Asset-Liability and Derivative Models

Jonathan Ziveyi1

1 UNSW
Australia
Risk and Actuarial Studies, UNSW Business School
j.ziveyi@unsw.edu.au

Module 1 Topic Notes

1/76 Version 2018. Copyright UNSW School of Risk and Actuarial Studies
 Module 1 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
2/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
3/76
 Topic 1: Utility Theory

Learning Outcome
Describe and apply methods for
I decision-making under uncertainty, and
I quantitative risk measurement
by effectively combining techniques from financial economics
and actuarial science.

We will consider:
I Expected utility theory and its applications
I Economic implications of utility functions
I Measures of investment risk

3/76
 Individual Preferences: Utility

Suppose we wish to model an individual’s decision making /

preferences.
I A utility function u (·) can be used to represent preferences
over alternatives. For 2 (deterministic) outcomes x and y :
I An individual prefers x to y if

u(x) > u(y )

I An individual is indifferent between x and y if

u(x) = u(y )

I In applications to financial modelling, x and y are typically

dollar amounts (representing wealth), and u(·) is usually a
tractable mathematical function.

4/76
 Preferences Under Uncertainty - Gamble

I We define a gamble between two risky alternatives, A and

B as an event whose outcome is A with a probability α and
B with probability (1 − α), that is

A prob α
gamble ≡ G(A, B; α) =
B prob (1 − α)

I The expected value of the gamble is

E(G) = αA + (1 − α)B. (1)

I For example, a gamble that offers $150,000 for a win and

$50,000 for a lose with equal probability has an expected
value of $100,000.

5/76
 Preferences Under Uncertainty - Example

I In most instances in practice however we are faced with

uncertain outcomes.
I For example: Suppose an investor has current wealth of
$10, 000, and can choose between the following 2
investments:
I A risk-free investment earning 5% interest.
I a risky investment whose outcome is normally distributed
with mean 106% and standard deviation 20%.
Which should he choose?

6/76
 Decision Making Under Uncertainty: Expected Utility
Theorem

I Under certain assumptions, the Expected Utility Theorem

can provide guidance for decision making under
uncertainty.
Expected Utility Theorem
Consider two (possibly random) wealth outcomes W1 and W2 .
An individual with utility function u(·) prefers W1 over W2 if and
only if
E [u (W1 )] ≥ E [u (W2 )] .

I An individual is indifferent between the two random

wealths if the expected utilities are equal.

7/76
 Preferences Under Uncertainty - Example cont...

Suppose the investor’s preferences can be represented by

u(w) = e−0.00005w

and has initial wealth of $10, 000. Which of the following is

preferred?
I A risk-free investment earning 5% interest.
I a risky investment whose outcome is normally distributed
with mean 106% and standard deviation 20%.

8/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
9/76
 Assumptions behind the Expected Utility Theorem
Axioms
1. Complete/Comparable: For an entire set of uncertain alternatives, an
individual can say either A is preferred to B i.e. (A  B) or B  A or
A∼B
2. Transitive: If A  B, and B  C, then A  C
3. Independent: If A ∼ B, and consider some D. The person is indifferent
between a gamble that gives

A wp p and D wp 1-p
B wp p and D wp 1-p

4. Measurability: If A  B  C then there exist a unique α such that

B ∼ G(A, C; α).

5. Ranking: If A  B  E and A  D  E,

B ∼ G(A, E; α1 ) and D ∼ G(A, E; α2 ) then α1 ≥ α2 ⇒ B  D

6. Certainty Equivalent: Everything (gamble) has a price.

9/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
10/76
 Applications of Expected Utility

There are many potential applications of (expected) utility

(some requiring further assumptions and/or generalisations).
For example:
I deciding on the optimal asset allocation
I pricing future cashflows
I placing a value on insurance contracts
I finding Pareto optimal risk exchanges

10/76
 Example: Insurance Choice

I A decision maker has initial wealth 10 and utility function

u (w) = w 2 , w > 0.

She faces a random loss L with mean 5 and variance 11.

