Navigator
Pre Blog Link
Robust multi-period portfolio selection(2)
Robust conterparts of multi-period portfolio problems
本节主要分析在非对称不确定集合
F
Ω
\mathcal{F}_\Omega
FΩ下,鲁棒投资组合模型
(
2.14
)
(2.14)
(2.14). 为此(To this end
),我们假设风险资产
i
i
i的
r
~
i
t
,
t
=
1
,
2
,
…
,
T
\tilde{r}_i^t, t=1,2,\dots, T
r~it,t=1,2,…,T由
m
m
m个随机因子
ξ
t
~
=
(
ξ
~
1
t
,
…
,
ξ
~
m
t
)
′
∈
R
m
\tilde{\pmb{\xi}^t}=(\tilde{\xi}_1^t, \dots, \tilde{\xi}_m^t)'\in\mathbb{R}^m
ξξξt~=(ξ~1t,…,ξ~mt)′∈Rm驱动,收益率向量
r
t
~
\tilde{\pmb{r}^t}
rrrt~可以表示为这些因子的线性组合.
从方程组
(
2.7
)
(2.7)
(2.7)可以知道,由于
r
0
\pmb{r}^0
rrr0是已知,可以得到收益率向量表达式如下
R
~
t
+
1
=
μ
t
+
1
+
A
t
+
1
ξ
t
~
,
t
=
1
,
…
,
T
(4.1)
\tilde{\pmb{R}}^{t+1}=\pmb{\mu}^{t+1}+\pmb{A}^{t+1}\tilde{\pmb{\xi}^t}, t=1,\dots, T\tag{4.1}
RRR~t+1=μμμt+1+AAAt+1ξξξt~,t=1,…,T(4.1)
其中
μ
t
+
1
\pmb{\mu}^{t+1}
μμμt+1为常数,
A
t
+
1
∈
R
n
×
m
\pmb{A}^{t+1}\in\mathbb{R}^{n\times m}
AAAt+1∈Rn×m是因子载荷矩阵,为了简化模型,设置
ξ
~
t
\tilde{\pmb{\xi}}^t
ξξξ~t为序列无关,对于每阶段
t
t
t满足标准条件
(
3.6
)
(3.6)
(3.6),且具有支撑集
S
t
∈
[
−
l
t
,
u
t
]
\mathbb{S}^t\in[-\pmb{l}^t, \pmb{u}^t]
St∈[−lllt,uuut]. 可以得到随机向量的期望为
E
[
R
~
t
+
1
]
=
μ
t
+
1
,
t
=
1
,
…
,
T
(4.2)
\mathbb{E}[\tilde{\pmb{R}}^{t+1}]=\pmb{\mu}^{t+1}, t=1,\dots, T\tag{4.2}
E[RRR~t+1]=μμμt+1,t=1,…,T(4.2)
可以将非对称不确定集合
F
Ω
\mathcal{F}_\Omega
FΩ扩展到多阶段的情况
F
Ω
t
=
{
ξ
t
:
∃
u
‾
t
,
u
‾
t
∈
R
+
m
,
ξ
t
=
u
‾
t
−
u
‾
t
,
∥
(
P
)
−
1
u
‾
+
(
Q
)
−
1
u
‾
∥
2
≤
Ω
t
,
ξ
t
∈
[
−
l
t
,
u
t
]
}
(4.3)
\mathcal{F}_\Omega^t=\{\pmb{\xi}^t:\exist \underline{\pmb{u}}^t, \overline{\pmb{u}}^t\in\mathbb{R}_+^m, \pmb{\xi}^t=\overline{\pmb{u}}^t-\underline{\pmb{u}}^t, \lVert(\pmb{P})^{-1}\overline{\pmb{u}}+(\pmb{Q})^{-1}\underline{\pmb{u}}\rVert_2\leq\Omega^t, \pmb{\xi}^t\in[-\pmb{l}^t, \pmb{u}^t]\}\tag{4.3}
FΩt={ξξξt:∃uuut,uuut∈R+m,ξξξt=uuut−uuut,∥(PPP)−1uuu+(QQQ)−1uuu∥2≤Ωt,ξξξt∈[−lllt,uuut]}(4.3)
其中,
P
t
=
d
i
a
g
(
p
1
t
,
…
,
p
m
t
)
\pmb{P}^t=diag(p_1^t, \dots, p_m^t)
PPPt=diag(p1t,…,pmt)且
Q
t
=
d
i
a
g
(
q
1
t
,
…
,
q
m
t
)
\pmb{Q}^t=diag(q_1^t, \dots, q_m^t)
QQQt=diag(q1t,…,qmt),矩阵元素满足
{
p
i
t
=
p
(
ξ
~
i
t
)
>
0
q
i
t
=
q
(
ξ
~
i
t
)
>
0
\begin{cases} p_i^t=p(\tilde{\xi}_i^t)>0\\ q_i^t=q(\tilde{\xi}_i^t)>0 \end{cases}
{pit=p(ξ~it)>0qit=q(ξ~it)>0
表示随机变量
ξ
i
t
\xi_i^t
ξit的forward
和backward
偏离值.
