Frontier Riskless CARA

Efficient Frontier

Wang Wei Mun

Lee Kong Chian School of Business

Singapore Management University

August 17, 2022

Wang Wei Mun Efficient Frontier 1 / 34

 Frontier Riskless CARA

Asset Allocation vs Asset Pricing

Asset allocation (or portfolio choice) is theory of how

investor allocates wealth amongst financial assets
For simplicity, assume investor is “price taker” =⇒ scale of
investment is small enough to not affect prices
Also for simplicity, assume perfect “frictionless” financial
market with no taxes or transaction costs, etc.
Asset allocation goes hand-in-hand with asset pricing: if we
know how all investors choose to allocate their wealth, then
we can find equilibrium prices to balance supply and demand
Harry Markowitz pioneered modern concept of asset allocation
with theory of mean-variance-efficient frontier in 1952

Wang Wei Mun Efficient Frontier 3 / 34

 Frontier Riskless CARA

Investment Environment

Financial market consists of n ≥ 2 risky tradable assets with

normal returns (and no riskless asset)
Let R = (R1 , . . . , Rn )′ be n × 1 vector of expected returns
Let V be n × n covariance matrix of returns, which consists of
variances on diagonal and covariances on off-diagonal
V must be symmetric: V′ = V, and positive definite:
z′ Vz > 0 for any non-zero n × 1 vector z
Assume no redundant assets, so returns must be linearly
independent and covariance matrix must be invertible:

∃ V−1 such that V−1 V = I

Then V−1 is also symmetric and positive definite

Wang Wei Mun Efficient Frontier 4 / 34

 Frontier Riskless CARA

Portfolio Weights

Let w = (w1 , . . . , wn )′ be n × 1 vector of portfolio weights,

which represents proportion of investor’s wealth allocated to
each tradable financial asset
No restriction on individual portfolio weights: positive weight
indicates normal investment (or “long position”) while
negative weight indicates short-selling (or “short position”)
Only restriction is that portfolio weights must sum to one:
w′ e = 1, where e = (1, . . . , 1)′ is n × 1 unit vector
Investor can allocate more than available wealth into any
individual financial asset, but cannot allocate more than
available wealth in aggregate
Then w′ R is expected return for investor’s portfolio, and
w′ Vw > 0 is variance of return for investor’s portfolio

Wang Wei Mun Efficient Frontier 5 / 34

 Frontier Riskless CARA

Asset Allocation Problem – Part 1

Investor’s ultimate goal is to create “optimal” portfolio that

maximises expected utility (of wealth)
Requires knowledge of investor’s utility function, which is
difficult to observe in reality
Since asset returns have normal distribution, investor’s
expected utility depends only on mean return and variance of
return for investor’s portfolio
Risk-averse investor will always prefer higher mean return and
lower variance of return
Hence simpler problem is to identify all“efficient” or “frontier”
portfolios that minimise risk for given mean return, which
includes specific optimal portfolio for any risk-averse investor

Wang Wei Mun Efficient Frontier 6 / 34

 Frontier Riskless CARA

Asset Allocation Problem – Part 2

Portfolio weights for portfolio with mean return of Rp must

satisfy two conditions:
Portfolio weights must sum to one: w′ e = 1
Portfolio must have mean return of Rp : w′ R = Rp
Out of all portfolios that satisfy these two conditions, frontier
portfolio has lowest variance of return: w′ Vw
In mathematical terms, finding portfolio weights for frontier
portfolio is constrained minimisation problem with quadratic
objective function and equality constraints
Fortunately, this minimisation problem is guaranteed to have
unique solution that is global minimum

Wang Wei Mun Efficient Frontier 7 / 34

 Frontier Riskless CARA

Asset Allocation Problem – Part 3

Solve constrained minimisation problem by adding Lagrange

multipliers to objective function (or “Langragian”) to ensure
that equality constraints are satisfied:

min L = w′ Vw + λ Rp − w′ R + γ 1 − w′ e



{w,λ,γ}

Here λ and γ are Lagrange multipliers for equality constraints

on mean return and portfolio weights, respectively
From economic perspective, Lagrange multiplier represents
marginal cost (“shadow price”) of relaxing constraint
Alternative “dual problem” is to maximise mean return for
specified variance of return, but more difficult to solve

Wang Wei Mun Efficient Frontier 8 / 34

 Frontier Riskless CARA

Frontier Portfolios – Part 1

Set partial derivative of Lagrangian to zero to get first-order

optimality condition for portfolio weights of frontier portfolio:

∂L
= Vw∗ − λR − γe = 0
∂w

Use other first-order optimality conditions (for Lagrange

multipliers) to confirm that equality constraints are satisfied:

