这学期会时不时更新一下伊曼纽尔·德曼(Emanuel Derman) 教授与迈克尔B.米勒(Michael B. Miller)的《The Volatility Smile》这本书,本意是协助导师课程需要,发在这里有意的朋友们可以学习一下,思路不一定够清晰且由于分工原因我是从书本第13章写起,还请大家见谅。
第20章 一些随机波动率模型的近似解
局部波动率模型的扩展
在局部波动率模型中,波动率是由股票价格决定的函数,而不是一个独立的随机变量。本节将从局部波动率和微笑曲线斜度出发,然后令微笑曲线的斜度本身服从随机流程
为了进一步解释该方法的原理,我们在 Hagan 等人开发的SABR模型(2002)基础上,建立一个简单的关于股票价格及其波动率的参数模型。具体而言,假设股票价格
S
S
S 的变动服从下列关系式:
d
S
S
=
α
S
β
−
1
d
W
d
α
=
ξ
α
d
Z
d
Z
d
W
=
ρ
d
t
\frac{dS}{S}=\alpha S^{\beta-1}dW\\ d\alpha=\xi\alpha dZ\\ dZdW=\rho dt
SdS=αSβ−1dWdα=ξαdZdZdW=ρdt
这里
W
,
Z
W,Z
W,Z 表示标准算术布朗运动,
ρ
\rho
ρ 表示二者之间的相关系数。
S
S
S 对数回报的波动率为
α
S
β
−
1
\alpha S^{\beta-1}
αSβ−1,
ξ
\xi
ξ 表示
α
\alpha
α 的波动率。
β
\beta
β 是一个模型参数,取值范围是0到1。当
ξ
=
0
,
β
=
1
\xi=0,\beta=1
ξ=0,β=1 时,模型就是一个普通的几何布朗运动,微笑曲线没有斜度。当
0
≤
β
<
1
,
ξ
=
0
0\leq\beta<1,\xi=0
0≤β<1,ξ=0 时,模型就变成了一个简单的局部波动率模型,微笑曲线有斜度
首先大致看一下这个模型的作用原理,先假设
ρ
=
0
\rho=0
ρ=0,
β
\beta
β 接近于1但是小于1,于是
1
−
β
1-\beta
1−β 就是一个非常小的正数,这会稍微偏离标准 BSM模型的设定条件。如果
ξ
=
0
\xi=0
ξ=0,根据此前对于局部波动率的研究,我们知道隐含波动率约等于当前股票价格和行权价格之间的局部波动率的平均值。因此,当
ξ
=
0
\xi=0
ξ=0 时,隐含波动率
Σ
L
V
\Sigma_{LV}
ΣLV 约等于:
Σ
L
V
(
S
,
t
,
K
,
T
,
α
,
β
)
≈
1
2
(
α
S
β
−
1
+
α
K
β
−
1
)
≈
α
S
β
−
1
1
2
[
1
+
(
K
S
)
β
−
1
]
\Sigma_{LV}(S,t,K,T,\alpha,\beta)\approx\frac{1}{2}(\alpha S^{\beta-1}+\alpha K^{\beta-1})\approx\alpha S^{\beta-1}\frac{1}{2}[1+(\frac{K}{S})^{\beta-1}]
ΣLV(S,t,K,T,α,β)≈21(αSβ−1+αKβ−1)≈αSβ−121[1+(SK)β−1]
β
=
1
\beta=1
β=1 时,对于上式右侧方括号中的第2项,可以用一阶泰勒展开式求得其近似表达式:
(
K
S
)
β
−
1
=
e
(
β
−
1
)
ln
(
K
S
)
≈
1
+
(
β
−
1
)
ln
(
K
S
)
(\frac{K}{S})^{\beta-1}=e^{(\beta-1)\ln(\frac{K}{S})}\approx1+(\beta-1)\ln(\frac{K}{S})
(SK)β−1=e(β−1)ln(SK)≈1+(β−1)ln(SK)
代入得:
Σ
L
V
(
S
,
t
,
K
,
T
,
α
,
β
)
≈
α
S
β
−
1
1
2
[
1
+
1
+
(
β
−
1
)
ln
(
K
S
)
]
≈
α
S
1
−
β
[
1
−
1
−
β
2
ln
(
K
S
)
]
\Sigma_{LV}(S,t,K,T,\alpha,\beta)\approx\alpha S^{\beta-1}\frac{1}{2}[1+1+(\beta-1)\ln(\frac{K}{S})]\\\approx\frac{\alpha}{S^{1-\beta}}[1-\frac{1-\beta}{2}\ln(\frac{K}{S})]
ΣLV(S,t,K,T,α,β)≈αSβ−121[1+1+(β−1)ln(SK)]≈S1−βα[1−21−βln(SK)]
根据上式,微笑曲线的斜度是一个与
ln
(
K
/
S
)
\ln(K/S)
ln(K/S) 相关的线性函数。由于
1
−
β
1-\beta
1−β 是正数,斜度就是负数(随着
K
K
K 增大,
Σ
\Sigma
Σ 下降),并且如果股价突然大幅下跌,平值期权的隐含波动率也会增加。我们知道,对于平值期权而言,
∂
Σ
/
∂
K
≈
∂
Σ
/
∂
S
\partial\Sigma/\partial K\approx\partial\Sigma/\partial S
∂Σ/∂K≈∂Σ/∂S,这跟任何一个局部波动率模型都是一致的
现在令
ξ
\xi
ξ 很小但是不等于0,于是上式中的随机波动率
α
\alpha
α 就成了一个随时间变动的参数,导致斜度也成了一个随机变量。由于
α
\alpha
α 的变动导致了期权的BSM价格在一定范围内变动,针对所有可能的
α
\alpha
α 值对应的期权价格,我们可以通过加权平均的方法计算得到看涨期权的价值,用
f
(
α
)
f(\alpha)
f(α) 表示
α
\alpha