Determine the maximum amount the she will pay for
complete insurance against loss L. (Assume that
0 ≤ P ≤ 5)

11/76
 Commonly used Utility Functions

I There are a number of commonly used utility functions in

the literature, e.g.:
I Linear utility function: U(x) = a + bx, b > 0.
I Quadratic utility function: U(x) = x − b2 x 2 , b > 0.
I Exponential utility function: U(x) = −e−bx , b > 0.
B 1
I Power utility function: U(x) = B−1 x 1− B , x > 0, b > 0.
I The choice of an appropriate utility function depends on
I economic properties implied by the chosen function
I the suitability of the above for the decision maker
I tractability
I The choice of the mathematical form of U() should not be
arbitrary. We now investigate some desirable properties
that a utility function should possess by relating
mathematical properties of functions to economic
principles.

12/76
 Principle of Non-Satiation

Non-Satiation
Individuals prefer more wealth to less

Mathematically - this requires a utility function which is an

increasing function of wealth. In other words, u 0 (w) > 0.

13/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
14/76
 Types of Investors: Risk Aversion

Risk-averse investors
Preferences exhibit risk aversion when the expectation of a risk is preferred to the risk
itself, i.e. E (W )  W .

Risk-neutral investors
An investor exhibits risk neutrality if he is indifferent between the expected value of a
risk and the risk itself, i.e. W ∼ E (W ).

Risk-loving investors
An individual exhibits risk loving behaviour when he prefers a gamble or risk over the
certain amount equal to the expected value, i.e. W  E (W ).

14/76
 Types of Investors - Example
I Consider a log utility function, U(W ) = ln(W ). Assume that
the gamble is given by

$5 prob 0.8
G(5, 30; 0.8) =
$30 prob 0.2

I The actuarial value of the gamble is the expected outcome.

The expected wealth is

E(W ) = 0.8($5) + 0.2($30) = $10

I The utility of the expected wealth is

U[E(W )] = ln(10) = 2.3

I If an individual with a log utility function could receive $10

with certainty, it would provide him with 2.3 utiles.
15/76
 Types of Investors - Example
I The other possibility is the utility of the gamble, which is
equal to the expected utility of wealth provided by the
gamble
E[U(W )] = 0.8U($5) + 0.2U($30) = 1.97
I Because we receive more utility from the actuarial value of
the gamble obtained with certainty, which is 2.3, than from
taking the gamble itself, which is 1.97, we are risk averse.
I In general:
I If U[E(W )] > E[U(W )], then we have risk aversion and it
00
turns out that U < 0;
I If U[E(W )] = E[U(W )], then we have risk neutrality with
00
U = 0.
I If U[E(W )] < E[U(W )], then we have risk loving with
00
U > 0.
I Intuitively
I If our utility function is strictly concave, we will be risk
averse;
I If it is linear, we are risk neutral;
16/76 I If it is convex, we are risk lovers.
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
17/76
 Risk Premium and Certainty Equivalent Wealth
I Risk premium is the maximum amount by which a risk
averse investor is willing to give up in order to avoid the
uncertainty.
I Suppose that Mr. Marshy is faced with the gamble
presented in the previous example and has current level of
wealth of $10, and a log utility function. How much will he
pay to avoid the gamble?
I If he takes the gamble, his expected utility of the gamble is
1.97 utiles and the level of wealth that provides 1.97 utiles
is $7.17(= e1.97 ).
I On the other hand, Mr. Marshy receives an expected level
of wealth $10 if he accepts the average of the gamble.
I Therefore, to avoid a gamble given a log utility function, he
will be willing to pay up to

$2.83 = $10 − $7.17 = E[W ] − exp(E[U(W )]).

17/76
 Risk Premium and Certainty Equivalent Wealth cont...

I The certainty equivalent wealth c(W ) is defined as

U(c(W )) = E[U(W )] ⇔ c(W ) = U −1 (E[U(W )]).

I The risk premium π(W , gamble) is defined as the
difference between
I an individual’s expected wealth E(W ), given the gamble,
and
I his or her certainty equivalent wealth, that is U −1 (E[U(W )])
I That is

π(W , gamble) = E(W ) − U −1 (E[U(W )]).