In the following, what we are interested in is the robust counterpart of ( 2.14 ) (2.14) (2.14) when the random factors ξ t ~ \tilde{\pmb{\xi}^t} ξξξt~ vary in the uncertainty set F Ω t \mathcal{F}_\Omega^t FΩt. To this end, we introduce some notation.
记
h
0
t
(
y
t
,
z
t
+
1
)
=
w
0
t
−
w
0
t
+
1
+
(
1
−
θ
R
0
t
+
1
z
s
,
t
+
1
−
1
+
θ
R
0
t
+
1
z
b
,
t
+
1
)
μ
t
+
1
t
=
1
,
…
,
T
−
1
(4.4)
h_0^t(\pmb{y}^t, \pmb{z}^{t+1})=w_0^t-w_0^{t+1}+\bigg(\frac{1-\theta}{R_0^{t+1}}\pmb{z}^{s, t+1}-\frac{1+\theta}{R_0^{t+1}}\pmb{z}^{b, t+1}\bigg)\pmb{\mu}^{t+1}\\ t=1, \dots, T-1\tag{4.4}
h0t(yyyt,zzzt+1)=w0t−w0t+1+(R0t+11−θzzzs,t+1−R0t+11+θzzzb,t+1)μμμt+1t=1,…,T−1(4.4)
以及
h
j
t
(
y
t
,
z
t
+
1
)
=
e
j
′
(
A
t
+
1
)
′
(
1
−
θ
R
0
t
+
1
z
s
,
t
+
1
−
1
−
θ
R
0
t
+
1
z
b
,
t
+
1
)
j
=
1
,
…
,
m
t
=
1
,
…
,
T
−
1
(4.5)
h_j^t(\pmb{y}^t, \pmb{z}^{t+1})=\pmb{e}_j'(\pmb{A}^{t+1})'\bigg(\frac{1-\theta}{R_0^{t+1}}\pmb{z}^{s, t+1}-\frac{1-\theta}{R_0^{t+1}}\pmb{z}^{b, t+1}\bigg)\\ j=1,\dots, m \quad t=1,\dots, T-1\tag{4.5}
hjt(yyyt,zzzt+1)=eeej′(AAAt+1)′(R0t+11−θzzzs,t+1−R0t+11−θzzzb,t+1)j=1,…,mt=1,…,T−1(4.5)
其中
y
t
=
(
w
0
t
;
w
t
)
∈
R
n
+
1
,
z
t
+
1
=
(
z
b
,
t
+
1
,
z
s
,
t
+
1
)
∈
R
2
m
\pmb{y}^t=(w_0^t; \pmb{w}^t)\in\mathbb{R}^{n+1}, \pmb{z}^{t+1}=(\pmb{z}^{b, t+1}, \pmb{z}^{s, t+1})\in\mathbb{R}^{2m}
yyyt=(w0t;wwwt)∈Rn+1,zzzt+1=(zzzb,t+1,zzzs,t+1)∈R2m,并且
e
j
\pmb{e}_j
eeej是一个
m
m
m维向量,其中位置
j
j
j的元素为1,其他的位置为0,根据
(
4.4
)
(4.4)
(4.4)和
(
4.5
)
(4.5)
(4.5)重写随机不等式约束
(
2.13
)
(2.13)
(2.13)
可以得到
w
0
t
−
w
0
t
+
1
+
1
−
θ
R
0
t
+
1
(
R
~
t
+
1
)
′
z
s
,
t
+
1
−
1
+
θ
R
0
t
+
1
(
R
~
t
+
1
)
′
z
b
,
t
+
1
≥
0
(4.6)
w_0^t-w_0^{t+1}+\frac{1-\theta}{R_0^{t+1}}(\tilde{\pmb{R}}^{t+1})'\pmb{z}^{s, t+1}-\frac{1+\theta}{R_0^{t+1}}(\tilde{\pmb{R}}^{t+1})'\pmb{z}^{b, t+1}\geq 0\tag{4.6}
w0t−w0t+1+R0t+11−θ(RRR~t+1)′zzzs,t+1−R0t+11+θ(RRR~t+1)′zzzb,t+1≥0(4.6)
将
(
4.1
)
(4.1)
(4.1)代入到随机约束
(
4.6
)
(4.6)
(4.6)中,由
(
4.1
)
∼
(
4.5
)
(4.1)\sim (4.5)
(4.1)∼(4.5)随机约束等效于如下关于
(
y
t
;
z
t
+
