Scribd
Upload
114 views3 pages

Maximizing The Sharpe Ratio

This document describes how to maximize the Sharpe ratio by formulating it as a convex quadratic program. It first defines the mean-variance portfolio optimization problem and considers maximizing the Sharpe ratio as an alternative objective. It then shows that under reasonable assumptions, this problem can be reduced to a standard convex quadratic program by making a variable substitution and observing the properties of the objective function. Specifically, it reformulates the problem as minimizing the portfolio variance subject to constraints on the mean return and linear inequality constraints, resulting in an optimization problem that can be solved efficiently.

Uploaded by

Dat Tran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
114 views3 pages

Maximizing The Sharpe Ratio

This document describes how to maximize the Sharpe ratio by formulating it as a convex quadratic program. It first defines the mean-variance portfolio optimization problem and considers maximizing the Sharpe ratio as an alternative objective. It then shows that under reasonable assumptions, this problem can be reduced to a standard convex quadratic program by making a variable substitution and observing the properties of the objective function. Specifically, it reformulates the problem as minimizing the portfolio variance subject to constraints on the mean return and linear inequality constraints, resulting in an optimization problem that can be solved efficiently.

Uploaded by

Dat Tran
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

IEOR 4500

Maximizing the Sharpe ratio

Suppose we have the setting for a mean-variance portfolio optimization problem:

µ, the vector of mean returns (1)


Q, the covariance matrix (2)
X
xj = 1, (proportions add to 1) (3)
j
Ax ≥ b, (other linear constraints). (4)
0 ≤ x. (5)

Note that we can use inequalities (4) to represent, in a generic way, many constraints,
including upper bounds on variables (constraints of the form xj ≤ uj ), as well as equations
and general inequalities of the form 00 ≤00 .
As an alternative to the standard mean-variance problem, we consider a different optimiza-
tion task. Let rf be the risk-free interest rate. Consider:

µT x − r f
maximize q (6)
xT Qx
s.t.
X
xj = 1,
j
Ax ≥ b.
0 ≤ x.

Problem (6) is difficult because of the nature of its objective. However, under a reasonable
assumption, it can be reduced to a standard convex quadratic program.
The assumption we make is: there exists a vector x satisfying (3)-(5) such that

µT x − rf > 0.

This assumption is reasonable: it simply says that our universe of assets is able to beat the
risk-free rate of return.
Our approach is as follows: given an asset vector x, define

µT x − r f
f (x) = q .
xT Qx

1
P
Since j xj = 1,

µT x − r f µT x − r f µ̂T x
P
j xj
f (x) = q = q = q ,
xT Qx xT Qx xT Qx

where for each index j, we define µ̂j = µj − rf .

Using this fact, we note:


P
Observation: For any vector x with j xj = 1, and any scalar λ > 0, f (λx) = f (x).
q q
To see this, check that if we write y = λx, then y T Qy = λ xT Qx, and similarly
µ̂T y = λµ̂T x.

Now we can state our optimization problem. Let  be the matrix whose i, j-entry is

aij − bi .

The problem we consider is:

1
maximize q (7)
y T Qy
s.t.
µ̂T y = 1, (8)
Ây ≥ 0. (9)
0 ≤ y. (10)

To see that problems (6) and (7) are indeed equivalent, suppose that ȳ is an optimal solution
P
to (7). Notice that because of (8), ȳ is not identically zero, and so by (10), j ȳj > 0. Define
the vector
ȳ
x̄ = P .
j ȳj

Then, by construction, X
x̄j = 1.
j

Further, since y satisfies (9), then for any row i we have


X
(aij − bi )ȳj ≥ 0,
j

or in other words, X X
aij ȳj ≥ ( ȳj )bi ,
j j

2
and as a consequence, X
aij x̄j ≥ bi .
j

Therefore, x̄ is feasible for problem (6). Further, as we observed before, f (x̄) = f (ȳ) =
√ 1 , since µ̂T ȳ = 1.
T
y Qy

In summary: the value of problem (6) is at least as large as the value of problem (7). The
converse is proved in a similar way. So, indeed, (6) and (7) are equivalent.

So we just have to solve (7). But this is clearly equivalent to:

minimize y T Qy
s.t.
µ̂T y = 1,
Ây ≥ 0.
0 ≤ y,

which is just a standard quadratic program.

You might also like