British Journal of Mathematics & Computer Science

10(6): 1-15, 2015, Article no.BJMCS.19469

ISSN: 2231-0851

SCIENCEDOMAIN international
www.sciencedomain.org

Free Assets and Their Relations with Riskless Assets

Reza Keykhaei1∗ and Mohammad Taghi Jahandideh2
1 Faculty of Mathematics and Computer, Khansar, University of Isfahan, Isfahan, Iran.
2 Department of Mathematics Sciences, Isfahan University of Technology, 84156-83111, Isfahan,

Iran.

Article Information
DOI: 10.9734/BJMCS/2015/19469
Editor(s):
(1) Zhenyou Huang, Dept. of Mathematics, Nanjing University of Science and Technology,
Jiangsu, P. R. China.
Reviewers:
(1) Anonymous, MingDao University, Taiwan.
(2) Sergey A. Surkov, International Institute of Management LINK Zukovsky, Moscow, Russia.
Complete Peer review History: http://sciencedomain.org/review-history/10461

Original Research Article

Received: 11 June 2015
Accepted: 08 July 2015
Published: 09 August 2015

Abstract
Tobin’s one-fund theorem states that, when a portfolio is consisting of some risky assets and a
riskless asset (with return rc ), then every efficient portfolio in the Mean-Variance optimization is
a combination of the tangency portfolio and the riskless asset. We introduce the notion of free
asset, which is an uncorrelated risky asset, and convert the problem for determining the tangency
portfolio to a problem with lower complexity, which requires smaller portfolio, by excluding free
assets with mean return rc from initial portfolio. We show that a set of free assets, with the same
mean return, can be replaced by one particular free asset with the mean return to obtain the
same results. We also show that free assets (or a set of free assets) with mean return rc and the
riskless asset have a close connection and under special conditions, they almost have the same
role in Mean-Variance portfolio selection problems.
Keywords: Mean-Variance optimization; efficient frontier; tangency portfolio; one-fund theorem.
2010 Mathematics Subject Classification: 91G10;90C20;90C29.

1 Introduction
The seminal work on Modern Portfolio Theory was originated by Markowitz [1] when he presented
his Mean-Variance (M-V) portfolio selection approach. It is a single-period investment theory in a

*Corresponding author: E-mail: r.keykhaei@math.iut.ac.ir

 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

set of risky assets that specifies the trade-off between the mean and the variance of a portfolio’s rate
of return, where the objective is to minimize the variance of portfolio’s return for a target level of
expected return. The idea of introducing a riskless asset is attributed to Tobin [2] who considered
cash in his portfolio selection problem. Since then, there has been a huge contributions about
portfolio optimization. Recently Steinbach [3] in a survey paper reviewed different models in this
case. The mathematics of solutions of optimal portfolios and efficient frontier in the Mean-Variance
asset allocation have been discussed by many authors. The paper of Merton [4] in 1972 was the
original full analytic treatment in this case. Merton [4] and Roll [5] developed the geometric aspects
of M-V analysis. In a recent article Bick [6] describes some results in portfolio selection using
analytical geometry theorems. Recently Keykhaei and Jahandideh [7] have considered portfolio
selection problem under condition of variable weights.

In this paper, we investigate analytically the effect of adding a new asset to the initial portfolio
such that its return is uncorrelated with the return of the other assets of the portfolio. The most
simple situation occurs when this uncorrelated asset be a cash account. We show that there is a
close connection between an uncorrelated asset and a cash account with the same mean return.
This concept helps us to derive most of our results. Also, we describe the situation where an
uncorrelated asset is replaced by a set of uncorrelated assets with an identical mean return. Under
special conditions, we show that this replacement yields the same outcomes.

The identification of the tangency portfolio is pivotal in the concept of one-found theorem, which
states that every efficient portfolio can be constructed by a combination of the tangency portfolio
and the cash account, and in deriving the Capital Allocation Line (CAL) (see Tütüncü [8]). We
show that the presence or absence of uncorrelated assets, with the same mean return with that of
the cash return, the tangency portfolio does not change. So we can reduce the initial portfolio to
a new portfolio, which does not contain such free assets, in order to reduce the complexity of the
procedure for finding the tangency portfolio, when a large number of assets exist in the portfolio.
In the following we use some notations and concepts from [7] throughout the paper.

The paper is organized as follows. In Section 2 we review the notion of risky assets, which was
introduced in [3]. This section contains all notations and problem formulations needed in other
sections. We present our problem formulation and analytical results in Section 3. The evaluation
of change in optimal solutions and the geometric of tangency portfolio are discussed in this section.
In Section 4 we examine the equivalence conditions for two portfolio selection problems in explicit
and limit sense.

