Navigator
Pre Link
On robust mean variance portfolios(1)
Robust MV model
Robust MV model
as follows, which we refer to the problem as RMVP2
:
max
x
min
r
∈
U
r
r
T
x
+
(
W
0
−
e
T
x
)
R
s
.
t
.
x
T
Σ
x
≤
t
2
\max_x\min_{r\in U_r}r^Tx+(W_0-e^Tx)R\\ s.t.\quad x^T\Sigma x\leq t^2
xmaxr∈UrminrTx+(W0−eTx)Rs.t.xTΣx≤t2
for an appropriately selected positive number
t
t
t, the model can be reformulated into
max
x
r
^
T
x
+
(
W
0
−
e
T
x
)
R
−
γ
∥
Σ
1
/
2
x
∥
2
s
.
t
.
x
T
Σ
x
≤
t
2
\max_x \hat{r}^Tx+(W_0-e^Tx)R-\gamma\lVert\Sigma^{1/2}x\rVert_2\\ s.t.\quad x^T\Sigma x\leq t^2
xmaxr^Tx+(W0−eTx)R−γ∥Σ1/2x∥2s.t.xTΣx≤t2
or equivalently based on the equation
r
^
T
x
−
e
T
x
R
=
μ
^
T
x
\hat{r}^Tx-e^TxR=\hat{\mu}^Tx
r^Tx−eTxR=μ^Tx
max
x
μ
^
T
x
+
W
0
R
−
γ
∥
Σ
1
/
2
x
∥
2
s
.
t
.
x
T
Σ
x
≤
t
2
\max_x\hat{\mu}^Tx+W_0R-\gamma\lVert\Sigma^{1/2}x\rVert_2\\ s.t.\quad x^T\Sigma x\leq t^2
xmaxμ^Tx+W0R−γ∥Σ1/2x∥2s.t.xTΣx≤t2
Proposition 2: If
H
≥
γ
H\geq \gamma
H≥γ then RMVP2
admits the unique optimal solution
x
∗
=
t
H
Σ
−
1
μ
^
x^*=\frac{t}{H}\Sigma^{-1}\hat{\mu}
x∗=HtΣ−1μ^
If
H
<
γ
H<\gamma
H<γ then it is optimal for an RMVP2
investor to keep all initial wealth in the riskless asset.
Proof: the Lagrange function using a non-negative multiplier
λ
\lambda
λ:
L
(
x
,
λ
)
=
μ
^
T
x
+
W
0
R
−
γ
∥
Σ
1
/
2
x
∥
2
+
λ
(
t
2
−
x
T
Σ
x
)
L(x, \lambda)=\hat{\mu}^Tx+W_0R-\gamma\lVert\Sigma^{1/2}x\rVert_2+\lambda(t^2-x^T\Sigma x)
L(x,λ)=μ^Tx+W0R−γ∥Σ1/2x∥2+λ(t2−xTΣx)
the optimal solution
x
∗
x^*
x∗ can be obtained according to the first-order condition and Slater condition
x
∗
=
(
σ
2
σ
λ
+
γ
)
Σ
−
1
μ
^
x^*=(\frac{\sigma}{2\sigma\lambda+\gamma})\Sigma^{-1}\hat{\mu}
x∗=(2σλ+γσ)Σ−1μ^
The definition of
σ
\sigma
σ reveal that
σ
2
=
σ
2
(
2
σ
λ
+
γ
)
2
H
2
\sigma^2=\frac{\sigma^2}{(2\sigma\lambda+\gamma)^2}H^2
σ2=(2σλ+γ)2σ2H2
Assuming the constraint to be active and observing that
σ
=
t
\sigma=t
σ=t, we have
λ
=
H
−
γ
2
t
\lambda=\frac{H-\gamma}{2t}
λ=2tH−γ
which is non-negative provided that
H
≥
γ
H\geq \gamma
H≥γ. If
H
<
γ
H<\gamma
H<γ then the only feasible choice for
λ
\lambda
λ is zero along with a dual objective function value equal to
W
0
R
W_0R
W0R which is attained in the primal by a riskless portfolio, i.e.
x
∗
=
0
x^*=0
x∗=0.
