captured by the
following utility function
1 γ
C
u (C t ) = t
1 γ
Suppose there is no uncertainty. Then
γ
1 C t +1
1+r =
δ Ct

such that
1. (real) interest rates are high when people are impatient;
2. (real) interest rates are high when consumption growth is high;
3. (real) interest rates are more sensitive to consumption growth when
investors are more risk averse (i.e. γ is high).
Example
Forward Contracts

A forward rate agreement (FRA) is a contract in which at time t + 1 one

party agrees to pay the forward price (or rate) F , while the other party agrees
to pay according to the then prevailing market price (‡oating rate) pt +1
I The net payment made at t + 1 is:

p t +1 F

I There is no exchange of cash ‡ows at initiation.

Example
Forward Contracts

We want to determine F that equates the initial expected value of the

transaction to zero (otherwise one of the two parties would walk away from the
deal).
Et [mt +1 (pt +1 F )] = 0.
This implies that

F = Et (pt +1 ) + cov (mt +1 , pt +1 ) (1 + rf )

Example
Forward Contracts

The forward price F is the expected future spot price of the asset to be
transacted at time t + 1
E t (p t +1 )
plus a term that re‡ects how the discount factor co-varies with the underlying
spot price, or a risk premium

cov (mt +1 , pt +1 )

grossed up by the risk-free interest rate to re‡ect that the contract price is set
at time t, but the payo¤ is received at time t + 1.
I Notice that if future prices are uncorrelated with the SDF mt +1 , i.e.
cov (mt +1 , pt +1 ) = 0, then

F = E t (p t +1 )

) forward prices ontracts are unbiased expectations of future prices.

Example
European options

A European option is a claim that pays at time T

max(St +T H, 0)

where St +T is the stock price at time t + T and H is the strike price set at
time t. The option price should be

C t = E t [ m t +1 max (St +T H, 0)] .

Example
European options

Alternatively, Ct is given by

max Ct = Et [m (S t +T H )]
m

subject to

m > 0
St = Et (mST )
1 = Et (m (1 + rf ))

Too cumbersome? We’ll come back next time...

The Stochastic Discount Factor
Risk corrections

Since the price of a stock

p t = E t (m t +1 X t +1 )

then
pt = Et (mt +1 )Et (Xt +1 ) + cov [mt +1 , Xt +1 ]
or
E t (X t +1 ) cov [δu 0 (Ct +1 ), Xt +1 ]
pt = +
1+r u 0 (C t )

An asset’s price is lowered if its payo¤ covaries positively with consumption.

An asset’s price is raised if its payo¤ covaries negatively with consumption.
The Stochastic Discount Factor
Risk corrections

I An asset whose payo¤ covaries positively with consumption is one that

pays o¤ well when you are wealthy, pays o¤ badly when you are already
poor.
I Such an asset makes consumption stream more volatile.
I It will sell for a low price.
I An asset whose payo¤ covaries negatively with consumption helps smooth
consumption better.
I It will sell for a high price.
The Stochastic Discount Factor
Idiosyncratic risk

Suppose we want to price another asset with payo¤ X̄t +1 such that

X̄t +1 = Xt +1 + εt +1

where Et (εt +1 ) = 0, vart (εt +1 ) = σ2ε , and mt +1 and εt +1 are independent.

What is the di¤erence in the price of the payo¤ Xt +1 and in the payo¤ X̄t +1 ?