α 的密度函数,我们可以得到:
C
S
L
V
≈
∫
C
B
S
M
(
Σ
L
V
(
S
,
t
,
K
,
T
,
α
,
β
)
)
f
(
α
)
d
α
C_{SLV}\approx\int C_{BSM}(\Sigma_{LV}(S,t,K,T,\alpha,\beta))f(\alpha)d\alpha
CSLV≈∫CBSM(ΣLV(S,t,K,T,α,β))f(α)dα
可以将
α
\alpha
α 分布的均值
α
ˉ
\bar\alpha
αˉ 代入得:
C
S
L
V
≈
∫
C
B
S
M
(
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
+
(
α
−
α
ˉ
)
,
β
)
)
f
(
α
)
d
α
≈
∫
[
C
B
S
M
(
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
,
β
)
)
+
∂
C
B
S
M
∂
α
∣
α
ˉ
(
α
−
α
ˉ
)
+
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
(
α
−
α
ˉ
)
2
]
f
(
α
)
d
α
≈
C
B
S
M
(
α
ˉ
)
+
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
C_{SLV}\approx\int C_{BSM}(\Sigma_{LV}(S,t,K,T,\bar\alpha+(\alpha-\bar\alpha),\beta))f(\alpha)d\alpha\\\approx\int[C_{BSM}(\Sigma_{LV}(S,t,K,T,\bar\alpha,\beta))+\frac{\partial C_{BSM}}{\partial\alpha}|_{\bar\alpha}(\alpha-\bar\alpha)+\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}(\alpha-\bar\alpha)^2]f(\alpha)d\alpha\\\approx C_{BSM}(\bar\alpha)+\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)
CSLV≈∫CBSM(ΣLV(S,t,K,T,αˉ+(α−αˉ),β))f(α)dα≈∫[CBSM(ΣLV(S,t,K,T,αˉ,β))+∂α∂CBSM∣αˉ(α−αˉ)+21∂α2∂2CBSM∣αˉ(α−αˉ)2]f(α)dα≈CBSM(αˉ)+21∂α2∂2CBSM∣αˉvar(α)
在这个随机局部波动率模型中,我们用
Σ
S
L
V
\Sigma_{SLV}
ΣSLV 表示隐含波动率,跟 BSM 波动率一样,这个隐含波动率就是使该模型期权价值等于 BSM 期权价值的波动率,于是:
C
S
L
V
≡
C
B
S
M
(
Σ
S
L
V
)
C_{SLV}\equiv C_{BSM}(\Sigma_{SLV})
CSLV≡CBSM(ΣSLV)
我们已经假设波动率的波动率值很小,
α
\alpha
α 值接近于
α
ˉ
\bar\alpha
αˉ 值。因此当
α
=
α
ˉ
\alpha=\bar\alpha
α=αˉ 时,
Σ
S
L
V
\Sigma_{{SLV}}
ΣSLV 与局部波动率
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
,
β
)
\Sigma_{LV}(S,t,K,T,\bar\alpha,\beta)
ΣLV(S,t,K,T,αˉ,β) 的差异会很小。于是可以将隐含波动率
Σ
S
L
V
\Sigma_{SLV}
ΣSLV 表示为
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
,
β
)
\Sigma_{LV}(S,t,K,T,\bar\alpha,\beta)
ΣLV(S,t,K,T,αˉ,β) 再加上一个数值很小的调整项,这个调整项是由随机变量
α
\alpha
α 产生的。因此有:
Σ
S
L
V
≡
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
,
β
)
+
(
Σ
S
L
V
−
Σ
L
V
(
S
,
t
,
K
,
T
,
α
ˉ
,
β
)
)
C
S
L
V
=
C
B
S
M
(
Σ
L
V
(
α
ˉ
)
+
(
Σ
S
L
V
−
Σ
L
V
(
α
ˉ
)
)
)
≈
C
B
S
M
(
α
ˉ
)
+
∂
C
B
S
M
∂
Σ
L
V
(
Σ
S
L
V
−
Σ
L
V
(
α
ˉ
)
)
\Sigma_{{SLV}}\equiv\Sigma_{LV}(S,t,K,T,\bar\alpha,\beta)+(\Sigma_{{SLV}}-\Sigma_{LV}(S,t,K,T,\bar\alpha,\beta))\\ C_{SLV}=C_{BSM}(\Sigma_{LV}(\bar\alpha)+(\Sigma_{{SLV}}-\Sigma_{LV}(\bar\alpha)))\\\approx C_{BSM}(\bar\alpha)+\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}(\Sigma_{{SLV}}-\Sigma_{LV}(\bar\alpha))