2 Risky Assets
Consider a portfolio consisting of n risky assets. Let R be the n-column random vector and Σ
be the covariance matrix of asset returns r1 , . . . , rn and let R̄ := E(R) be the n-column vector of
mean returns r̄1 , . . . , r̄n of these n assets. The mean returns and covariance matrix of the assets
are assumed to be known. Also let X and 1 denote the n-column vectors of asset weights and ones,
respectively. We display the ith asset by its weight xi and each portfolio by the vector of its asset
weights X = (x1 , . . . , xn )′ . For each portfolio X, its return is the random variable X′ R. Using [[3]],
we present the definition of reward and risk of a portfolio and basic assumptions as follows:

Definition 2.1. The reward of a portfolio is the mean of its return,

ρ(X) = E(X′ R) = X′ R̄,
and the risk of a portfolio is the variance of its return,
R(X) = σ 2 (X′ R) = X′ ΣX.

2
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Problem 1. We wish to solve the following mean-variance portfolio selection problem for the
specific reward ρ,
1
min X′ ΣX
X 2
s.t. X′ 1 = 1,
X′ R̄ = ρ.
In fact this problem minimizes risk subject to two constraints. The first constraint, budget equation,
means that all the wealth must be invested in portfolio and the second constraint, reward equation,
implies that portfolio must earn an expected return equal to the prescribed reward ρ.

We require the following basic assumptions:

• (A.1) The covariance matrix Σ is positive definite.
• (A.2) The mean vector R̄ is not a multiple of 1.
The first assumption implies that all assets are risky and that no asset can be represented as a
linear combination of the other assets. The second assumption means that n ≥ 2 and guarantees
a nondegenerate situation, otherwise if R̄ is a multiple of 1, say R̄ = k1, then Problem 1 can be
solved only for ρ = k.

We use the following notations which have been defined by Steinbach [[3]]:
α := 1′ Σ−1 1, β := 1′ Σ−1 R̄, γ := R̄′ Σ−1 R̄, δ := αγ − β 2 .
It is shown in Lemma 1.3 of [3] that α, γ and δ are positive.

On the first step we state Theorems 1.5 and Theorem 1.8 of [3], as the solution and optimal risk of
Problem 1, which the last is the variance of the return of the solution.
Theorem 2.1. Problem 1 has the unique primal-dual solution
X∗ := Σ−1 (λ∗ 1 + µ∗ R̄), λ∗ := (γ − βρ)/δ, µ∗ := (αρ − β)/δ. (2.1)
Also the optimal risk is
σ 2 (ρ) := λ∗ + µ∗ ρ = (αρ2 − 2βρ + γ)/δ, (2.2)
Proof. See Theorem 1.5 and Theorem 1.7 of [3].

As [3] we call λ∗ and µ∗ , the optimal budget multiplier and optimal reward multiplier,
respectively, and we call the whole of the graph of the optimal risk by efficient frontier.

3 Risky Assets and Free Assets

3.1 Risky Assets and One Free Asset
Consider n+1 risky assets x1 , . . . , xn , xf and let the return of the (n+1)th asset, xf , be uncorrelated
with the return of other n risky assets. We call such an asset free and display its return by rf . Let
R1 be the (n + 1)-column random vector of asset returns r1 , . . . , rn , rf and Σ1 be their covariance
matrix and R̄1 := E(R1 ) be the (n + 1)-column vector of mean return r̄1 , . . . , r̄n , r̄f for those n + 1
assets. Let 11 and X1 denote the (n + 1)-column vectors of ones and asset weights, respectively.
In fact the portfolio is X1 = (X′ , xf )′ , where X = (x1 , . . . , xn )′ . For convenience, we ascribe the
phrase “risky part” only to X and “optimal risky part” refers to the portion of risky part that
appears in the primal solution in all problems, throughout the paper. Also, let R̄ and Σ refer to
the risky part of the portfolio.

In this situation the assumptions are (A.1) and

3
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

• (A.3) The mean vector R̄1 is not a multiple of 11 .

As before we define the following notations:

α1 := 1′1 Σ−1 ′ −1 ′ −1
1 11 , β1 := 11 Σ1 R̄1 , γ1 := R̄1 Σ1 R̄1 , δ1 := α1 γ1 − β1 .
2

Assume that α, β, γ and δ are related to the risky part of the portfolio. By the assumptions we
have ( )
Σ 0
Σ1 = 2 ,
0 σf

where σf2 = V ar(rf ). Thus

( )
Σ−1 0
Σ−1
1 =
0 s

and

α1 = α + s, β1 = β + sr, γ1 = γ + sr2 , δ1 = δ + sδf (3.1)

where s := (σf2 )−1 , r := r̄f and δf := αr2 − 2βr + γ. Note that the definition of δf is similar to δ c
in Theorem 1.8 of [3] and it is positive, since δ and α are positive.

Although portfolio optimization under these assumptions is a special case of problem 1, since xf is
also risky, we state it as a new problem and explain its primal-dual solution.