1
)
(\pmb{y}^t; \pmb{z}^{t+1})
(yyyt;zzzt+1)和
(
ξ
~
t
)
(\tilde{\pmb{\xi}}^t)
(ξξξ~t)的双线性方程(bilinear function
)
G
t
(
y
t
,
z
t
+
1
;
ξ
t
~
)
=
h
0
t
(
y
t
,
z
t
+
1
)
+
∑
j
=
1
m
ξ
~
j
t
h
j
t
(
y
t
,
z
t
+
1
)
≥
0
t
=
1
,
…
,
T
−
1
(4.7)
G^t(\pmb{y}^t, \pmb{z}^{t+1}; \tilde{\pmb{\xi}^t})=h_0^t(\pmb{y}^t, \pmb{z}^{t+1})+\sum_{j=1}^m\tilde{\xi}_j^th_j^t(\pmb{y}^t, \pmb{z}^{t+1})\geq 0\\ t=1, \dots, T-1\tag{4.7}
Gt(yyyt,zzzt+1;ξξξt~)=h0t(yyyt,zzzt+1)+j=1∑mξ~jthjt(yyyt,zzzt+1)≥0t=1,…,T−1(4.7)
进一步
g
(
y
T
,
z
T
+
1
,
ξ
~
t
)
=
w
0
T
R
0
T
+
1
+
(
R
~
T
+
1
)
′
w
T
(4.8)
g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\pmb{\xi}}^t)=w_0^TR_0^{T+1}+(\tilde{\pmb{R}}^{T+1})'\pmb{w}^T\tag{4.8}
g(yyyT,zzzT+1,ξξξ~t)=w0TR0T+1+(RRR~T+1)′wwwT(4.8)
和
h
0
T
(
y
T
,
z
T
+
1
)
=
w
0
T
R
0
T
+
1
+
(
μ
T
+
1
)
′
w
T
(4.9)
h_0^T(\pmb{y}^T, \pmb{z}^{T+1})=w_0^TR_0^{T+1}+(\pmb{\mu}^{T+1})'\pmb{w}^T\tag{4.9}
h0T(yyyT,zzzT+1)=w0TR0T+1+(μμμT+1)′wwwT(4.9)
和
h
j
T
(
y
T
,
z
T
+
1
)
=
e
j
′
(
A
T
+
1
)
′
w
T
j
=
1
,
…
,
m
(4.10)
h_j^T(\pmb{y}^T, \pmb{z}^{T+1})=\pmb{e}_j'(\pmb{A}^{T+1})'\pmb{w}^T\\ j=1,\dots, m\tag{4.10}
hjT(yyyT,zzzT+1)=eeej′(AAAT+1)′wwwTj=1,…,m(4.10)
由
(
4.1
)
(4.1)
(4.1)可以得到
g
(
y
T
,
z
T
+
1
,
ξ
~
)
=
w
0
T
R
0
T
+
1
+
(
R
~
T
+
1
)
′
w
T
=
分离出非随机部分
(
w
0
T
R
0
T
+
1
+
(
μ
T
+
1
)
′
w
T
)
+
∑
i
=
1
m
ξ
~
j
(
e
j
′
(
A
T
+
1
)
′
w
T
)
=
代入4.9和4.10
h
0
T
(
y
T
,
z
T
+
1
)
+
∑
i
=
1
m
ξ
~
i
h
j
T
(
y
T
,
z
T
+
1
)
(4.11)
\begin{aligned} g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\pmb{\xi}})&=w_0^TR_0^{T+1}+(\tilde{\pmb{R}}^{T+1})'\pmb{w}^T\\ &\xlongequal{\text{分离出非随机部分}}(w_0^TR_0^{T+1}+(\pmb{\mu}^{T+1})'\pmb{w}^T)+\sum_{i=1}^m\tilde{\xi}_j(\pmb{e}_j'(\pmb{A}^{T+1})'\pmb{w}^T)\\ &\xlongequal{\text{代入4.9和4.10}}h_0^T(\pmb{y}^T, \pmb{z}^{T+1})+\sum_{i=1}^m\tilde{\xi}_ih_j^T(\pmb{y}^T, \pmb{z}^{T+1})\tag{4.11} \end{aligned}
g(yyyT,zzzT+1,ξξξ~)=w0TR0T+1+(RRR~T+1)′wwwT分离出非随机部分(w0TR0T+1+(μμμT+1)′wwwT)+i=1∑mξ~j(eeej′(AAAT+1)′wwwT)代入4.9和4.10h0T(yyyT,zzzT+1)+i=1∑mξ~ihjT(yyyT,zzzT+1)(4.11)