18/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
19/76
 Risk Aversion and Utility Functions

I For relate risk aversion (or otherwise) for the mathematical

properties of u(w) we need Jensen’s inequality:

Jensen’s inequality
Let X be a random variable with density function fX (·) and let
00
g (·) be any concave function, eg. g (x) < 0 for all x. Then, we
have
E [g (X )] < g [E (X )] .

I Applied to expected utility, we see that a risk averse

individual possesses a utility function which satisfies
u 00 (w) < 0.

19/76
 Absolute Risk Aversion

I A third property of utility functions relates to how the

investor’s behaviour varies with changes in wealth.

Absolute Risk Aversion

00
u (w)
A (w) = − 0
u (w)

I An individual with decreasing absolute risk aversion will

increase his or her $ investment in risky assets as his/her
wealth increases.

20/76
 Example

I An individual has decreasing absolute risk aversion. He is

currently willing to pay up to $100 to insure against the risk
of damage to his home.
I Suppose his wealth increases. Would the maximum
premium he is willing to pay be greater/equal/less then
$100?

21/76
 Relative Risk Aversion

Relative Risk Aversion

00
u (w) d 0
R (w) = wA (w) = −w = −w [log u (w)]
u 0 (w) dw

I Relative risk aversion refers to the change in the proportion

of wealth invested in risky assets as wealth changes.

22/76
 Example

I Consider an individual with a power utility function (with

γ < 0):
wγ
u(w) =
γ
I How will his investment in risky assets change as his
wealth increases?

23/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
24/76
 Topic Summary

I In this topic, we explored methods and techniques which

can help us to quantify (and hence make decisions) in the
face of uncertainty.
I These included, in particular
I Expected Utility Theory
I how to apply the Theory
I its assumptions and implications
I the related idea of risk measurement was also considered.
I These ideas will underpin many of the more complicated
situations that we will investigate in the coming weeks, and
hence form an important foundation for the rest of the
course.

24/76
 Topic 2: Investment Risk Measures

I Consider an investment with random outcome X .

I Generally we are interested in quantifying the riskiness of
the position.
Risk Measure
A risk measure ϑ(·) is a mapping from the random variable that
generally represents the risk to the set of real numbers:

ϑ : X → R.
I A risk measure is supposed to provide a value for the
degree of risk or uncertainty associated with the random
variable.

25/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
26/76
 Examples of Different Risk Measures

Variance:
I Variance is a measure of how variable the outcome is relative to
the mean.
h i
2
Var (X ) = E (X − µ) = E X 2 − µ2


Z ∞
2
= (x − µ) fX (x) dx
−∞

Semi-Variance of Return:
I It is sometimes called the downside semi-variance. It accounts
only for the variation below the mean.
Z µ
2
semi-var (X ) = (x − µ) fX (x) dx.
−∞

26/76
 Examples of Different Risk Measures cont...
Expected Shortfall:
I It gives the shortfall below a certain level, say l, called the
benchmark. It is useful for monitoring a fund’s exposure to
risk.
Z l
ES = (l − x) fX (x) dx
−∞

Skewness:
I It measures the extent to which a probability distribution is
asymmetric about its mean.
h i
S (X ) = E (X − µ)3
Z ∞
= (x − µ)3 fX (x) dx
−∞

27/76
 Examples of Different Risk Measures cont...
Kurtosis:
I It measures the “peakedness” or “pointedness” of a
distribution.
h i
K (X ) = E (X − µ)4
Z ∞
= (x − µ)4 fX (x) dx
−∞

Shortfall Measures:
I It provides a measure of the expected underperformance
relative to a given benchmark.
Z L
g (L − x) fX (x) dx
−∞

where g (·) is some function.

28/76
 Examples of Different Risk Measures cont...

Special cases for this function include:

1. If g (L − x) = 1, then we have the so-called shortfall
probability
Z L
p (L) = fX (x) dx = Pr (X ≤ L) .
−∞

2. If g (L − x) = (L − x), then we have the so-called expected

shortfall as defined above.
3. If g (L − x) = (L − x)2 , then we have the so-called shortfall
variance Z L
(L − x)2 fX (x) dx.
−∞

29/76
 An Important Risk Measure: Value-at-Risk

I The Value-at-risk (at level α) of a position is the answer to the

following question: What is the maximum loss that can be
incurred from holding a security or a portfolio of securities over a
given time period so that there only is a low probability 1 − αthat
the actual loss will be larger?
I It describes the quantile of the projected portfolio loss
distribution.
I It has been/is widely used in the financial markets as a risk
measure, and in particular as a determinant of the economic
capital banks need to hold.