ΣSLV≡ΣLV(S,t,K,T,αˉ,β)+(ΣSLV−ΣLV(S,t,K,T,αˉ,β))CSLV=CBSM(ΣLV(αˉ)+(ΣSLV−ΣLV(αˉ)))≈CBSM(αˉ)+∂ΣLV∂CBSM(ΣSLV−ΣLV(αˉ))
上述是对隐含波动率进行了一阶泰勒展开,联系上述式子可以得到:
C
S
L
V
≈
C
B
S
M
(
α
ˉ
)
+
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
C
S
L
V
≈
C
B
S
M
(
α
ˉ
)
+
∂
C
B
S
M
∂
Σ
L
V
(
Σ
S
L
V
−
Σ
L
V
(
α
ˉ
)
)
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
≈
∂
C
B
S
M
∂
Σ
L
V
(
Σ
S
L
V
−
Σ
L
V
(
α
ˉ
)
)
→
Σ
S
L
V
≈
Σ
L
V
(
α
ˉ
)
+
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
∂
C
B
S
M
∂
Σ
L
V
C_{SLV}\approx C_{BSM}(\bar\alpha)+\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)\\ C_{SLV}\approx C_{BSM}(\bar\alpha)+\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}(\Sigma_{{SLV}}-\Sigma_{LV}(\bar\alpha))\\ \frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)\approx\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}(\Sigma_{{SLV}}-\Sigma_{LV}(\bar\alpha))\\ \to\Sigma_{SLV}\approx\Sigma_{LV}(\bar\alpha)+\frac{\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)}{\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}}
CSLV≈CBSM(αˉ)+21∂α2∂2CBSM∣αˉvar(α)CSLV≈CBSM(αˉ)+∂ΣLV∂CBSM(ΣSLV−ΣLV(αˉ))21∂α2∂2CBSM∣αˉvar(α)≈∂ΣLV∂CBSM(ΣSLV−ΣLV(αˉ))→ΣSLV≈ΣLV(αˉ)+∂ΣLV∂CBSM21∂α2∂2CBSM∣αˉvar(α)
当总方差值
σ
2
τ
\sigma^2\tau
σ2τ 很小且位于平值期权附近的时候,就存在
Σ
L
V
(
α
ˉ
)
≈
α
ˉ
/
S
1
−
β
\Sigma_{LV}(\bar\alpha)\approx\bar\alpha/S^{1-\beta}
ΣLV(αˉ)≈αˉ/S1−β。根据链式法则可以证明:
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
≈
(
1
S
1
−
β
)
2
∂
2
C
B
S
M
∂
σ
2
∣
σ
=
Σ
L
V
≈
(
Σ
L
V
α
ˉ
)
2
∂
2
C
B
S
M
∂
σ
2
∣
σ
=
Σ
L
V
\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}\approx(\frac{1}{S^{1-\beta}})^2\frac{\partial^2C_{BSM}}{\partial\sigma^2}|_{\sigma=\Sigma_{LV}}\approx(\frac{\Sigma_{LV}}{\bar\alpha})^2\frac{\partial^2C_{BSM}}{\partial\sigma^2}|_{\sigma=\Sigma_{LV}}
∂α2∂2CBSM∣αˉ≈(S1−β1)2∂σ2∂2CBSM∣σ=ΣLV≈(αˉΣLV)2∂σ2∂2CBSM∣σ=ΣLV
上式右侧是正数,意味着看涨期权对于
α
\alpha
α 成凸性。又因为
v
a
r
(
α
)
≈
α
ˉ
2
ξ
2
τ
var(\alpha)\approx\bar\alpha^2\xi^2\tau
var(α)≈αˉ2ξ2τ,因此之前式子里的
v
a
r
(
α
)
var(\alpha)
var(α) 可以用近似表示代替:
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
∂
C
B
S
M
∂
Σ
L
V
≈
1
2
[
(
Σ
L
V
α
ˉ
)
2
∂
2
C
B
S
M
∂
σ
2
∂
C
B
S
M
∂
σ
∣
σ
=
Σ
L
V
α
ˉ
2
ξ
2
τ
]
≈
1
2
Σ
L
V
2
∂
2
C
B
S
M
∂
σ
2
∂
C
B
S
M
∂
σ
∣
σ
=
Σ
L
V
ξ
2
τ
\frac{\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)}{\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}}\approx\frac{1}{2}[(\frac{\Sigma_{LV}}{\bar\alpha})^2\frac{\frac{\partial^2C_{BSM}}{\partial\sigma^2}}{\frac{\partial C_{BSM}}{\partial\sigma}}|_{\sigma=\Sigma_{LV}}\bar\alpha^2\xi^2\tau]\\\approx\frac{1}{2}\Sigma_{LV}^2\frac{\frac{\partial^2C_{BSM}}{\partial\sigma^2}}{\frac{\partial C_{BSM}}{\partial\sigma}}|_{\sigma=\Sigma_{LV}}\xi^2\tau
