16-Port Selection Mean Var Page 486 Wednesday, February 4, 2004 1:09 PM
486 The Mathematics of Financial Modeling and Investment Management
Suppose asset returns are determined by a multifactor model (as
described in Chapter 18) so that the expected return of the i-th security
is a linear combination of p factors. We can then write
p
μi = αi + ∑βijfj , j = 1,2,..., p
j = 1
where μi are expected returns and fj are the expectations of factors.
Exposure to the j-th factor can be controlled by constraining the
beta βaj of portfolio a relative to that factor:
N
∑ waiβij = βaj
i = 1
where wai are the weights of portfolio a.
A portfolio manager might want to maximize a portfolio’s return
given a target level of risk. This problem would lead to maximizing a
linear function subject to quadratic constraints of the form
wa′ΣΣΣΣwa = wa
In practice, however, a portfolio manager prefers to minimize a
function of the type:
wa′ΣΣΣΣwa – λwa ′μμμμ
where μμμμis the vector of securities’ expected returns and λ is a risk-aver-
sion parameter. A function of this type implements a compromise
between risk and returns.
Finally, a portfolio manager needs to impose lower thresholds on
portfolio weights to avoid portfolios being made up of a large number
of small holdings. This implies the constraints wai ≥ bi. In practice,
therefore, mean-variance portfolio selection leads to a quadratic optimi-
zation problem of the following type:
Minimize
wa′ΣΣΣΣwa – λwa ′μμμμ
subject to