因此可以将原问题
(
2.14
)
(2.14)
(2.14)改写如下MLPM
min
(
y
t
,
z
t
+
1
)
∈
X
ρ
(
a
,
λ
;
g
(
y
T
,
z
T
+
1
,
ξ
t
~
)
+
)
=
−
E
[
g
(
y
T
,
z
T
+
1
,
ξ
~
t
)
]
+
λ
E
[
(
a
−
g
(
y
T
,
z
T
+
1
,
ξ
t
~
)
)
+
]
s
.
t
.
{
G
t
(
y
t
,
z
t
+
1
;
ξ
t
~
)
=
h
0
t
(
y
t
,
z
t
+
1
)
+
∑
j
=
1
m
ξ
j
t
~
h
j
t
(
y
t
,
z
t
+
1
)
≥
0
t
=
1
,
…
,
T
−
1
\begin{aligned} &\min_{(y^t, z^{t+1})\in\mathcal{X}} \rho(a, \lambda; g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\pmb{\xi}^t})_+)\\ &=-\mathbb{E}[g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\xi}^t)]+\lambda\mathbb{E}[(a-g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\pmb{\xi}^t}))_+]\\ &s.t.\begin{cases} G^t(\pmb{y}^t, \pmb{z}^{t+1}; \tilde{\pmb{\xi}^t})=h_0^t(\pmb{y}^t, \pmb{z}^{t+1})+\sum\limits_{j=1}^m\tilde{\xi_j^t}h_j^t(\pmb{y}^t, \pmb{z}^{t+1})\geq 0\\ t=1, \dots, T-1 \end{cases} \end{aligned}
(yt,zt+1)∈Xminρ(a,λ;g(yyyT,zzzT+1,ξξξt~)+)=−E[g(yyyT,zzzT+1,ξ~t)]+λE[(a−g(yyyT,zzzT+1,ξξξt~))+]s.t.⎩⎨⎧Gt(yyyt,zzzt+1;ξξξt~)=h0t(yyyt,zzzt+1)+j=1∑mξjt~hjt(yyyt,zzzt+1)≥0t=1,…,T−1
其中
y
t
=
(
w
0
t
;
w
t
)
∈
R
n
+
1
,
z
t
+
1
=
(
z
b
,
t
+
1
;
z
s
,
t
+
1
)
∈
R
2
n
\pmb{y}^t=(w_0^t; \pmb{w}^t)\in\mathbb{R}^{n+1}, \pmb{z}^{t+1}=(\pmb{z}^{b, t+1}; \pmb{z}^{s, t+1})\in\mathbb{R}^{2n}
yyyt=(w0t;wwwt)∈Rn+1,zzzt+1=(zzzb,t+1;zzzs,t+1)∈R2n,并且在确定性约束条件的可行集
X
\mathcal{X}
X为
X
=
{
(
y
t
,
z
t
+
1
)
:
R
0
1
w
0
1
+
∑
i
=
1
m
R
i
1
w
i
1
=
1
,
i
=
1
,
…
,
n
w
i
t
+
1
=
w
i
t
+
z
i
b
,
t
+
1
−
z
i
s
,
t
+
1
w
i
t
≥
0
z
i
b
,
t
≥
0
z
i
s
,
t
≥
0
}
\mathcal{X}= \left\{ (\pmb{y}^t, \pmb{z}^{t+1}):\quad \begin{aligned} &R_0^1w_0^1+\sum_{i=1}^mR_i^1w_i^1=1, i=1,\dots, n\\ &w_i^{t+1}=w_i^t+z_i^{b, t+1}-z_i^{s, t+1}\\ &w_i^t\geq 0\\ &z_i^{b, t}\geq 0\\ &z_i^{s, t}\geq 0 \end{aligned} \right\}
X=⎩⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎧(yyyt,zzzt+1):R01w01+i=1∑mRi1wi1=1,i=1,…,nwit+1=wit+zib,t+1−zis,t+1wit≥0zib,t≥0zis,t≥0⎭⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎫
定理4.1 模型
(
2.14
)
(2.14)
(2.14)(MLPM
)的目标函数具有紧的上界(tight upper bound
)
ρ
=
−
E
[
g
(
y
T
,
z
T
+
1
,
ξ
~
T
)
]
+
λ
E
[
(
a
−
g
(
y
T
,
z
T
+
1
,
ξ
T
~
)
)
+
]
≤
−
(
1
+
λ
)
h
0
T
(
y
T
,
z
T
+
1
)
+
λ
a
−
λ
(
γ
−
1
)
β
^
+
λ
(
−
β
^
)
+
+
λ
γ
(
(
h
0
T
(
y
T
,
z
T
+
1
)
−
a
)
+
max
ξ
T
∈
F
Ω
∑
j
=
1
m
h
j
T
(
y
T
,
z
T
+
1
)
ξ
j
T
)
(4.12)
\begin{aligned} \rho&=-\mathbb{E}[g(\pmb{y}^T, \pmb{z}^{T+1}, \tilde{\pmb{\xi}}^T)]+\lambda\mathbb{E}[(a-g(\pmb{y}^T, \pmb{z}^{T+1},\tilde{\pmb{\xi}^T}))_+]\\ &\leq -(1+\lambda)h_0^T(\pmb{y}^T, \pmb{z}^{T+1})+\lambda a-\lambda(\gamma-1)\hat{\beta}+\lambda(-\hat{\beta})_+\\ &+\lambda\gamma\bigg((h_0^T(\pmb{y}^T, \pmb{z}^{T+1})-a)+\max_{\xi^T\in\mathcal{F}_\Omega}\sum_{j=1}^mh_j^T(\pmb{y}^T, \pmb{z}^{T+1})\xi_j^T\bigg)\tag{4.12} \end{aligned}
ρ=−E[g(yyyT,zzzT+1,ξξξ~T)]+λE[(a−g(yyyT,zzzT+1,ξξξT~))+]≤−(1+λ)h0T(yyyT,zzzT+1)+λa−λ(γ−1)β^+λ(−β^)++λγ((h0T(yyyT,zzzT+1)−a)+ξT∈FΩmaxj=1∑mhjT(yyyT,zzzT+1)ξjT)(4.12)
其中方程参数定义如下
证明过程见论文的Online Appendix.
记
ψ
(
a
,
λ
;
g
(
y
T
,
z
T
+
1
,
ξ
T
)
)
\psi(a, \lambda; g(\pmb{y}^T, \pmb{z}^{T+1}, \pmb{\xi}^T))
ψ(a,λ;g(yyyT,zzzT+1,ξξξT))
表示方程
(
4.12
)
(4.12)
(4.12)的右端,即
ρ
\rho
ρ的紧的上界(tight upper bound
). 因此可以转换原问题
(
2.14
)
(2.14)
(2.14)的目标函数为
ψ
\psi
ψ. 从定理
4.1
4.1
4.1的证明过程中可以知道
max
ξ
T
∈
F
Ω
∑
j
=
1
m
h
j
T
(
y
T
,
z
T
+
1
)
ξ
j
T
=
min
d
T
{
Ω
∥
d
T
∥
:
P
−
1
d
T
≥
h
T
,
Q
−
1
d
h
≥
−
h
T
}
\max_{\xi^T\in\mathcal{F}_\Omega}\sum_{j=1}^mh_j^T(\pmb{y}^T, \pmb{z}^{T+1})\xi_j^T=\min_{d^T}\{\Omega\lVert \pmb{d}^T\rVert:\pmb{P}^{-1}\pmb{d}^T\geq \pmb{h}^T, \pmb{Q}^{-1}\pmb{d}^h\geq-\pmb{h}^T\}
ξT∈FΩmaxj=1∑mhjT(yyyT,zzzT+1)ξjT=dTmin{Ω∥dddT∥:PPP−1dddT≥hhhT,QQQ−1dddh≥−hhhT}
由
(
x
)
+
=
x
+
(
−
x
)
+
(x)_+=x+(-x)_+
(x)+=x+(−x)+
代入
ψ
\psi
ψ表达式可以得到
Next we return to the stochastic inequality constraints of problem (
MLPM
). With the uncertainty set F Ω \mathcal{F}_\Omega FΩ and 3.10 3.10 3.10,the robust counterpart of the stochastic inequality constraint is the determinate inequality.