Problem 2. Consider a portfolio consists of n risky assets and an additional asset xf . Then
the risk and reward are R(X1 ) = X′ ΣX + x2f σf2 = X′1 Σ1 X1 and ρ(X1 ) = X′ R̄ + xf r = X′1 R̄1 ,
respectively, and the optimization problem reads
( )′ ( )( )
1 X Σ 0 X 1 ′ 1
min = (X ΣX + x2f σf2 ) = X′1 Σ1 X1
X,xf 2 xf 0 σf2 xf 2 2

s.t. X′ 1 + xf = X′1 11 = 1,
X′ R̄ + xf r = X′1 R̄1 = ρ.

We want to know that for a prescribed reward, how the optimal portfolio and optimal risk will
be effected by the presence of this free asset. It will soon be shown that changes are regular and
tractable. For this we start with a more simple situation. Although a cash account is uncorrelated
with any risky asset, we except it from the set of free assets, because its variance is zero and it is
not risky, and denote it by xf0 . We add a cash account to the initial portfolio and investigate the
effect of its presence in the optimal portfolios in more details.

Replace xf by xf0 in Problem 2. Notations (X∗0 ′ , x∗f0 )′ , λ∗0 , µ∗0 and σ02 (ρ) denote the primal and
dual solutions and optimal risk of Problem 2 when σf2 = 0. By Theorem 1.8 of [[3]]

X∗0 = Σ−1 (λ∗0 1 + µ∗0 R̄), x∗f0 = 1 − µ∗0 (β − rα),

(3.2)
λ∗0 = −rµ∗0 , µ∗0 = (ρ − r)/δf .

In this case the optimal risk is

σ02 (ρ) = (ρ − r)2 /δf = µ∗0 (ρ − r). (3.3)

4
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Henceforth let X∗ (r), λ∗ (r) and µ∗ (r) denote the primal and dual solutions of Problem 1, for
prescribed reward r. The difference between optimal budget multipliers is
γ − βρ r(r − ρ)
λ∗ − λ∗0 = −
δ δf
1
= (γ − βr)((αr − β)ρ + (γ − βr))
δδf
δ γ − βr αr − β γ − βr
= ( )(( )ρ + ( ))
δf δ δ δ
∗
= k0 λ (r),
∗ ∗
where k0 := (δ/δf )(λ (r) + µ (r)ρ). Similarly we see that
µ∗ − µ∗0 = k0 µ∗ (r).
Therefore
X∗ − X∗0 = Σ−1 (λ∗ 1 + µ∗ R̄) − Σ−1 (λ∗0 1 + µ∗0 R̄)
= Σ−1 ((λ∗ − λ∗0 )1 + (µ∗ − µ∗0 )R̄)
= k0 Σ−1 (λ∗ (r)1 + µ∗ (r)R̄)
= k0 X∗ (r).
We summarize the above result in the following theorem.
Theorem 3.1. The difference between the primal and dual solutions of Problem 1 and that of
problem 2 when σf2 = 0, are equal to the primal and dual solutions of Problem 1 for prescribed
reward r, multiplied by k0 , namely
X∗ − X∗0 = k0 X∗ (r), λ∗ − λ∗0 = k0 λ∗ (r), µ∗ − µ∗0 = k0 µ∗ (r),
where k0 := (δ/δf )(λ∗ (r) + µ∗ (r)ρ).
Now we are able to calculate the difference between the optimal risks. The following seems to
be an extension of Theorem 1.10 of [3] (see also propositions 9.1 and 9.2 of [6]).
Theorem 3.2. For σf2 = 0, in contrast with Problem 1, Problem 2 almost always has lower optimal
risk. Precisely if β ̸= rα, then
δ
σ 2 (ρ) − σ02 (ρ) = ( )−1 k02
δf
and the efficient frontiers are tangent to each other at the point
γ − rβ δf
(ρT , σ 2 (ρT )) := ( , ),
β − rα (β − rα)2
where both problems have identical solutions, i.e., (X∗0 ′ , x∗f0 )′ = (X∗′ , 0)′ . If β = rα, then x∗f0 = 1
and X∗0 1 = 0. Moreover
1
σ 2 (ρ) − σ02 (ρ) = .
α
For a geometrical interpretation of Theorem 3.2, see Figure 3.1 and Figure 3.1.

Proof. By Theorem 2.1, equation 3.3 and Lemma 3.1, we have

σ 2 (ρ) − σ02 (ρ) = (λ∗ + µ∗ ρ) − µ∗0 (ρ − r)
= (λ∗ − λ∗0 ) + (µ∗ − µ∗0 )ρ
= k0 (λ∗ (r) + µ∗ (r)ρ)
δ
= ( )−1 k02 .
δf
For the proof of the other assertions see the proof of Theorem 1.10 in [[3]].

5
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Here we illustrate the connection between a free asset and a cash account with the same return:
r = rc , where rc is the return of the cash account. By the assumption r = rc , we show that there
are interesting relations between optimal portfolios and optimal risks in the following three cases:
• (C.1) risky assets only,
• (C.2) risky assets and a free asset,
• (C.3) risky assets and a cash account.
In fact we establish relations between the solutions and consequently optimal risk of Problem 1 and
Problem 2 under the conditions σf2 ̸= 0 or σf2 = 0.