The price p̄t of the asset with payo¤ X̄t +1 is

p̄t = Et (mt +1 )Et (Xt +1 ) + cov [mt +1 , Xt +1 ] = pt

p̄t = pt

An investor is compensated only for holding systematic risk: idiosyncratic risk

does not generate a risk correction!
to compute the optimal portfolio

m t +1
Let us now use the same framework to derive the optimal portfolio choice rule.
I Suppose there are n assets available paying X1,t +1 , X2,t +1 , ..., Xn,t +1 at
date t + 1.
I The price of each asset is p1,t , p2,t , ..., pn,t .
Given pt , what is the optimal portfolio of these n assets? We need to specify
U 0 (C )
mt +1 = δ U 0 (Ct +)1 :
t

max U (Ct ) + δEt [U (Ct +1 )]

θi

subject to the following budget constraints:

n
Ct = Kt ∑ pi ,t θ i
i =1
n
C t +1 = K t +1 + ∑ Xii,t +1 θ i
i =1
Portfolios of assets

We …nd the optimal composition of the portfolio from the n FOC

U 0 (C t +1 )
pi ,t = Et δ X
U 0 (Ct ) i ,t +1
Example

I Suppose that each of the 3 possible states (ω 1 , ω 2 , ω 3 ) occurs with the

following probabilities π t (1) = 12 , π t (2) = π t (3) = 14 .
I There are three assets (D, E, F) in the economy, each returning a payo¤
that depends on the particular state realized at t + 1.

State ω i (t + 1 )
Asset ω1 ω2 ω2
D 6 8 4
E 13 9 8
F 1 1 1

I The price of each stock at time t is

ptD = 6 ptE = 10 ptF = 1

Example
Portfolio optimization

Which portfolio of stocks D, E , F would you choose to hold to maximize your

consumption plan for today t and tomorrow t + 1?
I Suppose that your utility function is U (C ) = ln C .
I Your only income tomorrow is the return of the portfolio (Kt +1 = 0)
I The time preference rate is 5%? (Note: the time preference rate is not
the same as the discount factor)
Example
Portfolio optimization

Maximize utility

max fln(Ct ) + δE [ln(Ct +1 )]g

θ
s.t. Ct = Kt θ D ptD θ E ptE θ F ptF
Ct +1 (i ) = θ D XtD+1 (i ) + θ E XtE+1 (i ) + θ F XtF+1 (i ) for i=1,2,3.

I At the moment we have 4 budget conditions: one at time t, and three at

time t + 1.
I We take pt and Xt +1 as given, and we solve for
fθ D , θ E , θ F g, Ct , fCt +1 (1), Ct +1 (2), Ct +1 (3)g: 7 unknowns.
I Since we need to solve for 7 unknown, we need 3 more conditions. These
come from the optimality conditions (…rst order conditions):

U 0 (C t +1 )
pt (i ) = E t δ X t +1 (i ) for i = 1, 2, 3.
U 0 (C t )

which we’ll write as δEt [U 0 (Ct +1 )Xt +1 (i )] = U 0 (Ct )pt (i ).

Example
Portfolio optimization

The three FOC (one for each asset) are:

1 6 1 8 1 4 6
δ + + =
2 C t +1 (1 ) 4 C t +1 (2 ) 4 C t +1 (3 ) Ct
1 13 1 9 1 8 10
δ + + =
2 C t +1 (1 ) 4 C t +1 (2 ) 4 C t +1 (3 ) Ct
1 1 1 1 1 1 1
δ + + =
2 C t +1 (1 ) 4 C t +1 (2 ) 4 C t +1 (3 ) Ct

plus the (3 + 1) budget constraints: in total 7 equations in 7 unknowns.

Example
Portfolio optimization

Now, we can solve for the weights of each asset

5
θD = Kt
246
10
θE = Kt
123
25
θF = Kt
123
The Stochastic Discount Factor
The data

I Most models are just special cases of the fundamental pricing equation
pt = Et (mt +1 Xi ,t +1 ).
I A “model” is equivalent to the speci…cation of the necessary structure to
determine the the stochastic discount factor mt +1
I We will show that both the CAPM and multi-factor models are just
di¤erent speci…cations of mt +1.
I In the previous example, m = δ C (Ct +t 1 ) . Notice that m is a stochastic
process.
The Stochastic Discount Factor
Asset returns