∂ΣLV∂CBSM21∂α2∂2CBSM∣αˉvar(α)≈21[(αˉΣLV)2∂σ∂CBSM∂σ2∂2CBSM∣σ=ΣLVαˉ2ξ2τ]≈21ΣLV2∂σ∂CBSM∂σ2∂2CBSM∣σ=ΣLVξ2τ
根据在第19章用到的vega和volga的表达式:
∂
2
C
B
S
M
∂
σ
2
∂
C
B
S
M
∂
σ
=
1
σ
[
1
σ
2
τ
(
ln
(
S
K
)
)
2
−
σ
2
τ
4
]
\frac{\frac{\partial^2C_{BSM}}{\partial\sigma^2}}{\frac{\partial C_{BSM}}{\partial\sigma}}=\frac{1}{\sigma}[\frac{1}{\sigma^2\tau}(\ln(\frac{S}{K}))^2-\frac{\sigma^2\tau}{4}]
∂σ∂CBSM∂σ2∂2CBSM=σ1[σ2τ1(ln(KS))2−4σ2τ]
总方差
σ
2
τ
\sigma^2\tau
σ2τ 很小且期权位于平值状态附近时,
[
ln
(
S
/
K
)
2
]
[\ln(S/K)^2]
[ln(S/K)2] 本身也很小,与
σ
2
τ
\sigma^2\tau
σ2τ 差不多,于是上式进一步改写为:
∂
2
C
B
S
M
∂
σ
2
∂
C
B
S
M
∂
σ
≈
1
σ
[
1
σ
2
τ
(
ln
(
S
K
)
)
2
]
≈
1
σ
3
τ
(
ln
(
S
K
)
)
2
\frac{\frac{\partial^2C_{BSM}}{\partial\sigma^2}}{\frac{\partial C_{BSM}}{\partial\sigma}}\approx\frac{1}{\sigma}[\frac{1}{\sigma^2\tau}(\ln(\frac{S}{K}))^2]\approx\frac{1}{\sigma^3\tau}(\ln(\frac{S}{K}))^2
∂σ∂CBSM∂σ2∂2CBSM≈σ1[σ2τ1(ln(KS))2]≈σ3τ1(ln(KS))2
于是有:
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
∂
C
B
S
M
∂
Σ
L
V
≈
1
2
Σ
L
V
2
∂
2
C
B
S
M
∂
σ
2
∂
C
B
S
M
∂
σ
∣
σ
=
Σ
L
V
ξ
2
τ
≈
1
2
Σ
L
V
2
[
1
σ
3
τ
(
ln
(
S
K
)
)
2
]
∣
σ
=
Σ
L
V
ξ
2
τ
=
1
2
ξ
2
Σ
L
V
(
ln
(
S
K
)
)
2
\frac{\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)}{\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}}\approx\frac{1}{2}\Sigma_{LV}^2\frac{\frac{\partial^2C_{BSM}}{\partial\sigma^2}}{\frac{\partial C_{BSM}}{\partial\sigma}}|_{\sigma=\Sigma_{LV}}\xi^2\tau\approx\frac{1}{2}\Sigma_{LV}^2[\frac{1}{\sigma^3\tau}(\ln(\frac{S}{K}))^2]|_{\sigma=\Sigma_{LV}}\xi^2\tau=\frac{1}{2}\frac{\xi^2}{\Sigma_{LV}}(\ln(\frac{S}{K}))^2
∂ΣLV∂CBSM21∂α2∂2CBSM∣αˉvar(α)≈21ΣLV2∂σ∂CBSM∂σ2∂2CBSM∣σ=ΣLVξ2τ≈21ΣLV2[σ3τ1(ln(KS))2]∣σ=ΣLVξ2τ=21ΣLVξ2(ln(KS))2
将上式代入下式:
Σ
S
L
V
≈
Σ
L
V
(
α
ˉ
)
+
1
2
∂
2
C
B
S
M
∂
α
2
∣
α
ˉ
v
a
r
(
α
)
∂
C
B
S
M
∂
Σ
L
V
\Sigma_{SLV}\approx\Sigma_{LV}(\bar\alpha)+\frac{\frac{1}{2}\frac{\partial^2C_{BSM}}{\partial\alpha^2}|_{\bar\alpha}var(\alpha)}{\frac{\partial C_{BSM}}{\partial\Sigma_{LV}}}
ΣSLV≈ΣLV(αˉ)+∂ΣLV∂CBSM21∂α2∂2CBSM∣αˉvar(α)
当距离到期日时间
τ
\tau
τ 很近,且接近平值期权时,可以得到如下近似表达式:
Σ
S
L
V
≈
Σ
L
V
(
α
ˉ
)
+
1
2
ξ
2
Σ
L
V
(
α
ˉ
)
(
ln
(
S
K
)
)
2
≈
Σ
L
V
[
1
+
1
2
(
ξ
Σ
L
V
(
α
ˉ
)
)
2
(
ln
(
S
K
)
)
2
]