Let X∗1 , λ∗1 , µ∗1 and σ12 (ρ) denote the primal and dual solutions and optimal risk of Problem 2. By
Theorem 2.1
X∗1 = Σ−1 ∗ ∗
1 (λ1 11 + µ1 R̄1 ), (3.4)

δ ∗ δ
λ∗1 = (γ1 − β1 ρ)/δ1 = λ + (1 − )λ∗0 , (3.5)
δ1 δ1

δ ∗ δ
µ∗1 = (α1 ρ − β1 )/δ1 = µ + (1 − )µ∗0 . (3.6)
δ1 δ1
If X∗1 = (X′ , xf )′ , then
δ ∗ δ
X = Σ−1 (λ∗1 1 + µ∗1 R̄) = X + (1 − )X∗0 , (3.7)
δ1 δ1

δ ∗
xf = s(λ∗1 + rµ∗1 ) = (1 − )xf0 . (3.8)
δ1
Also the optimal risk is
δ 2 δ
σ12 (ρ) = λ∗1 + µ∗1 ρ = σ (ρ) + (1 − )σ02 (ρ). (3.9)
δ1 δ1
Although we denote the optimal solution of the Problem 1 by X∗ , we may denote it by (X∗′ , 0)′ .
This means that there is no free asset or its value is zero. Considering this and the fact that
0 < δ < δ1 , we can infer from equations 3.5-3.9 that, under the condition σf2 ̸= 0, the primal-dual
solution (resp. optimal risk) of problem 2 is a convex combination of the primal-dual solution (resp.
optimal risk) of Problem 1 and the primal-dual solution (resp. optimal risk) of Problem 2, when
σf2 = 0. Observe that the coefficients of these convex combinations are all the same. Moreover, δ/δ1
and 1 − δ/δ1 are independent of the choice of ρ and are valid for any target reward. We summarize
the above discussion in the following lemma.
Lemma 3.3. For any prescribed reward, the primal-dual solution (resp. optimal risk) of problem
2 for σf2 ̸= 0 is a convex combination of the primal-dual solution (resp. optimal risk) of Problem 1
and the primal-dual solution (resp. optimal risk) of Problem 2, when σf2 = 0. In all of these convex
combinations, coefficients are δ/δ1 and 1 − δ/δ1 .
This lemma enables us to state other versions of Theorems 3.1 and 3.2 for risky assets and a
free asset.
Theorem 3.4. Let X be the optimal risky part of Problem 2. Then
X∗ − X = k1 X∗ (r), λ∗ − λ∗1 = k1 λ∗ (r), µ∗ − µ∗1 = k1 µ∗ (r),
where
k1 := s(δ/δ1 )(λ∗ (r) + µ∗ (r)ρ).

6
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Proof. First we derive the difference between optimal budget multipliers. Using Lemma 3.3 and
Theorem 3.1, we have
δ sδf δ
λ∗ − λ∗1 = (1 − )(λ∗ − λ∗0 ) = k0 λ∗ (r) = s (λ∗ (r) + µ∗ (r)ρ)λ∗ (r) = k1 λ∗ (r),
δ1 δ1 δ1
where k1 := s(δ/δ1 )(λ∗ (r) + µ∗ (r)ρ). The proof of the remaining parts are similar.

Theorem 3.5. In contrast to Problem 1, Problem 2 almost always has lower optimal risk. Precisely,
if β ̸= rα, then
δ
σ 2 (ρ) − σ12 (ρ) = (s )−1 k2
δ1
and the efficient frontiers are tangent to each other at the point
γ − rβ δf
(ρT , σ 2 (ρT )) := ( , ),
β − rα (β − rα)2
where both problems have identical solutions, i.e., X∗1 = (X∗′ , 0)′ . Finally if β = rα, then in X∗1
X′ 1 = α/α1 , xf = s/α1 .
Moreover
s
σ 2 (ρ) − σ12 (ρ) = .
αα1
Proof. We define the difference between optimal risks by
f (ρ) := σ 2 (ρ) − σ12 (ρ).
By Lemma 3.3 we have
δ
f (ρ) = (1 − )(σ 2 (ρ) − σ02 (ρ)).
δ1
Since δ < δ1 , Theorem 3.2 implies that f (ρ) ≥ 0 and so σ 2 (ρ) ≥ σ12 (ρ).

If β ̸= rα, then by Theorem 3.2,

δ δ δ
f (ρ) = (1 − )( )−1 k02 = (s )−1 k12
δ1 δf δ1
and f has a root in ρT := (γ − rβ)/(β − rα). Hence σs2 (ρT ) = σ 2 (ρT ) = δf /(β − rα)2 . For ρ = ρT
both problems have the same optimal reward multiplier and the same optimal budget multiplier ,
i.e.,
1 r
µ∗ = µ∗1 = , λ∗ = λ∗1 = − .
β − rα β − rα
Let X∗1 = (X′ , xf )′ . Substitution of λ∗1 and µ∗1 in 3.8 implies that xf = 0. This substitution in 3.7
and 2.1 implies that the values of X and X∗ are identical, in both problems.