Consider a stock i for which the price satis…es

pt = Et (mt +1 Xi ,t +1 )

Then
Xi ,t +1
1 = Et m t +1 = Et (mt +1 (1 + ri ,t +1 )) .
pt
1
Since Et (mt +1 ) = 1 +r f

1
1= Et (1 + ri ,t +1 ) + cov [mt +1 , ri ,t +1 ]
1 + rf
or
Et (ri ,t +1 ) rf = (1 + rf ) cov [mt +1 , ri ,t +1 ] .
The Stochastic Discount Factor
CAPM

CAPM states that the expected return of asset j is higher as its beta is higher

E (rj ) = rf + β j [E (rm ) rf ]

where rj and rm denote one-period returns on asset j and the market, rf

denotes the risk free interest rate, and βj is given by

cov (rj , rm )
βj =
σ2m

where σ2m is the variance of the market portfolio return.

The Stochastic Discount Factor
Example 1: CAPM

The CAPM is mathematically identical to a speci…cation of the discount factor

that is linear in the market return rm

m = ā b̄rm

Exercise. Prove that indeed the CAPM is immediately obtained when the SDF
is linear in rm : to prove it you just need to show that there exists two set of
constants ā and b̄ that enforce the CAPM condition.
The Stochastic Discount Factor
Example 1: CAPM

Proof: Substitute m = ā b̄rm into the pricing equation:

Et (ri ,t +1 ) rf = (1 + rf ) cov [(ā b̄rm ), ri ,t +1 ]

then

Et (ri ,t +1 ) rf = (1 + rf ) cov [ b̄rm , ri ,t +1 ]

= (1 + rf ) b̄cov [rm , ri ,t +1 ]
cov [rm , ri ,t +1 ]
= (1 + rf ) b̄ σ2m
σ2m
= (1 + rf ) b̄ σ2m βi
E (r m ) r f
if b̄ = (1 +rf )σ2m
, then

(1 + rf ) b̄ σ2m = [E (rm ) rf ]
so that we recover the CAPM model!

Et (ri ,t +1 ) rf = βi [E (rm ) rf ]
h i
(Additional condition ā = 1 +1rf + b̄E (1 + rm ) )
The Stochastic Discount Factor
Example 2: APT

The APT Pricing Equation is:

E (rj ) = λ0 + λ1 βj ,1 + ... + λk βj ,k

where
I λi is the factor risk premia which tells you how much extra return you get
for each extra unit of risk your portfolio has (there exists one λ for each
factor in the economy, plus one extra λ0 )
I βj ,i s represent factor loadings.
The Stochastic Discount Factor
Example 2: APT

The APT is mathematically identical to a speci…cation of the discount factor

that is linear in the factors f
m = a + b0 f
where b and f are k 1 vectors.
Choosing a Portfolio of Risky and Risk-free Assets

Suppose we can invest both in a portfolio P consisting of

I a risk-free asset with return rf
I a risky asset i , with a return ri , an expected return E (ri ) and standard
deviation σi
Which fraction w of the portfolio we invest in asset i ?
Choosing a Portfolio of Risky and Risk-free Assets

The return and expected return on a portfolio with weight w on the risky
security and (1 w ) on the risk-free asset is:

rP = wri + (1 w )rf
rP = rf + w (ri rf )
| {z }
rie

E (rP ) = rf + wE (rie )

The risk (variance) of the combined portfolio is:

h i
σ2P = E (rP E (rP ))2

= w 2 σ2i
Which Portfolio?

The optimal portfolio is the solution to the following problem:

h γ 2i
max U (rP ) = max E (rP ) σ
w w 2 P
where
E (rP ) = rf + w (ri rf ) and σ2P = w 2 σ2i
Which Portfolio?

The optimal portfolio is given by

E (ri ) rf
w =
γσ2i
Which Portfolio?

Di¤erent levels of risk aversion lead to di¤erent choices.