G t ( y t , z t + 1 , ξ t ) ≥ 0 , ξ t ∈ F Ω t (4.13) G^t(\pmb{y}^t, \pmb{z}^{t+1}, \pmb{\xi}^t)\geq 0, \pmb{\xi}^t\in\mathcal{F}^t_{\Omega}\tag{4.13} Gt(yyyt,zzzt+1,ξξξt)≥0,ξξξt∈FΩt(4.13)
In summary, we have the robust counterpart of original problem, denoted by RLPM:
定理4.2 对于任意
λ
≥
1
,
ε
>
0
\lambda\geq 1, \varepsilon>0
λ≥1,ε>0,
ε
t
=
ε
/
(
T
−
1
)
,
Ω
=
Ω
t
=
−
2
ln
(
ε
t
)
,
γ
=
exp
(
−
Ω
2
/
2
)
\varepsilon_t=\varepsilon/(T-1),\Omega=\Omega_t=\sqrt{-2\ln(\varepsilon_t)},\gamma=\exp(-\Omega^2/2)
εt=ε/(T−1),Ω=Ωt=−2ln(εt),γ=exp(−Ω2/2),对于RLPM
的任意可行解
(
y
t
,
z
t
+
1
)
(\pmb{y}^t, \pmb{z}^{t+1})
(yyyt,zzzt+1),原问题的随机约束满足
作者对定理 ( 4.2 ) (4.2) (4.2)的评论如下(主要分析了其内在的保守性)
The impact of Theorem 4.2 is double. On the one hand, one can always take a desired confidence level ε \varepsilon ε such as ε = 0.05 \varepsilon=0.05 ε=0.05,such that the probability guarantee (4.16) holds. However, on the other hand, this will lead to a very conservative individual constraint, in other words, if ε = 0.05 \varepsilon=0.05 ε=0.05,then we must take ε t ≤ ε / ( T − 1 ) = 0.0005 ; P { G t ( y t , z t + 1 ; ξ t ) ≥ 0 } ≥ 0.9995 \varepsilon_t\leq \varepsilon/(T-1)=0.0005; \mathbb{P}\{G^t(\pmb{y}^t, \pmb{z}^{t+1};\pmb{\xi}^t)\geq 0\}\geq 0.9995 εt≤ε/(T−1)=0.0005;P{Gt(yyyt,zzzt+1;ξξξt)≥0}≥0.9995 for T = 101 T=101 T=101. Indeed, this is too conservative and is unpractical.
The relationship between RLPM and CVaR
定义关于投资组合 x ∈ X ⊂ R n + 1 \pmb{x}\in\mathcal{X}\subset \mathbb{R}^{n+1} xxx∈X⊂Rn+1和随机向量 ξ ~ ∈ R m \tilde{\pmb{\xi}}\in\mathbb{R}^m ξξξ~∈Rm的损失函数 f ( x , ξ ~ ) f(\pmb{x}, \tilde{\pmb{\xi}}) f(xxx,ξξξ~).
VaR
is defined as the minimal level β \beta β, such that the probability that the portfolio loss f ( x , ξ ~ ) f(\pmb{x}, \tilde{\pmb{\xi}}) f(xxx,ξξξ~) exceeds β \beta β is below α \alpha α, where α ∈ ( 0 , 1 ) \alpha\in(0, 1) α∈(0,1) is the probability level specified by the user (typically, α = 1 % \alpha=1\% α=1% or 5 % 5\% 5%).