\Sigma_{SLV}\approx\Sigma_{LV}(\bar\alpha)+\frac{1}{2}\frac{\xi^2}{\Sigma_{LV}(\bar\alpha)}(\ln(\frac{S}{K}))^2\approx\Sigma_{LV}[1+\frac{1}{2}(\frac{\xi}{\Sigma_{LV}(\bar\alpha)})^2(\ln(\frac{S}{K}))^2]
ΣSLV≈ΣLV(αˉ)+21ΣLV(αˉ)ξ2(ln(KS))2≈ΣLV[1+21(ΣLV(αˉ)ξ)2(ln(KS))2]
上式说明,当波动率服从随机流程时,局部波动率微笑表达式会加入一个二次项
ln
(
S
/
K
)
2
\ln(S/K)^2
ln(S/K)2,该二次项的系数与随机波动率
ξ
\xi
ξ 值和波动率
α
\alpha
α 值相对大小有关
由于我们是以局部波动率微笑曲线为基础开始分析的,微笑曲线的斜度并不要求 α \alpha α 和股票价格之间存在相关性。但是考虑了 α \alpha α 和股票价格之间的相关性之后,微笑曲线可以得到进一步修正
BSM模型扩展:根据复制原则对随机波动率期权进行估值
假设股票及其波动率服从下列一般随机变化流程:
d
S
=
μ
S
d
t
+
σ
S
d
W
d
σ
=
p
(
S
,
σ
,
t
)
d
t
+
q
(
S
,
σ
,
t
)
d
Z
d
W
d
Z
=
ρ
d
t
dS=\mu Sdt+\sigma SdW\\d\sigma=p(S,\sigma,t)dt+q(S,\sigma,t)dZ\\dWdZ=\rho dt
dS=μSdt+σSdWdσ=p(S,σ,t)dt+q(S,σ,t)dZdWdZ=ρdt
其中
p
(
S
,
σ
,
t
)
,
q
(
S
,
σ
,
t
)
p(S,\sigma,t),q(S,\sigma,t)
p(S,σ,t),q(S,σ,t) 可以满足几何布朗运动、均值回归或者更为一般化的变化过程
现在,假设有某期权的价值是
V
(
S
,
σ
,
t
)
V(S,\sigma,t)
V(S,σ,t),另一个期权的价值是
U
(
S
,
σ
,
t
)
U(S,\sigma,t)
U(S,σ,t),两个期权的标的资产都是同一只股票,但是行权价和/或到期日不同。我们以此构建了一个组合,卖出
Δ
\Delta
Δ 份股票
S
S
S,卖出
δ
\delta
δ 份期权
U
U
U 的合约:
Π
=
V
−
Δ
S
−
δ
U
\Pi=V-\Delta S-\delta U
Π=V−ΔS−δU。根据伊藤引理:
d
Π
=
∂
V
∂
t
d
t
+
∂
V
∂
S
d
S
+
∂
V
∂
σ
d
σ
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
d
t
+
1
2
∂
2
V
∂
σ
2
q
2
d
t
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
d
t
−
Δ
d
S
−
δ
(
∂
U
∂
t
d
t
+
∂
U
∂
S
d
S
+
∂
U
∂
σ
d
σ
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
d
t
+
1
2
∂
2
U
∂
σ
2
q
2
d
t
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
d
t
)
=
[
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
−
δ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
)
]
d
t
+
[
∂
V
∂
S
−
δ
∂
U
∂
S
−
Δ
]
d
S
+
[
∂
V
∂
σ
−
δ
∂
U
∂
σ
]
d
σ
d\Pi=\frac{\partial V}{\partial t}dt+\frac{\partial V}{\partial S}dS+\frac{\partial V}{\partial\sigma}d\sigma+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2dt+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2dt+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho dt\\-\Delta dS-\delta(\frac{\partial U}{\partial t}dt+\frac{\partial U}{\partial S}dS+\frac{\partial U}{\partial\sigma}d\sigma+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2dt+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2dt+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho dt)\\=[\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho\\-\delta(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho)]dt\\+[\frac{\partial V}{\partial S}-\delta\frac{\partial U}{\partial S}-\Delta]dS+[\frac{\partial V}{\partial\sigma}-\delta\frac{\partial U}{\partial\sigma}]d\sigma