The robust MV portfolio is identical to the MV portfolio obtained as a solution to the problem
max
x
μ
^
T
x
+
W
0
R
s
.
t
.
x
T
Σ
x
≤
t
2
\max_x\hat{\mu}^Tx+W_0R\\ s.t.\quad x^T\Sigma x\leq t^2
xmaxμ^Tx+W0Rs.t.xTΣx≤t2
That is, the investor maximizing robust expected return under a variance constraint makes a MV portfolio choice when his/her confidence in the estimation of the mean is high, i.e.
γ
\gamma
γ is smaller than
H
H
H.(这种情况下对均值的置信程度很高).
Corollary 1:
Let
H
≥
γ
H\geq \gamma
H≥γ.Then
- choosing a maximum variance
t
=
T
−
W
0
R
H
−
γ
t=\frac{T-W_0R}{H-\gamma}
t=H−γT−W0R, the
RMVP2
investor holds an optimal portfolio identical to theRMVP1
investor with a target wealth equal to T T T. - choosing a minimum target wealth equal to
T
=
W
0
R
+
t
(
H
−
γ
)
T=W_0R+t(H-\gamma)
T=W0R+t(H−γ) the
RMVP1
investor holds an optimal portfolio identical to theRMVP2
investor with a variance cap equal to t 2 t^2 t2.
Consider the following problem
max
x
min
r
∈
U
r
r
T
x
+
(
W
0
−
e
T
x
)
R
−
ρ
2
x
T
Σ
x
\max_x\min_{r\in U_r}r^Tx+(W_0-e^Tx)R-\frac{\rho}{2}x^T\Sigma x
xmaxr∈UrminrTx+(W0−eTx)R−2ρxTΣx
where
ρ
\rho
ρ is a positive scalar. Expanding the inner
m
i
n
min
min problem from the point of worst-case
, we have
max
x
r
^
T
x
+
(
W
0
−
e
T
x
)
R
−
γ
∥
Σ
1
/
2
x
∥
2
\max_x \hat{r}^Tx+(W_0-e^Tx)R-\gamma\lVert\Sigma^{1/2}x\rVert_2
xmaxr^Tx+(W0−eTx)R−γ∥Σ1/2x∥2
which refer to as RMVP3
.
Proposition 3: If
γ
ρ
<
H
\gamma\rho<H
γρ<H then RMVP3
admits the unique optimal solution
x
∗
=
(
H
−
γ
ρ
ρ
H
)
Σ
−
1
μ
^
x^*=(\frac{H-\gamma\rho}{\rho H})\Sigma^{-1}\hat{\mu}
x∗=(ρHH−γρ)Σ−1μ^
If
γ
ρ
>
H
\gamma\rho>H
γρ>H then it is optimal for RMVP3
investor to keep all initial wealth in the riskless asset.
Proof: The function is strictly concave.(RMVP3问题是非凸优化), the first-order necessary conditions yield the candidate solution:
x
=
(
σ
γ
+
σ
ρ
)
Σ
−
1
μ
^
x=(\frac{\sigma}{\gamma+\sigma\rho})\Sigma^{-1}\hat{\mu}
x=(γ+σρσ)Σ−1μ^
Developing the right-hand side, we obtain a quadratic equation
ρ
2
σ
2
+
2
γ
ρ
σ
+
γ
2
−
H
2
=
0
\rho^2\sigma^2+2\gamma\rho\sigma+\gamma^2-H^2=0
ρ2σ2+2γρσ+γ2−H2=0
with the positive root
σ
+
=
H
−
γ
ρ
ρ
2
\sigma_+=\frac{H-\gamma\rho}{\rho^2}
σ+=ρ2H−γρ
provided that
γ
ρ
<
H
\gamma\rho<H
γρ<H.