30/76
 Example

I Suppose a random loss Y is normally distributed with

mean µ and standard deviation σ. Find the Value at risk at
97.5% confidence level of this position.

31/76
 Relationship between Risk Measures and Utility
Functions
I Suppose an investor has a quadratic utility function of the
form
u (w) = w + dw 2 ,
initial wealth w0 and intends to invest with return X with
mean E (X ) = µ and variance var (X ) = σ 2 . His expected
utility then is
h i
2
E [u (w0 + X )] = E (w0 + X ) + d (w0 + X )

= E (w0 + X ) + d w02 + 2w0 X + X 2




dw02 + w0
+ (1 + 2dw0 ) E (X )


=
+ dE X 2
which clearly is a combination of the first two moments of
the distribution of the rate of return. Thus, in this case, the
32/76 variance of return is a suitable measure.
 Topic 3: The Mean-Variance Portfolio Theory

Learning Outcome
Apply mean-variance criteria to determine the optimal asset al-
location for long term investors, including insurers and superan-
nuation funds.

We will consider:
I Why Mean-Variance?
I Mean-Variance with 2 risky assets.
I The general Markowitz problem.
I Inclusion of a riskless asset.

33/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
34/76
 Mean Variance Portfolio Theory

I Idea - find the ‘best’ asset portfolio in terms of

mean-variance tradeoff.
I Sometimes also referred to as Modern Portfolio Theory
(MPT)
I Developed by Markowitz (1952) - hence sometimes
referred to as the Markowitz mean-variance model
I We will:
1. Look at some motivations for mean-variance analysis
2. Start by looking at a simple 2 asset case
3. generalize to n assets

34/76
 Why Mean - Variance?
I Consider an individual with utility function u (·). We can use a
Taylor series expansion around the expected end of period
wealth W :
u (W ) = u (E [W ]) + u 0 (E [W ]) (W − E [W ])
1 2
+ u 00 (E [W ]) (W − E [W ])
2
∞
X 1 (n) n
+ u (E [W ]) (W − E [W ])
n!
n=3
2
which indicates a preference for E [W ] and dislike for σW [recall
the properties for u (·)], everything else being equal.
I Taking expectations yields
1 00 2
E [u (W )] = u (E [W ]) + u (E [W ]) σW + E[R3 ]
2
where
"∞ #
X 1 n
E[R3 ] = E u (n) (E [W ]) (W − E [W ])
n!
n=3
35/76
 Why Mean - Variance cont...

I Hence, expected utility depends on just the mean and

variance if either
1. u n (E[W ]) = 0 for n > 2. This holds if utility function is
quadratic, Or
2. The distribution of returns is normal since then all moments
depend on the mean and variance.

36/76
 Discussion

I Based on the previous derivations, suggest two situations

where the investor’s preference for E [W ] and dislike for
2 are actually exact.
σW

37/76
 Example
I Consider a portfolio P of two securities A and B. Suppose
we have $1,000 to invest and put $400 in A and $600 in B.
If E(rA ) = 10% and E(rB ) = 6%, what is the expected
return of our portfolio?
I The expected profit is

$400 × 0.1 + $600 × 0.06 = $76

I Then the expected return is

   
76 400 600
= × 0.1 + × 0.06 = 7.6%.
1, 000 1, 000 1, 000
I This can be expressed as

rP = wA rA + wB rB

where wA = 400/1, 000 = 0.4 and wB = 600/1, 000 = 0.6

are the fractions of money invested in A and B respectively.
38/76
 Example cont...
I The fractions are called portfolio weights. Hence

µP = E[rP ] = wA E[rA ] + wB E[rB ]

= wA µA + wB µB
= 0.4 × 0.1 + 0.6 × 0.06 = 7.6%.
I The variance of the portfolio can also be represented as