dΠ=∂t∂Vdt+∂S∂VdS+∂σ∂Vdσ+21∂S2∂2Vσ2S2dt+21∂σ2∂2Vq2dt+∂S∂σ∂2VσqSρdt−ΔdS−δ(∂t∂Udt+∂S∂UdS+∂σ∂Udσ+21∂S2∂2Uσ2S2dt+21∂σ2∂2Uq2dt+∂S∂σ∂2UσqSρdt)=[∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ−δ(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ)]dt+[∂S∂V−δ∂S∂U−Δ]dS+[∂σ∂V−δ∂σ∂U]dσ
要使
Π
\Pi
Π 成为无风险组合,就需要消掉
d
S
,
d
σ
dS,d\sigma
dS,dσ 对应的各项,需满足如下条件:
∂
V
∂
S
−
δ
∂
U
∂
S
−
Δ
=
0
∂
V
∂
σ
−
δ
∂
U
∂
σ
=
0
\frac{\partial V}{\partial S}-\delta\frac{\partial U}{\partial S}-\Delta=0\\ \frac{\partial V}{\partial\sigma}-\delta\frac{\partial U}{\partial\sigma}=0
∂S∂V−δ∂S∂U−Δ=0∂σ∂V−δ∂σ∂U=0
于是对冲比率等于:
Δ
=
∂
V
∂
S
−
δ
∂
U
∂
S
δ
=
∂
V
∂
σ
∂
U
∂
σ
\Delta=\frac{\partial V}{\partial S}-\delta\frac{\partial U}{\partial S}\\ \delta=\frac{\dfrac{\partial V}{\partial\sigma}}{\dfrac{\partial U}{\partial\sigma}}
Δ=∂S∂V−δ∂S∂Uδ=∂σ∂U∂σ∂V
因此,对冲组合价值的变动可以表示为:
d
Π
=
[
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
−
δ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
)
]
d
t
d\Pi=[\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho\\-\delta(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho)]dt
dΠ=[∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ−δ(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ)]dt
根据不存在无风险套利原则,投资于该无风险组合所能够获得的回报就是无风险利率
r
r
r,于是:
d
Π
=
r
Π
d
t
=
r
(
V
−
Δ
S
−
δ
U
)
d
t
d\Pi=r\Pi dt=r(V-\Delta S-\delta U)dt
dΠ=rΠdt=r(V−ΔS−δU)dt
于是有:
[
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
−
δ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
)
]
d
t
=
r
(
V
−
Δ
S
−
δ
U
)
d
t
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
−
r
V
−
δ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
−
r
U
)
+
r
Δ
S
=
0
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
−
r
V
−
δ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
−
r
U
)
+
r
(
∂
V
∂
S
−
δ
∂
U
∂
S
)
S
=
0
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
+
∂
V
∂
S
r
S
−
r
V
=
∂
V
∂
σ
∂
U
∂
σ
(
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
+
∂
U
∂
S
r
S
−
r
U
)