∂L
= Rp − w′ R = 0 =⇒ w ′ R = Rp
∂λ
∂L
= 1 − w′ e = 0 =⇒ w′ e = 1
∂γ

Wang Wei Mun Efficient Frontier 9 / 34

 Frontier Riskless CARA

Frontier Portfolios – Part 2

Pre-multiply first-order optimality condition for frontier

portfolio weights by V−1 and rearrange:

Vw∗ − λR − γe = 0 =⇒ w∗ = λV−1 R + γV−1 e

Pre-multiply by R′ and apply constraint for mean return:

R′ w∗ = λR′ V−1 R + γR′ V−1 e = Rp

Pre-multiply by e′ and apply constraint for portfolio weights:

e′ w∗ = λe′ V−1 R + γe′ V−1 e = 1

Wang Wei Mun Efficient Frontier 10 / 34

 Frontier Riskless CARA

Frontier Portfolios – Part 3

Solve simultaneous equations to find Lagrange multipliers:

δRp − α ζ − αRp
λ= ; γ=
ζδ − α2 ζδ − α2

Here α is scalar, while ζ and δ are strictly positive scalars:

α = R′ V−1 e; ζ = R′ V−1 R; δ = e′ V−1 e

Confirm that denominator is strictly positive:

(αR − ζe)′ V−1 (αR − ζe) = ζ ζδ − α2 > 0

Wang Wei Mun Efficient Frontier 11 / 34

 Frontier Riskless CARA

Frontier Portfolios – Part 4

Substitute for λ and γ to find portfolio weights for frontier

portfolio with mean return of Rp :
   
∗ δRp − α −1 ζ − αRp
w = V R+ V−1 e
ζδ − α2 ζδ − α2

Rearrange to get linear relationship: w∗ = a + bRp , where:

ζV−1 e − αV−1 R δV−1 R − αV−1 e

a= ; b=
ζδ − α2 ζδ − α2

Minimum-variance frontier consists of portfolios with lowest

amount of risk, for different values of Rp

Wang Wei Mun Efficient Frontier 12 / 34

 Frontier Riskless CARA

Efficient Frontier – Part 1

Variance of return for frontier portfolio:

σp2 = (w∗ )′ Vw∗ = (a + bRp )′ V (a + bRp )

δRp2 − 2αRp + ζ
=
ζδ − α2
1 δ
= + (Rp − Rmv )2
δ ζδ − α2

α
Rmv = δ is mean return for global minimum-variance portfolio
Minimum-variance frontier is parabola when plotted with
variance of return on y -axis and expected return on x-axis
Standard practice to flip axes, as shown on next slide

Wang Wei Mun Efficient Frontier 13 / 34

 Frontier Riskless CARA

Rp
Efficient Frontier

Rmv

σp2

Wang Wei Mun Efficient Frontier 14 / 34

 Frontier Riskless CARA

Efficient Frontier – Part 2

Top half of minimum-variance frontier (where Rp ≥ Rmv ) is

known as efficient frontier, consisting of portfolios with
highest mean return for given amount of risk
If use standard deviation of return (instead of variance of
return) on x-axis, then minimum-variance frontier is hyperbola
with center at (0, Rmv ) and asymptotes:

1
α2 2

Rp = Rmv ± ζ− σp
δ

To maximise expected utility, investor will choose optimal

portfolio where indifference curve is tangent to frontier

Wang Wei Mun Efficient Frontier 15 / 34

 Frontier Riskless CARA

Efficient Frontier
Expected Return

Std Dev of Return

Wang Wei Mun Efficient Frontier 16 / 34

 Frontier Riskless CARA

Portfolio Separation – Part 1

Affine combination is linear combination with (positive or

negative) coefficients that sum to one, so affine combination
of two portfolios is also portfolio:

[κw1 + (1 − κ) w2 ]′ e = κw1′ e + (1 − κ) w2′ e = 1

If p1 and p2 are both frontier portfolios, then affine

combination of p1 and p2 is also frontier portfolio:

κw1∗ + (1 − κ) w2∗ = κ (a + bRp1 ) + (1 − κ) (a + bRp2 )

= a + b [κRp1 + (1 − κ) Rp2 ]
= a + bRp3

Wang Wei Mun Efficient Frontier 17 / 34

 Frontier Riskless CARA

Portfolio Separation – Part 2

Here p3 is frontier portfolio with mean return of

Rp3 = κRp1 + (1 − κ) Rp2
No restriction on value of κ, so investor can generate entire
minimum-variance frontier with different affine combinations
of any two frontier portfolios
Two-fund (or mutual fund) separation theorem: investor
can construct optimal portfolio with appropriate affine
combination of any two frontier portfolios
In theory, more convenient since investor can generate
efficient frontier and construct optimal portfolio without
knowing R and V for individual risky assets