σP2 = wA2 σA2 + wB2 σB2 + 2wA wB σAB

I The variance of the portfolio is a weighted average of the
variances of the individual assets plus two times the
product of the portfolio weights times the covariance
between the assets.
I If the portfolio weights are both positive, then a positive
covariance will tend to increase the portfolio variance,
because both returns tend to move in the same direction,
and a negative covariance will tend to reduce the portfolio
39/76
variance.
 The Efficient Frontier

I The efficient frontier represents the efficient portfolios,

given the set of securities used to construct the portfolios.
Efficient portfolios are those with the highest expected
return for a given level of risk.
I Inefficient portfolios are portfolios such that there is at
least one other feasible portfolio that has the same risk
(σP ) but a higher expected return.
I The inefficient portfolios are the feasible portfolios that lie
below the global minimum variance portfolio.
I The efficient portfolios are those portfolios that lie above
the global minimum variance portfolio.

40/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
41/76
 Globally minimum variance portfolio - Two Risky
Assets
I The global minimum variance portfolio (MVP) is the
portfolio with lowest level of risk possible, regardless of
portfolio expected return, given the population of securities
used to construct the portfolio.
I It separates the minimum variance set into the efficient and
inefficient sets.
I For a two asset portfolio, the weights of the global
minimum variance portfolio are determined as follows:
I Solve the constrained optimization problem

min σP2 = wA2 σA2 + wB2 σB2 + 2wA wB σAB ,

wA ,wB

s.t.

wA + wB = 1.

41/76
 Global Minimum Variance Portfolio cont...
I Substituting wB = 1 − wA into the formula for σP2 reduces
the problem to

min σP2 = wA2 σA2 + (1 − wA )2 σB2 + 2wA (1 − wA )σAB ,

wA

I The first order condition for a minimum are

dσP2
= 2wAMVP σA2 − 2(1 − wAMVP )σB2 + 2σAB (1 − 2wAMVP ) = 0
dwA
I Hence the weight of asset A and B in the MVP can be
determined by
σB2 − σAB
wAMVP =
σA2 + σB2 − 2σAB
wBMVP = 1 − wAMVP
I Calculate the weights, expected return and the variance of
the MVP for the data provided in the previous slides.
42/76
 Special Cases of Two Asset Portfolios
I Correlation of returns measures the strength of the
relationship between the returns of two assets:
σAB
ρAB = , ⇒ σAB = ρAB σA σB .
σA σB
I The shape of the investment possibilities set is very
sensitive to the correlation between assets A and B.
I The portfolio standard deviation is less than the weighted
average of the component security standard deviations,
unless the two securities are perfectly positively correlated.
I Portfolios with less than perfectly correlated securities
always offer better risk-return opportunities than the
individual component securities on their own.
I The lower the correlation between the assets the greater
the gain to diversification.
43/76
 Special Cases of Two Asset Portfolios – Perfect
Positive Correlation

I There is no gain from diversification in this case (unless

short selling is possible).

σP2 = wA2 σA2 + (1 − wA )2 σB2 + 2wA (1 − wA )σAB

= (wA σA + (1 − wA )σB )2

I This implies that

σP = (wA σA + (1 − wA )σB )

44/76
 Special Cases of Two Asset Portfolios – Perfect
Negative Correlation
I Can create a portfolio without risk (no short-selling)

σP2 = wA2 σA2 + (1 − wA )2 σB2 + 2wA (1 − wA )σAB

= (wA σA − (1 − wA )σB )2

I Implying that

σP = ± [wA σA − (1 − wA )σB ]

I The standard deviation can be interpreted as follows:

σB
σP = [wA σA − (1 − wA )σB ] , if wA ≥ ,
σA + σB
σB
σP = − [wA σA − (1 − wA )σB ] , if wA < .
σA + σB
I Homework: Do the zero correlation case.
45/76
 Three Risky Assets

I Suppose an investor can choose from any combination of

three risky assets 1, 2, and 3 with returns e
r1 , e
r2 , and e
r3 ,
respectively. Let the respective means, variances, and
covariances be given by




E er1 = z1 , E er2 = z2 , and E er3 = z3 ,

r1 = σ12 , Var e
r2 = σ22 , and Var e
r3 = σ32 ,




Var e
and


Cov e
r1 , e
r2 = σ12 , Cov e
r1 , e
r3 = σ13 ,
and

Cov e
r2 , e
r3 = σ23 .