If β = rα, then δf = δ/α and by Theorem 3.2,

δ 1 s
f (ρ) = (1 − ) = ,
δ1 α αα1
which is independent of ρ. Therefore,
s
σ 2 (ρ) − σ12 (ρ) = .
αα1
Now by Theorem 3.2, x∗f0 = 1 and the proof is complete by 3.8.

7
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Note. As observed in Theorem 3.5, the tangent point between two efficient frontiers is
independent of the variance of the free asset. On the other hand, by Theorem 3.2 we see in
both situations that tangent points are identical, if r is the same in both theorems. Also optimal
portfolios associated to this point are identical. In fact in cases (C.1), (C.2) and (C.3) the efficient
frontiers are tangent at the point (ρT , σ 2 (ρT )), where they have the same optimal portfolio, (see
Figure 3.1). Also see Figure 3.1 for a condition that there is no tangency.

σ 2(ρ)
σ12(ρ)
σi2(ρ)
σ02(ρ)
Variance

(ρT , σ 2(ρT ))

r Mean
Fig. 1. Efficient frontiers when r ̸= β/α. The global minimum variance portfolios
are represented by black disks and tangent point is marked by a circle.

3.2 Risky Assets and More than One Free Asset

Let us now consider a portfolio with n ≥ 1 risky assets x1 . . . , xn , and m ≥ 1 free assets xf1 , . . . , xfm
with returns rf1 , . . . , rfm . We suppose that all free assets have the same expected return r. At first
we add the first free asset to the portfolio of n risky assets and solve the portfolio selection problem,
as stated in Section 3.1. We continue this procedure by adding ith free asset to the portfolio
consisting of n risky assets and previous i − 1 free assets and solve the portfolio selection problem
again, for i = 2, . . . , m. In each stage the portfolio is Xi = (X′ , X̂′i )′ , where X = (x1 , . . . , xn )′
and X̂i = (xf1 , . . . , xfi )′ . Let R̄i and Σi are the mean vector and the covariance matrix of asset
returns, respectively, and 1i is an (n + i)-column vector of ones. We have the following problem
and we denote it by Problem i′ :
1
min X′i Σi Xi
Xi 2

s.t. X′i 1i = 1,
X′i R̄i = ρ.

8
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

σ 2(ρ)
σ12(ρ)
σi2(ρ)
s/(αα1)
σ02(ρ) (αi − α)/(ααi)
Variance

1/α
1/α1
1/αi
r Mean
Fig. 2. Efficient frontiers when r = β/α. The global minimum variance portfolios
are represented by black disks.

Note that Problems 2 and 1′ are equivalent. Suppose the assumptions (A.1) and (A.3) hold and
define the following notations:

αi := 1′i Σ−1
i 1i , βi := 1′i Σ−1
i R̄i , γi := R̄′i Σ−1
i R̄i ,

and
(i)
δi := αi γi − βi2 , δf := αi r2 − 2βi r + γi ,

for i = 1, . . . , m. Also, let si := (V ar(rfi ))−1 . As a reminder, we mention that Σ, R̄ and the
constants α, β, γ and δ are related to the risky part of the portfolio.

Following conditions (C.1), (C.2) and (C.3), we denote the special situation that the mean return
of free assets, r, is equal with rc by
• (C.2′ ) risky assets and more than one free assets.
In the following we intend to present the relations between the cases (C.1), (C.2′ ) and (C.3).

Lemma 3.6. For i = 1, . . . , m,

1. αi = α + s1 + . . . + si ,
2. βi = β + (s1 + . . . + si )r,
3. γi = γ + (s1 + . . . + si )r2 ,
(i)
4. δf = δf ,
5. δi = δ + (s1 + . . . + si )δf .

9
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Proof. Parts (1)-(4) are obvious. We use induction to prove part (5).
δi+1 = αi+1 γi+1 − βi+1
2

= (αi + si+1 )(γi + si+1 r2 ) − (βi + si+1 r)2

= (αi γi − βi2 ) + si+1 (αi r2 − 2βi r + γi )
(i)
= δi + si+1 δf
= δ + (s1 + . . . + si+1 )δf .

Consider a portfolio consisting of n risky assets and free assets xf1 , . . . , xfi and a cash account
xf0 whit return r. In this case, we display each portfolio by X0i = (X′i , xf0 )′ = (X′ , X̂′i , xf0 )′ . Now
our portfolio selection problem is Problem i∗ :
1 ′
min Xi Σi Xi
X0i 2
s.t. X′0i 1i+1 = 1,
X′0i R̄i+1 = ρ.

We display the primal and dual solutions and optimal risk of Problem i∗ by X∗0i , λ∗0i , µ∗0i and σ0i
2
(ρ).