[\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho\\-\delta(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho)]dt=r(V-\Delta S-\delta U)dt\\ \frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho-rV\\-\delta(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho-rU)+r\Delta S=0\\ \frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho-rV\\-\delta(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho-rU)+r(\frac{\partial V}{\partial S}-\delta\frac{\partial U}{\partial S})S=0\\ \frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial V}{\partial S}rS-rV\\=\frac{\dfrac{\partial V}{\partial\sigma}}{\dfrac{\partial U}{\partial\sigma}}(\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial U}{\partial S}rS-rU)
[∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ−δ(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ)]dt=r(V−ΔS−δU)dt∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ−rV−δ(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ−rU)+rΔS=0∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ−rV−δ(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ−rU)+r(∂S∂V−δ∂S∂U)S=0∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ+∂S∂VrS−rV=∂σ∂U∂σ∂V(∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ+∂S∂UrS−rU)
推导可得:
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
+
∂
V
∂
S
r
S
−
r
V
∂
V
∂
σ
=
∂
U
∂
t
+
1
2
∂
2
U
∂
S
2
σ
2
S
2
+
1
2
∂
2
U
∂
σ
2
q
2
+
∂
2
U
∂
S
∂
σ
σ
q
S
ρ
+
∂
U
∂
S
r
S
−
r
U
∂
U
∂
σ
\frac{\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial V}{\partial S}rS-rV}{\dfrac{\partial V}{\partial\sigma}}\\=\frac{\frac{\partial U}{\partial t}+\frac{1}{2}\frac{\partial^2U}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2U}{\partial\sigma^2}q^2+\frac{\partial^2U}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial U}{\partial S}rS-rU}{\dfrac{\partial U}{\partial\sigma}}
∂σ∂V∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ+∂S∂VrS−rV=∂σ∂U∂t∂U+21∂S2∂2Uσ2S2+21∂σ2∂2Uq2+∂S∂σ∂2UσqSρ+∂S∂UrS−rU
上式左侧部分是一个只与期权
V
V
V 有关的函数,右侧部分是一个只与期权
U
U
U 有关的函数。期权
U
,
V
U,V
U,V 对应的行权价和到期期限完全相互独立,因此,对于任意的
U
,
V
U,V
U,V 而言,要使上式成立的唯一办法就是等式两侧的各项跟期权的参数完全无关。换句话说,上式两侧的部分仅仅是与
S
,
σ
,
t
S,\sigma,t
S,σ,t 相关的函数
引入一个未知函数
ϕ
(
S
,
σ
,
t
)
\phi(S,\sigma,t)
ϕ(S,σ,t),并令上式两侧都等于
−
ϕ
(
S
,
σ
,
t
)
-\phi(S,\sigma,t)
−ϕ(S,σ,t),于是可以得到期权
V
V
V 的估值方程式:
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
+
∂
V
∂
S
r
S
−
r
V
+
∂
V
∂
σ
ϕ
(
S
,
σ
,
t
)
=
0
\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial V}{\partial S}rS-rV+\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t)=0
∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ+∂S∂VrS−rV+∂σ∂Vϕ(S,σ,t)=0
这是用于随机波动率期权估值的偏微分方程
从夏普比率的角度来解释 ϕ ( S , σ , t ) \phi(S,\sigma,t) ϕ(S,σ,t) 的含义
根据伊藤引理,我们可以将期权的价值变动表示为:
d
V
=
∂
V
∂
t
d
t
+
∂
V
∂
S
d
S
+
∂
V
∂
σ
d
σ