46/76
 Three Risky Assets - Matrix Notation

I When dealing with 3 or more assets, techniques from

linear algebra can simplify the calculations/notations
significantly.
I Define the mean vector as
 
z1
z =  z2 
z3

and variance-covariance matrix

 2 
σ1 σ12 σ13
Σ =  σ12 σ22 σ23 
σ13 σ23 σ32

47/76
 Three Risky Assets cont...
 
w1
I If the vector w =  w2  gives the portfolio weights, then
w3
the portfolio mean is
 
X 3 z 1
wk zk = w1 w2 w3  z2  = w0 z


µP =
k =1 z3

and portfolio variance is

3
σP2 wi2 σi2 +
P PP
= 2wi wj σij
i=1 i<j,i=1,2,3
σ12 σ12 σ13
  
w1
 σ12 σ 2 σ23   w2 


= w1 w2 w3 2
σ13 σ23 σ32 w3
= w0 Σw
48/76
 Discussion Question

I Suppose that the 3 assets have different means 1 . By

using the results from the 2 assets case, graphically derive
the general shape of the feasible set faced the investor.

1
49/76 and that the variance-covariance matrix is full rank
 Minimum Variance Set, and Efficient Frontier
I A portfolio is said to be a minimum variance portfolio if it
has the smallest variance among portfolios that have the
same expected rate of return.
I A portfolio is said to be an efficient portfolio if it has the
maximum expected return among portfolios that have the
same variance.
I Intuitively, we can graphically deduce the general shape of
the above sets. (Question - what do they look like?)

50/76
 Finding Minimum Variance Portfolios - Three Risky
Assets

I The optimization (minimization) problem is

1
min w0 Σw
w1 ,w2 ,w3 2

subject to
w0 1 = 1,
and
w0 z = µ
 
1
where µ is known and fixed, and 1 =  1  is a vector of
1
ones.

51/76
 Example

I Suppose the market consists of three stocks whose vector

of expected returns and the covariance structure of their
returns are given respectively
   
0.05 0.25 0.15 0.17
µ =  0.1  , Σ =  0.15 0.21 0.09 
0.15 0.17 0.09 0.28

I Find the minimum variance portfolio P having µP = 0.1

I Solution Method 1
I Let w3 = 1 − w1 − w2 such that

σP2 = w0 Σw
= 0.19w12 + 0.31w22 − 0.22w1 − 0.38w2 + 0.34w1 w2 + 0.28

52/76
 Example cont...

I The Markowitz Problem:

min σP2
w1 ,w2

subject to the constraint which takes the form

0.05w1 + 0.1w2 + 0.15(1 − w1 − w2 ) = 0.1

I We use the Lagrange multiplier techniques by defining

1 2
L(w1 , w2 , λ) = σ + λ[0.1 − (0.05w1 + 0.1w2 + 0.15(1 − w1 − w2 ))]
2 P

I NOTE that including the 1/2 before the σP2 is optional.

I Computing 1st derivatives with respect to w1 , w2 , and λ we
obtain the optimal weight that solves the equations in the next
slide.

53/76
 Example cont...

I The first order conditions become

0.38w1∗ + 0.34w2∗ − 0.22 + 0.10λ = 0

0.34w1∗ + 0.62w2∗ − 0.38 + 0.05λ = 0
0.10w1∗ + 0.05w2∗ − 0.05 = 0

I Solving the system yields

w1∗ = 0.24, w2∗ = 0.52, w3∗ = 0.24, λ = −0.48

I The corresponding portfolio expected return and variance

are

µP = 0.1, and σP2 = 0.1668.