Theorem 3.7. Free assets don’t appear in the optimal portfolio of Problem i∗ . precisely, when σf2 =
0, Problem i∗ and Problem 2 have identical primal-dual solution. Therefore if X∗0i = (X′ , X̂′i , xf0 )′ ,
then
X = X∗0 , X̂i = (0, 0, . . . , 0)′ , xf0 = x∗f0 , λ∗0i = λ∗0 , µ∗0i = µ∗0 .
Consequently both problems have the same optimal risk.

Proof. Let X∗0i = (X′i , xf0 )′ . By Theorem 1.8 of [[3]]

Xi = Σ−1 ∗ ∗
xf0 = 1 − µ∗0i (β − rα), λ∗0i = −rµ∗0i , µ∗0i = (ρ − r)/δf ,
(i)
i (λ0i 1i + µ0i R̄i ),

and optimal risk is

(i)
σ02 (ρ) = (ρ − r)2 /δf .
Then
X = Σ−1 (λ∗0i 1 + µ∗0i R̄)
and
X̂i = (λ∗0i + rµ∗0i )(s1 , . . . , si )′ = (0, 0, . . . , 0)′ .
Now all statements can be obtained from Lemma 3.6 (4), 3.2 and 3.3.

Note. If we apply Theorem 3.2 for Problem i′ and Problem i∗ we see that σi2 (ρTi ) = σ0i 2
(ρTi ),
∗ ∗
where ρTi = (γi − rβi )/(βi − rαi ) and its related optimal portfolio of Problem i has xf0 = 0.
Also, it can be easily seen that ρTi = ρT . Now by Theorems 3.2 and 3.7 for ρTi we get X∗0i =
(X∗0 ′ , 0 . . . , 0, 0)′ = (X∗′ , 0 . . . , 0, 0)′ . Thus we conclude that if a portfolio contains a cash, then
in the absence or the presence of one or more than one free assets with mean return equal to
cash return, the tangency portfolio remains fixed. So, for determining the tangency portfolio, we
can constitute a smaller portfolio by excluding such free assets. This may provide a more simple
procedure with lower complexity in determining the tangency portfolio.

In the following let X∗i , λ∗i , µ∗i and σi2 (ρ) be the primal and dual solutions and optimal risk of
Problem i′ .

10
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Lemma 3.8. Let X∗i = (X′ , xf1 , . . . , xfi )′ , then

δ ∗ δ δf ∗
X= X + (1 − )X∗0 , xfj = sj xf , j = 1, . . . , i. (3.10)
δi δi δi 0
Also
∑
i
δ ∗
xfj = (1 − )xf0 . (3.11)
j=1
δi

Moreover
δ ∗ δ δ ∗ δ
λ∗i = λ + (1 − )λ∗0 , µ∗i = µ + (1 − )µ∗0 , (3.12)
δi δi δi δi

δ 2 δ
σi2 (ρ) = σ (ρ) + (1 − )σ02 (ρ). (3.13)
δi δi

Proof. The proof proceeds by induction on i. We do this for optimal weights. Other assertions can
be proved similarly. The case i = 1 is obvious from Lemma 3.3. Suppose the assertion is true for i.
We show that it holds for i + 1. By applying Lemma 3.3 for Problems i′ and (i + 1)′ we have

δi δi
X∗i+1 = (X∗i ′ , 0)′ + (1 − )(X∗0i ′ , x∗f0 )′ .
δi+1 δi+1

By Theorem 3.7 and the induction hypothesis we see that in X∗i+1 ,

δi δ ∗ δ δi δ δ
X= ( X + (1 − )X∗0 ) + (1 − )X∗0 = X∗ + (1 − )X∗0 ,
δi+1 δi δi δi+1 δi+1 δi+1
and
δi sj δf ∗ δf ∗
xfj = xf0 = sj xf , j = 1, . . . , i,
δi+1 δi δi+1 0
δi δf ∗
xfi+1 = (1 − )x∗f0 = si+1 xf .
δi+1 δi+1 0

Here, we state some results that can be derived from Lemma 3.8.

Theorem 3.9. Let X be the optimal risky part of Problems i′ . Then

X∗ − X = ki X∗ (r), λ∗ − λ∗i = ki λ∗ (r), µ∗ − µ∗i = ki µ∗ (r),

where
δ
ki := (s1 + . . . + si )( )(λ∗ (r) + µ∗ (r)ρ).
δi

Proof. By Lemma 3.8 and Theorem 3.1

δ δ δ
λ∗ − λ∗i = (1 − )(λ∗ − λ∗0 ) = (1 − ) (λ∗ (r) + µ∗ (r)ρ)λ∗ (r) = ki λ∗ (r),
δi δi δf

where ki = (s1 + . . . + si )(δ/δi )(λ∗ (r) + µ∗ (r)ρ). The proof of the other claims are similar.