Portfolios of Assets

In reality we invest in more than one asset.

Let E (R ) and Rf be the vector of returns of the risky and risk free asset. Let
w and wf the portfolio weights for the risky assets and risk free assets. They
have to add up to 1: thus, w 0 i + wf = 1. The expected return ERp,w of the
portfolio with weights w and wf is equal to

ERp,w = w 0 E (R ) + wf Rf = Rf + w 0 [E (R ) Rf ]

Thus, the excess return and variance of this portfolio is

ERp,w Rf = w 0 [E (R ) Rf ]
0
Var (Rp,w ) = w Σw

For simplicity, let use a star to denote an excess return, so that the excess
return of the portfolio is ERp,w = E (Rp,w ) Rf and the excess return of each
risky asset is ERn = E (Rn ) Rf for n = 1, 2...N.
Portfolios of Assets

The MVP is the solution of the linear-quadratic problem in which we minimize

the portfolio variance w 0 Σw for a given level of portfolio excess return µ0 :

Minw w 0 Σw s.t. ERp,w = µ0

Write the Lagrangian L (w , λ) of the previous problem: this is given by

L (w , λ) = w 0 Σw + λ(w 0 ER µ0 )

The …rst order conditions are:

∂L (w , λ )
= 2Σw + λER = 0
∂w
∂L (w , λ )
= (w 0 ER µ0 ) = 0
∂λ

Solve for (w , λ) by substitution. Start from the …rst equation and solve for w :

1 1
w = λΣ ER
2
Portfolios of Assets

Now, substitute in the second equation and solve for λ:

1
λER 0 Σ 1
ER µ0 = 0
2
h i 1
λ = 2 ER 0 Σ 1
ER µ0

Substitute back and we have the …nal result

Σ 1 ER
w = µ0
[ER 0 Σ 1 ER ]
Σ 1 [E (R ) Rf ]
= µ0
[E (R ) Rf ]0 Σ 1 [E (R ) Rf ]

This is the amount to be invested in the risky assets. How much should you
invest in the risk-free? What is left from your total wealth: the sum of all the
portfolio weights should be 1, thus

wf = 1 i 0w
Portfolios of Assets
What is the portfolio weight wT for the tangency portfolio? This is the portfolio
which is invested 100% in the risky assets, thus i 0 w = 1. This implies that
h i 1
µ ER 0 Σ 1 ER [i 0 Σ 1 ER ] = 1

Just solve for µ

ER 0 Σ 1 ER
µ=
[i 0 Σ 1 ER ]
Now substitute back
Σ 1 ER
w = µ
[ER 0 Σ 1 ER ]
ER 0 Σ 1 ER Σ 1 ER
wT =
[i 0 Σ 1 ER ] [ER 0 Σ 1 ER ]
Here is the tangency portfolio:
Σ 1 ER
wT =
[i 0 Σ 1 ER ]
Σ 1 [E (R ) R f ]
=
i 0 Σ 1 [E (R ) R f ]
Portfolios of Assets: xample

In practice, notice that the numerator Σ 1 [E (R ) Rf ] gives you the

proportions to be invested in each assets. The denominator is simply rescaling
the portfolio weights so that they sum up to 1.
0.10 0.40 0.07
Suppose for instance that Σ = and [E (R ) Rf ] = ,
0.40 0.15 0.05
then
1
1 0.10 0.02 0.12 1. 164 4
Σ [E (R ) Rf ] = =
0.02 0.15 0.05 0.178 08
The weights do not sum to 1, as 1.1644 + 0.17808 = 1. 342 5, thus

1 1. 164 4 0.867 34
wT = =
1. 342 5 0.178 08 0.132 65
Optimal Portfolio Choice
Exponential utility

Question: What is the connection between the SDF approach mt +1 to …nd

optimal portfolios and the mean-variance approach? Equivalently, w hich
preferences over consumption U (C ) justify the mean-variance approach, which
is written in terms of asset returns as
h γ 2i
max U (rP ) = max E (rP ) σ
w w 2 P