54/76
 Example - Solution Method 2

I We need to solve the following minimization problem under

two constraints

min σP2
w1 ,w2 ,w3

subject to the constraints

0.05w1 + 0.1w2 + 0.15w3 = 0.1

w1 + w2 + w3 = 1

I First define the Lagrangian:

1 2
L[w1 , w2 , w3 , λ, γ] = σ + λ[1 − (w1 + w2 + w3 )]
2 P
+ γ[0.1 − (0.05w1 + 0.1w2 + 0.15w3 )]

55/76
 Example - Solution Method 2 cont...
I Then the first order conditions are

0.25w1 + 0.15w2 + 0.17w3 = λ + 0.05γ (2)

0.15w1 + 0.21w2 + 0.09w3 = λ + 0.1γ (3)
0.17w1 + 0.09w2 + 0.28w3 = λ + 0.15γ (4)
0.05w1 + 0.1w2 + 0.15w3 = 0.1 (5)
w1 + w2 + w3 = 1. (6)

I In matrix form, (2) - (4) can be written as

Σw = λ1 + γz (7)

and conditions (5) and (6) can be written as

w0 z = 0.1 and w0 1 = 1 (8)

56/76
 Finding Minimum Variance Portfolios - n Risky Assets

I The optimization problem is:

1
min w0 Σw
w 2

subject to
w0 1 = 1,
and
w0 z = µ.

57/76
 Solving the Minimization Problem

I Set up the Lagrangian:

1 0
w Σw + λ 1 − 10 w + γ µ − z0 w



L=
2
take derivatives and set to zero:
∂L
= Σw − λ1 − γz = 0
∂w
∂L
= 1 − 10 w = 0
∂λ
∂L
= µ − z0 w = 0
∂γ
I We now solve the n + 2 equations for the n + 2 unknowns.

58/76
 Solving the Minimization Problem cont...

I For simplicity of notation, define the constants

A = 10 Σ−1 1
B = 10 Σ−1 z = z0 Σ−1 1
C = z0 Σ−1 z
∆ = AC − B 2

I From the equation for ∂L

∂w we have

Σw − λ1 − γz = 0

which on rearrangement gives

w = λΣ−1 1 + γΣ−1 z (9)

59/76
 Solving for λ, γ
We can apply the original constraints to solve for the constants:
I Using the budget constraint: Multiply both sides of (9) by 10
:
λ10 Σ−1 1 + γ10 Σ−1 z = 10 w = 1

I Using the mean requirement: Multiply both sides of (1) by

z0 :
λz0 Σ−1 1 + γz0 Σ−1 z = z0 w = µ

This gives two equation and two unknowns. On solving

(exercise) we get
C − µB
λ =
∆
µA − B
γ =
∆
60/76 which subsequently gives a full solution for w.
 Exercise: The Minimum Variance Frontier

For a portfolio on the minimum variance set, show that the

variance σp2 is a quadratic function of µ.

61/76
 Implementation Example - Excel

62/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
63/76
 Global Minimum Variance Portfolio

I What is the portfolio that has the lowest variance?

I We want
∂σ 2  2Aµg − 2B
µ=µg = =0
∂µ ∆
I Hence
B
µg =
A
and on substitution, we have (note: λg , γg represents the
values of λ, γ corresponding to µg ):

wg = λg Σ−1 1 + γg Σ−1 z
1 −1
= Σ 1
A

63/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
64/76
 ‘Spanning’ the minimum variance frontier
I We know that any minimum variance portfolio satisfies
w = λΣ−1 1 + γΣ−1 z (10)
I Now we know that
Σ−1 1
wg =
A
and suppose we consider another portfolio, with
Σ−1 z
wd =
B
then Equation (10) becomes
w = λAwg + γBwd
but we also know that
λA + γB = 1
hence we see that wg and wd can be combined to form
64/76
any other minimum variance portfolio!
 Two Fund Theorem

I We see that wg and wd can be used to ‘span’ the whole

minimum variance frontier.
I Indeed, we can also use any other two minimum variance
portfolio to ’span’ the frontier. To see this - consider, for any
constants a, b,

wa = (1 − a) wg + awd
wb = (1 − b) wg + bwd

and observe that a portfolio of these two can be combined

to form any other minimum variance portfolio.

65/76
 A Caveat: Non-Negative Constraints

I So far the signs of the weights are not restricted (beyond

summing to 1).
I In practice this could create issues as some weights could
be negative (and potentially highly negative).
I A negative weight implies that we need to short sell the
asset. (possible, but some practical issues).
I It is possible to include non-negativity constraints:
I The optimization becomes a ‘quadratic program’ and will
often need to be solved numerically.
I In practice, it is often observed that many assets gain a 0
weight if such constraints are in force.