11
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Theorem 3.10. In contrast to Problem 1, Problem i′ almost always has lower optimal risk.
Precisely, if β ̸= rα, then
δ −1 2
σ 2 (ρ) − σi2 (ρ) = [(s1 + . . . + si )( )] ki
δi
and the efficient frontiers are tangent at the point
γ − rβ δf
ρT = , σ 2 (ρT ) = ,
β − rα (β − rα)2
where both problems have identical solutions, i.e., X∗i = (X∗′ , X̂′i )′ , where X̂i = (0, 0, . . . , 0)′ .
Finally if β = rα, then in X∗i
sj
X′ 1 = α/αi , xfj = , j = 1, . . . , i.
αi
Moreover
αi − α
σ 2 (ρ) − σi2 (ρ) = .
ααi
The geometrical interpretation of Theorem 3.10 is given in Figures 3.1 and 3.1.

Proof. The claims about the tangent point can be proved by applying Theorem 3.5 inductively to
Problems j ′ and (j + 1)′ for all j < i, Since
γj − rβj γ − rβ
= .
βj − rαj β − rα

The other assertions can be obtained from Lemma 3.8 and Theorem 3.2. Note that if β = rα, then
δf = δ/α and x∗f0 = 1.

Note. If β ̸= rα, then by Theorem 3.10 we see that in each Problem i′ , for target reward ρT
all optimal portfolios have the same construction. So we may attribute the single phrase tangency
portfolio to all these optimal portfolios, which also appears in Theorem 3.2.

By a simple differentiation of optimal risk in Problem 1 (Problem i′ ) with respect to ρ, we see

that the return of the global minimum variance (GMV) portfolio, which has the smallest possible
variance, is equal to β/α (βi /αi ). It is shown that every optimal portfolio in Problem 1 (Problem
i′ ) with return ρ ,except the global minimum variance portfolio, has a unique orthogonal optimal
portfolio, in the sense that their covariance is zero, with return ρ̂ = (γ − ρβ)/(β − ρα) (ρ̂ =
(γi − ρβi )/(βi − ραi )) (see Corollary 3 of Roll [5] or Proposition 9.1 of Bick [6]). Geometrically, the
line passing between the optimal portfolio with return ρ̂ and the GMV portfolio in the mean-variance
plan, intersects mean axis in ρ (see Corollary 3.A of Roll [5]).
Proposition 3.1. If β ̸= rα, in Problem 1 and any Problem i′ , the optimal portfolio associated
to the reward r and tangency portfolio are orthogonal. Also, in the mean-variance plan, all GMV
portfolios corresponding to foregoing problems, lie on the line which passes from tangency portfolio
and intersects the mean axis in r (see Figure 3.1). If β = rα, then all GMV portfolios lie on the
line ρ = r (see Figure 3.1).
Proof. It is obvious that, β ̸= rα if and only if βi ̸= rαi . Let β ̸= rα and ρ = r. Then by the
discussion prior to this proposition, we see that in each of the above problems, the optimal portfolio
corresponding to reward r is not GMV portfolio and has an orthogonal portfolio with return
γi − rβi γ − rβ
ρ̂ = = = ρT ,
βi − rαi β − rα
which is the return of tangency portfolio. Thus the line passing between the tangency portfolio
and the global minimum variance portfolio of problem i′ , intersects mean axis in r. Thus all global
minimum variance portfolios lie on this line. The case β = rα is trivial.

12
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

Note. By Theorems 3.9 and 3.10 we can directly check the orthogonality stated in Proposition
3.1. Observe that

X∗i (r)′ Σi X∗i (ρT ) = (1 + ki )X∗ (r)′ ΣX∗ (ρT )

1 + ki
= [(γ − βr)1′ + (αr − β)R̄′ ]Σ−1 [(γ − βρT )1 + (αρT − β)R̄]
δ2
( ′ ) ( )
1 + ki ( ) 1 −1 ( ) γ − βρT
= γ − βr αr − β ′ Σ 1 R
δ2 R αρT − β
( )( )
1 + ki ( ) γ −β 1
= 1 r
δ −β α ρT
= 0,

where in the first line of above equations, we denote each optimal wight by its related reward.

4 Equivalent Conditions
We mention that two portfolio selection problems are equivalent if they generate the same efficient
frontier. In the case that both problems contain only risky assets, 2.2 implies that if for each
reward ρ, those two problems have the same optimal budget multiplier and the same optimal
reward multiplier, then they are equivalent.

Theorem 4.1. Let j < i. Problem i′ with the set of free assets {xf1 , . . . , xfi } and Problem j ′ with
the set of free assets {xf1 , . . . , xfj−1 , xf ∗ } (for j = 1 this set contains xf ∗ only) are equivalent,
∑
where s∗ = (V ar(rf ∗ ))−1 = ik=j sk and rf ∗ is the return of xf ∗ . Indeed both problems have the
same optimal budget multiplier and the same optimal reward multiplier. Also, for target reward ρ,
in optimal portfolios of both problems, individual weights of common assets are identical and

∑
i
xf ∗ = xfk .
k=j

Proof. Let {xf1 , . . . , xfj−1 , xf ∗ } be the set of free assets in Problem j ′ . Then for this set

δj = δ + (s1 + . . . + sj−1 + s∗ )δf = δ + (s1 + . . . + si )δf = δi .