Wang Wei Mun Efficient Frontier 18 / 34

 Frontier Riskless CARA

Orthogonal Frontier Portfolios – Part 1

Covariance of return between two frontier portfolios:

(w1∗ )′ Vw2∗ = (a + bRp1 )′ V (a + bRp2 )

1 δ
= + (Rp1 − Rmv ) (Rp2 − Rmv )
δ ζδ − α2

For given p1 , set covariance to zero to find mean return for

p2 , which is frontier portfolio that is “orthogonal” to p1 :

ζδ − α2
Rp2 = Rmv −
δ 2 (Rp1 − Rmv )

If p1 is efficient, then p2 must be “inefficient” (and vice versa)

Wang Wei Mun Efficient Frontier 19 / 34

 Frontier Riskless CARA

Orthogonal Frontier Portfolios – Part 2

Can measure slope at any point of minimum-variance frontier:

∂Rp ζδ − α2
= σp
∂σp δ (Rp − Rmv )

Evaluate at (σp1 , Rp1 ) to get slope of minimum-variance

frontier at point corresponding to specific frontier portfolio p1
Hence equation for line (with unknown y -intercept of R0 ) that
is tangent to frontier at point corresponding to p1 :

ζδ − α2
 
Rp = R0 + σp σp
δ (Rp1 − Rmv ) 1

Wang Wei Mun Efficient Frontier 20 / 34

 Frontier Riskless CARA

Orthogonal Frontier Portfolios – Part 3

Evaluate at (σp1 , Rp1 ) and solve for y -intercept:

ζδ − α2
R0 = Rp1 − σ2
δ (Rp1 − Rmv ) p1
ζδ − α2
 
1 δ 2
= Rp1 − + (Rp1 − Rmv )
δ (Rp1 − Rmv ) δ ζδ − α2
ζδ − α2
= Rmv − 2
δ (Rp1 − Rmv )
= Rp2

Hence y -intercept for tangent line at p1 also shows mean

return for frontier portfolio that is orthogonal to p1

Wang Wei Mun Efficient Frontier 21 / 34

 Frontier Riskless CARA

Investment Environment with Riskless Asset

Financial market consists of n ≥ 2 risky assets (with normal

returns) and riskless asset with risk-free rate of Rf
Let w be vector of portfolio weights for risky assets, so that
1 − w′ e is proportion of wealth invested in riskless asset
If w′ e < 1, then investor is lending money (to other investors,
through bank) at risk-free rate
If w′ e > 1, then investor is borrowing money (from other
investors, through bank) at risk-free rate
Expected return for investor’s portfolio:

w′ R + 1 − w′ e Rf = Rf + w′ (R − Rf e)

Here R − Rf e represents n × 1 vector of risk premiums

Wang Wei Mun Efficient Frontier 23 / 34

 Frontier Riskless CARA

Asset Allocation with Riskless Asset – Part 1

Lagrangian for asset allocation problem:

1
min L = w′ Vw + λ Rp − Rf − w′ (R − Rf e)
 
{w,λ} 2

Use optimality condition to find frontier portfolio weights:

Vw∗ − λ (R − Rf e) = 0 =⇒ w∗ = λV−1 (R − Rf e)

Use expected return to solve for Lagrange multiplier:

Rp = Rf + λ (R − Rf e)′ V−1 (R − Rf e)
Rp − Rf
= Rf + λ ζ − 2αRf + δRf2 =⇒ λ =


ζ − 2αRf + δRf2

Wang Wei Mun Efficient Frontier 24 / 34

 Frontier Riskless CARA

Asset Allocation with Riskless Asset – Part 2

Variance of return for portfolio on minimum-variance frontier:

σp2 = (w∗ )′ Vw∗ = λ2 (R − Rf e)′ V−1 (R − Rf e)

(Rp − Rf )2
=
ζ − 2αRf + δRf2

Minimum-variance frontier in (σp , Rp )-space:

21
Rp = Rf ± ζ − 2αRf + δRf2 σp

Hence efficient frontier consists of straight line with

1
y -intercept of Rf and positive slope of ζ − 2αRf + δRf2 2

Wang Wei Mun Efficient Frontier 25 / 34

 Frontier Riskless CARA

Efficient Frontier
Expected Return

Std Dev of Return

Without Riskless Asset With Riskless Asset

Wang Wei Mun Efficient Frontier 26 / 34

 Frontier Riskless CARA

Portfolio Separation with Riskless Asset – Part 1

If Rf < Rmv = αδ , then efficient frontier (with riskless asset) is

tangent to top half of risky-asset-only frontier
To verify this result, consider adding riskless asset to existing
frontier generated by n risky assets
Capital allocation line (CAL) is line in (σp , Rp )-space
joining riskless asset to any risky portfolio
CAL shows mean return and std dev of return for different
affine combinations of risky portfolio and riskless asset
Slope of CAL shows Sharpe ratio for all affine combinations
“Tangency” portfolio is unique risky portfolio where CAL is
tangent to existing frontier generated by n risky assets

Wang Wei Mun Efficient Frontier 27 / 34

 Frontier Riskless CARA

Portfolio Separation with Riskless Asset – Part 2

Tangency portfolio must have highest Sharpe ratio out of all

possible risky portfolios
Set y -intercept to Rf , and use previous result for equation of
tangent line to get mean return for tangency portfolio:

ζδ − α2 αRf − ζ
Rtg = Rmv − 2
=
δ (Rf − Rmv ) δRf − α

Hence risk premium for tangency portfolio:

αRf − ζ ζ − 2αRf + δRf2

Rtg − Rf = − Rf = −
δRf − α δ (Rf − Rmv )

Wang Wei Mun Efficient Frontier 28 / 34

 Frontier Riskless CARA

Portfolio Separation with Riskless Asset – Part 3

Variance of return for tangency portfolio:

2 1 δ (Rtg − Rmv )2 1 ζδ − α2
σtg = + = +
δ ζδ − α2 δ δ 3 (Rf − Rmv )2
ζδ − α2 ζ − 2αRf + δRf2
 
1
= 1+ =
δ (δRf − α)2 δ 2 (Rf − Rmv )2

Choose positive square root to get std dev of return:

1
ζ − 2αRf + δRf2 2
σtg =−
δ (Rf − Rmv )

Wang Wei Mun Efficient Frontier 29 / 34

 Frontier Riskless CARA

Portfolio Separation with Riskless Asset – Part 4

Hence slope of CAL and Sharpe ratio for tangency portfolio:

 
δRf2 
 
Rtg − Rf ζ − 2αRf + δ (Rf − Rmv )
= − −
1

σtg δ (Rf − Rmv ) ζ − 2αRf + δRf2 2

1
2 2
= ζ − 2αRf + δRf

Confirms that CAL for tangency portfolio is efficient frontier

(with riskless asset), since risk-averse investors prefer to hold
risky portfolio with highest possible “reward-to-risk” ratio
Opposite result for Rf > Rmv : tangency portfolio lies on
bottom half of existing frontier and has lowest Sharpe ratio

Wang Wei Mun Efficient Frontier 30 / 34

 Frontier Riskless CARA

CARA Utility: Economic Environment

Financial market consists of n ≥ 2 risky assets (with normal

returns) and riskless asset with risk-free rate of Rf
Let R̃ be n × 1 vector of asset returns, so portfolio return:
 
R̃p = Rf + w′ R̃ − Rf e

Investor has constant absolute risk aversion, so let br = bW0

be coefficient of relative risk aversion (at initial wealth)
Hence investor’s utility of (random) final wealth:

−b W̃
 
U W̃ = −e −bW̃ = −e r W0 = −e −br R̃p

Wang Wei Mun Efficient Frontier 32 / 34

 Frontier Riskless CARA

CARA Utility: Asset Allocation

Asset returns have joint normal distribution, so utility of final

wealth has lognormal distribution:
h  i h i
E U W̃ = E −e −br R̃p
′ 1 2 ′
= −e −br [Rf +w (R−Rf e)]+ 2 br w Vw

Exponential function is monotonically increasing, so minimise

(negative) exponent to maximise expected utility:
 
1
max w′ (R − Rf e) − br w′ Vw
w 2

Wang Wei Mun Efficient Frontier 33 / 34

 Frontier Riskless CARA

CARA Utility: Optimal Portfolio

Use optimality condition to find optimal portfolio weights:

1 −1
R − Rf e − br Vw∗ = 0 =⇒ w∗ = V (R − Rf e)
br

Pre-multiply by W0 e′ to find absolute (dollar) amount of

wealth invested in risky assets:

1
W0 e′ w ∗ = (α − δRf )
b

Notice that W0 doesn’t show up on RHS =⇒ all investors

with same coefficient of absolute risk aversion will invest same
(dollar) amount in risky assets, regardless of initial wealth

Wang Wei Mun Efficient Frontier 34 / 34