66/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
67/76
 One Fund Theorem - Inclusion of a risk-free asset

Consider for simplicity the case with only 1 risky asset, but
assume now that we have a risk free asset (with known return
rf ). The resulting portfolio (with w invested in the risky asset,
and 1 − w in the risk-free):
I has mean
(1 − w)rf + wz
I has standard deviation
wσ
Hence we see that the feasible set is a straight line.

67/76
 Discussion Question

I Suppose we have n risky assets, as well as a risk free.

Use a graphical argument to deduce what the efficient
frontier will look like.

68/76
 One Fund Theorem

I From previous discussions we see that the efficient set is a

straight line linking the risk-free rate (with standard
deviation 0) with (σt , µt ) (representing the characteristics of
the ‘tangency’ portfolio’.
I In particular, all efficient portfolios are combinations of the
tangency portfolio and the risk free asset!
This is the ‘One Fund Theorem’.

69/76
 Discussion: Finding a Minimum Variance Portfolio

I The optimization problem when there is a risky asset can

be represented by
1
min w0 Σw
w 2

subject to
(z − rf 1)0 w = µ − rf
where µ is the required mean return.
I Question: The constraint appears different at first glance to
that considered in the risky asset only case. Explain the
differences and/or similarities.

70/76
 Solution - Worked Example

I Form the Lagrangian:

1 0
w Σw + γ µ − rf − (z − rf 1)0 w


L=
2
take derivatives and set to zero
∂L
= Σw − γ (z − rf 1) = 0
∂w
∂L
= µ − rf − (z − rf 1)0 w = 0
∂γ

71/76
 Solution cont...

I Implying that the solution is

w = γΣ−1 (z − rf 1)

And we can solve for γ. From the constraint:

µ − rf = (z − rf 1)0 w
= γ (z − rf 1)0 Σ−1 (z − rf 1)
 
= γ z0 Σ−1 z − 2rf 10 Σ−1 z + rf2 10 Σ−1 1
 
= γ C − 2rf B + rf2 A

Hence
µ − rf
γ=
C − 2rf B + rf2 A

72/76
 Solution cont...

I Upon substitution, the minimum variance set is

σ 2 = w0 Σw
0
= γ 2 (z − rf 1) Σ−1 ΣΣ−1 (z − rf 1)
(µ − rf )2
=
C − 2rf B + rf2 A

which is a line on µ − σ space, and in particular

 1
2
µ = rf ± C − 2rf B + rf2 A σ

73/76
 Solving for the Tangency portfolio
I The tangency portfolio by definition consists only of risky
assets. Hence must have

1 = 10 wt
I This implies that we want

1 = 10 wt
= 10 γt Σ−1 (z − rf 1)
 
= γt 10 Σ−1 z − rf 10 Σ−1 1
1
γt =
B − Arf

and so
1
wt = Σ−1 (z − rf 1)
B − Arf
74/76
 Implementation Example - Excel - continued

I Finding the tangency portfolio

75/76
 Plan
Topic 1: Utility Theory
Utility Theory
The Utility Axioms
Applications of Expected Utility
Types of Investors
Risk Premium and Certainty Equivalent Wealth
Risk Aversion and Utility Functions
Topic Summary
Topic 2: Investment Risk Measures
Examples of Different Risk Measures
Topic 3: The Mean-Variance Portfolio Theory
Motivation
Deriving the Optimal Portfolio
Global Minimum Variance Portfolio
Two Fund Theorem
One Fund Theorem
Topic Summary
76/76
 Topic Summary

I In this topic, we investigated techniques to determine the

optimal asset allocation for an investment fund using a
mean-variance frontier.
I In particular, we discussed
I The motivations behind a mean-variance approach
I The mathematical derivations of the optimal asset
allocation in situations with and without a riskless asset
I Associated results that can be used to numerically compute
the associated portfolios
I These techniques and findings will allow us to compute the
optimal portfolios, and furthermore to extend these results
to more complicated situations (e.g. when liabilities are
included!)

76/76