Now the claim can be proved by using Lemma 3.8. For instance in optimal portfolio of Problem j ′

δf δf ∑ δf
i ∑ i
xf ∗ = s∗ xf ∗ = s∗ xf0∗ = (sk xf0∗ ) = xfk .
δj 0 δi δi
k=j k=j

Note. Theorem 4.1 is helpful for deriving those differences stated in Theorems 3.9 and 3.10.
For this, let j = 1 and define the dummy asset xf ∗ with the conditions of Theorem 4.1. Now the
assertions can directly be proved by using Theorems 3.4 and 3.5.

Another kinds of equivalence relation can be presented between conditions (C.2) and (C.3), also
(C.2′ ) and (C.3). Indeed they are equivalent in the limit sense. If we posses one free asset, (cases
(C.2) and (C.3)), then by 3.9
δ 2
σ12 (ρ) − σ02 (ρ) = (σ (ρ) − σ02 (ρ)).
δ1

13
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

We display the left side of the above equation by R(s). It is obvious that R(s) decreases to zero
when s increases to infinity. Similarly, if we have m > 1 free assets, (cases (C.2′ ) and (C.3)), then
by 3.13
δ
σm2
(ρ) − σ02 (ρ) = (σ 2 (ρ) − σ02 (ρ)).
δm
∑
In this case we display the left side of the above equation by R(s∗ ), where s∗ = m j=1 sj . Again,
we see that by increasing s∗ , σm 2
coincides with σ02 . This can occur when more and more free
assets with low variances are added into the portfolio. By the above analysis we have the following
theorem:

Theorem 4.2. In the M-V portfolio selection problem, the following conditions are equivalent in
the limit:
1. The cases (C.2) and (C.3), if σf2 tends to zero.
∑m
2. The cases (C.2′ ) and (C.3), if (s∗ )−1 tends to zero, when s∗ = j=1 sj .

Note. If the condition of part (1) of Theorem 4.2 holds, we see that the primal-dual solution
of (C.2) converges to the primal-dual solution of (C.3) by 3.5-3.8. Also, if the condition of part (2)
of Theorem 4.2 holds, then 3.10 and 3.12 implies that the optimal risky part and the dual solutions
of (C.2′ ) converge to the optimal risky part and the dual solutions of (C.3), respectively. Moreover,
by 3.11 we can check that the sum of optimal weights of free assets tends to the optimal weight of
cash account.

5 Discussion
The major assumption in this paper is that the portfolio contains an uncorrelated asset. We can
discuss about the existence of such asset in the market. Moreover it is assumed that the cash account
and the free asset have the same expected return. It is relevant to study the Mean-Variance portfolio
optimization problem where the expected return of the free asset and the cash account are not equal.
Also we can study the case where the portfolio contains a set of free assets with different expected
returns.

6 Conclusions
In this paper we investigate Mean-Variance portfolio optimization problem when the portfolio
contains an uncorrelated asset (free asset). The main results are summarized as follows.
I. It is shown in subsection (3.1) that the free asset does not appear in the tangency portfolio if
the free asset and the cash account have the same expected return. It is shown in subsection
(3.2) that we can reduce the optimal risk by using more free assets.
II. In the section (4) we discuss the situation for which a set of free assets act like a single free asset.

Competing Interests
The authors declare that no competing interests exist.

References
[1] Markowitz HM. Portfolio selection. The Journal of Finance. 1952;7:77-91.

14
 Keykhaei & Jahandideh; BJMCS, 10(6), 1-15, 2015; Article no.BJMCS.19469

[2] Tobin J. Liquidity preference as a behavior towards risk. Review of Economic Studies.
1958;25:65-86.
[3] Steinbach MC. Markowitz revisited: Mean-variance models in financial portfolio analysis. Siam
Review. 2001;43:31-85.
[4] Merton RC. An analytic derivation of the efficient portfolio frontier. Journal of Financial and
Quantitative Analysis. 1972;7:1851-1872.
[5] Roll R. A critique of the asset pricing theory’s tests. Part I. On past and potential testability
of the theory. Journal of Financial Economics. 1977;4:129-176.
[6] Bick A. The mathematics of the portfolio frontier: A geometry-based approach. Quarterly
Review of Economics and Finance. 2004;44(2):337-361.
[7] Keykhaei R, Jahandideh MT. Portfolio selection under condition of variable weights. Applied
Mathematics. 2012;3(10A):1505-1515.
[8] Tütüncü RH. A note on calculating the optimal risky portfolio. Finance and Stochastics.
2001;5:413-417.
——————————————————————————————————————————————–
⃝2015
c Keykhaei & Jahandideh; This is an Open Access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted
use, distribution, and reproduction in any medium, provided the original work is properly cited.

Peer-review history:
The peer review history for this paper can be accessed here (Please copy paste the total link in your browser
address bar)
http://sciencedomain.org/review-history/10461

15