Answer: If preference over consumption U (C ) are exponential, then the

E (r i ) r f
optimal portfolio is of the form w = γσ 2 , i.e. the mean-variance solution.
i
Optimal Portfolio Choice
Exponential utility

Proof: Suppose the utility function of investor is given by

U (C ) = exp( γC ), where γ represents the coe¢ cient of absolute risk
aversion. Portfolio allocation problem:
I Investor maximize expected utility (consumption takes place only at t + 1,
for simplicity
max EU (Ct +1 )
C t +1
I The investor has an endowment Kt to allocate between a riskless asset,
that returns (1 + rf ) at date t + 1, and a risky asset with uncertain payo¤
X t +1
Investors consume at date t + 1 the return of their portfolio

C t +1 = θ t X t +1 + (K t θ t p t ) (1 + rf )
| {z } | {z }
Risky Payo¤ Risk Free

where θ t is the number of shares of the risky asset the investor acquires;
θ t pt is the dollar value of these shares.
I Let’s assume, morever, that

X t +1 N (µX , σ2X )
Optimal Portfolio Choice
Exponential utility

Portfolio allocation problem:

I Investor maximize expected utility (consumption takes place only at t + 1,
for simplicity
max EU (Ct +1 )
C t +1

I Investors consume at date t + 1 the return of their portfolio

C t +1 = θ t p t (1 + rt +1 ) + (K t θ t p t ) (1 + rf )

where θ t is the number of shares of the risky asset the investor acquires, or

C t +1 = K t [ w t rt +1 + (1 w t ) rf ] + K t

where wt is the fraction of the initial wealth invested in the risky asset
(wt Kt = θ t pt ).
Optimal Portfolio Choice
Exponential utility

I Let’s re-write the problem in terms of the fraction wt of the initial wealth
invested in the risky asset: wt Kt = θ t pt
I In this case, we have Ct +1 = Kt rP + Kt
I Investors consume at date t + 1 the return of their portfolio

C t +1 = K t [ w t rt +1 + (1 w t ) rf ] + K t

where rP is the return of the portfolio and rP N (E (rP ), σ P ).

Optimal Portfolio Choice
Exponential utility

Portfolio allocation problem re-written

I Investor choose wt to maximize expected utility

max E f exp [ γ (Kt rP + Kt )]g

wt

I Note that if a random variable V N ( µV , σ V )

1
E (exp V ) = exp µV + σ2V .
2

I Then...
Optimal Portfolio Choice
Exponential vs. Quadratic Utility

... Then if V = Kt rP + Kt ,

E [exp (γV )]
1 2
= exp (γ(µv γσ )
2 v

I Investors’preferences are monotonic in µV + 12 σ2V , just as in the

mean-variance case: Nice and easy!
Risk-return Trade-o¤
Mean-variance optimization

The case of one-period exponential utility is the same as when investors

I like high expected returns E (r )
I dislike high variance σ2r (that is, investors are risk averse)

Their utility or happiness from a pattern of returns rP is:

γ 2
U = E (rP ) σ
2 P
where γ is a measure the investor’s level of risk-aversion (the higher γ, the
higher an investor’s dislike of risk).
Indi¤erence Curves

I Investor preferences can be depicted as indi¤erence curves.

I Each curve represents di¤erent utility levels for …xed risk aversion
γ(= 1.6).
I Each curve traces out the combinations of E (rP ), σ2P yielding the same
level of utility U.
Indi¤erence Curves

I Each curve plots the same utility level for di¤erent risk aversion γ.
I Higher γ implies that for a given σ, investors require higher mean return
to achieve same level of utility.
Next Time

I For next time, read the article by Brandt.

I We will talk about the practice of optimal portfolio choice and strategic
asset management.