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
=
min
{
β
∣
P
(
f
(
x
,
ξ
~
)
)
≤
α
}
VaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))=\min\{\beta\mid \mathbb{P}(f(\pmb{x}, \pmb{\tilde{\xi}}))\leq \alpha\}
VaR1−α(f(xxx,ξξξ~))=min{β∣P(f(xxx,ξ~ξ~ξ~))≤α}
CVaR
度量了在
f
(
x
,
ξ
~
)
≥
V
a
R
1
−
α
f(\pmb{x}, \tilde{\pmb{\xi}})\geq VaR_{1-\alpha}
f(xxx,ξξξ~)≥VaR1−α下的条件期望损失,数学上的定义可以参考Rockafellar, R & Uryasev (2000)1给出.
C
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
=
E
[
f
(
x
,
ξ
~
)
∣
f
(
x
,
ξ
~
)
≥
V
a
R
1
−
α
]
CVaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))=\mathbb{E}[f(\pmb{x}, \tilde{\pmb{\xi}})\mid f(\pmb{x}, \tilde{\pmb{\xi}})\geq VaR_{1-\alpha}]
CVaR1−α(f(xxx,ξξξ~))=E[f(xxx,ξξξ~)∣f(xxx,ξξξ~)≥VaR1−α]
对于给定的
α
\alpha
α,如果设置LPM中的目标a
的值为
a
=
1
−
V
a
R
1
−
α
a=1-VaR_{1-\alpha}
a=1−VaR1−α,并且
g
(
x
,
ξ
~
)
=
1
−
f
(
x
,
ξ
~
)
g(\pmb{x}, \tilde{\pmb{\xi}})=1-f(\pmb{x}, \tilde{\pmb{\xi}})
g(xxx,ξξξ~)=1−f(xxx,ξξξ~),那么可以得到
L
P
M
(
a
;
g
(
x
,
ξ
~
)
)
=
E
[
a
−
g
(
x
,
ξ
~
)
]
+
=
E
[
a
−
g
(
x
,
ξ
~
)
∣
a
−
g
(
x
,
ξ
~
)
≥
0
]
=
E
[
f
(
x
,
ξ
~
)
−
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
∣
f
(
x
,
ξ
~
)
≥
V
a
R
1
−
α
]
\begin{aligned} LPM(a; g(x, \tilde{\xi}))&=\mathbb{E}[a-g(\pmb{x}, \tilde{\pmb{\xi}})]_+=\mathbb{E}[a-g(\pmb{x}, \tilde{\pmb{\xi}})\mid a-g(\pmb{x}, \tilde{\pmb{\xi}})\geq 0]\\ &=\mathbb{E}[f(\pmb{x}, \tilde{\pmb{\xi}})-VaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))\mid f(\pmb{x}, \tilde{\pmb{\xi}})\geq VaR_{1-\alpha}] \end{aligned}
LPM(a;g(x,ξ~))=E[a−g(xxx,ξξξ~)]+=E[a−g(xxx,ξξξ~)∣a−g(xxx,ξξξ~)≥0]=E[f(xxx,ξξξ~)−VaR1−α(f(xxx,ξξξ~))∣f(xxx,ξξξ~)≥VaR1−α]
如果
α
\alpha
α的取值使得
V
a
R
1
−
α
VaR_{1-\alpha}
VaR1−α为非负值,那么可以得到
L
P
M
(
a
;
g
(
x
,
ξ
~
)
)
≤
C
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
LPM(a; g(\pmb{x}, \tilde{\pmb{\xi}}))\leq CVaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))
LPM(a;g(xxx,ξξξ~))≤CVaR1−α(f(xxx,ξξξ~))
当
α
\alpha
α和
ξ
~
\tilde{\xi}
ξ~的分布为给定时,即
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
VaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))
VaR1−α(f(xxx,ξξξ~))为确定值,可以得到
L
P
M
(
a
;
g
(
x
,
ξ
~
)
)
=
C
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
−
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
LPM(a; g(\pmb{x}, \tilde{\pmb{\xi}}))=CVaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))-VaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))
LPM(a;g(xxx,ξξξ~))=CVaR1−α(f(xxx,ξξξ~))−VaR1−α(f(xxx,ξξξ~))
In this case, the minimization of
CVaR
is equivalent to the minimization of the sum ofLPM
andVaR
. Thus, the differences betweenmean-CVaR
andmean-LPM
are clear.
First
, in the minimization ofCVaR
problem, we can understand that the target V a R 1 − α ( f ( x , ξ ~ ) ) VaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}})) VaR1−α(f(xxx,ξξξ~)) is in fact endogenous. But in LPM, the target a a a is exogenous.
Second
, the advantage of exogenous a a a can avoid the effect of uncertainty set in robust portfolio choice which can reduce a chance constraint in multi-period case.
根据CVaR
的一致性,可以得到
min
x
C
V
a
R
1
−
α
(
f
(
x
,
ξ
~
)
)
=
min
x
,
γ
0
{
γ
0
∣
C
V
a
R
1
−
α
f
(
x
,
ξ
~
)
+
γ
0
≤
0
}
(4.17)
\min_x CVaR_{1-\alpha}(f(\pmb{x}, \tilde{\pmb{\xi}}))=\min_{x, \gamma_0}\bigg\{ \gamma_0\mid CVaR_{1-\alpha}f(\pmb{x}, \tilde{\pmb{\xi}})+\gamma_0\leq 0\bigg\}\tag{4.17}
xminCVaR1−α(f(xxx,ξξξ~))=x,γ0min{γ0∣CVaR1−αf(xxx,ξξξ~)+γ0≤0}(4.17)
The right hand side of problem ( 4.17 ) (4.17) (4.17) includes a
CVaR
constraint which is still difficult to be expressed explicitly when the distribution information ξ ~ \tilde{\pmb{\xi}} ξξξ~ is only given by ( 3.7 ) (3.7) (3.7)
可以将CVaR
约束看做机会约束的近似
C
V
a
R
1
−
ε
(
f
(
x
,
ξ
~
)
+
γ
0
)
≤
0
→
P
{
f
(
x
,
ξ
~
)
+
γ
0
≥
0
}
≥
1
−
ε
(4.18)
CVaR_{1-\varepsilon}(f(\pmb{x}, \tilde{\pmb{\xi}})+\gamma_0)\leq 0\to\mathbb{P}\bigg\{ f(\pmb{x}, \tilde{\pmb{\xi}})+\gamma_0\geq 0\bigg\}\geq 1-\varepsilon\tag{4.18}
CVaR1−ε(f(xxx,ξξξ~)+γ0)≤0→P{f(xxx,ξξξ~)+γ0≥0}≥1−ε(4.18)
但是这并不是充分条件.
As stated in Natarajan et al. (2008) to solve the minimization of CVaR under uncertainty set F Ω \mathcal{F}_\Omega FΩ, it needs to add a new constraint (
the right hand side of (4.18)
) , which will lead to a more conservative solution from ( 4.2 ) (4.2) (4.2) than LPM.
Third, when the exogenous target a is taken not too extreme value
, it can help LPM reduce the variance of portfolio. For example, ifa
is close to the mean return of portfolio, then LPM in fact controls the down-side semi-variance which clearly is important to limit the variance. (如果目标值a与投资组合的收益均值接近,那么LPM实际上控制了下行方差(半方差)的风险).
可以由VaR
给出外生变量a
的参考值. 令投资组合的收益率
r
~
p
(
ξ
~
)
\tilde{r}_p(\tilde{\pmb{\xi}})
r~p(ξξξ~)服从正态分布. 令
a
=
1
−
V
a
R
a=1-VaR
a=1−VaR2,可以得到
a
=
1
−
Z
1
−
α
σ
r
~
p
(4.19)
a=1-\mathcal{Z}_{1-\alpha}\sigma_{\tilde{r}_p}\tag{4.19}
a=1−Z1−ασr~p(4.19)
在
normal distribution
下可以计算VaR
值为 V a R = Z 1 − α σ r ~ p + μ r ~ p VaR=\mathcal{Z}_{1-\alpha}\sigma_{\tilde{r}_p}+\mu_{\tilde{r}_p} VaR=Z1−ασr~p+μr~p,本文的数值实验中,使用Hull (2015)的结果,他在计算过程中忽略了期望收益 μ r ~ p \mu_{\tilde{r}_p} μr~p,因为和波动率相比,期望收益的值太小了. ↩︎