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
d
t
+
1
2
∂
2
V
∂
σ
2
q
2
d
t
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
d
t
=
(
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
)
d
t
+
∂
V
∂
S
d
S
+
∂
V
∂
σ
d
σ
dV=\frac{\partial V}{\partial t}dt+\frac{\partial V}{\partial S}dS+\frac{\partial V}{\partial\sigma}d\sigma+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2dt+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2dt+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho dt\\=(\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho)dt+\frac{\partial V}{\partial S}dS+\frac{\partial V}{\partial\sigma}d\sigma
dV=∂t∂Vdt+∂S∂VdS+∂σ∂Vdσ+21∂S2∂2Vσ2S2dt+21∂σ2∂2Vq2dt+∂S∂σ∂2VσqSρdt=(∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ)dt+∂S∂VdS+∂σ∂Vdσ
将:
d
S
=
μ
S
d
t
+
σ
S
d
W
d
σ
=
p
(
S
,
σ
,
t
)
d
t
+
q
(
S
,
σ
,
t
)
d
Z
d
W
d
Z
=
ρ
d
t
dS=\mu Sdt+\sigma SdW\\ d\sigma=p(S,\sigma,t)dt+q(S,\sigma,t)dZ\\ dWdZ=\rho dt
dS=μSdt+σSdWdσ=p(S,σ,t)dt+q(S,σ,t)dZdWdZ=ρdt
中的
d
S
,
d
σ
dS,d\sigma
dS,dσ 代入得:
d
V
=
(
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
)
d
t
+
∂
V
∂
S
(
μ
S
d
t
+
σ
S
d
W
)
+
∂
V
∂
σ
(
p
(
S
,
σ
,
t
)
d
t
+
q
(
S
,
σ
,
t
)
d
Z
)
=
(
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
+
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
)
d
t
+
∂
V
∂
S
σ
S
d
W
+
∂
V
∂
σ
q
(
S
,
σ
,
t
)
d
Z
≡
μ
V
V
d
t
+
V
σ
V
,
S
d
W
+
V
σ
V
,
σ
d
Z
dV=(\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho)dt\\+\frac{\partial V}{\partial S}(\mu Sdt+\sigma SdW)+\frac{\partial V}{\partial\sigma}(p(S,\sigma,t)dt+q(S,\sigma,t)dZ)\\=(\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t))dt\\+\frac{\partial V}{\partial S}\sigma SdW+\frac{\partial V}{\partial\sigma}q(S,\sigma,t)dZ\\\equiv\mu_VVdt+V\sigma_{V,S}dW+V\sigma_{V,\sigma}dZ
dV=(∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ)dt+∂S∂V(μSdt+σSdW)+∂σ∂V(p(S,σ,t)dt+q(S,σ,t)dZ)=(∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ+∂S∂VμS+∂σ∂Vp(S,σ,t))dt+∂S∂VσSdW+∂σ∂Vq(S,σ,t)dZ≡μVVdt+VσV,SdW+VσV,σdZ
当
V
V
V 服从几何布朗运动时,其预期回报和波动率分别为:
μ
V
=
1
V
(
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
+
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
)
σ
V
,
S
=
∂
V
∂
S
S
V
σ
σ
V
,
σ
=
∂
V
∂
σ
q
(
S
,
σ
,
t
)
V
σ
V
=
σ
V
,
S
2
+
σ
V
,
σ
2
+
2
ρ
σ
V
,
S
σ
V
,
σ
\mu_V=\frac{1}{V}(\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho+\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t))\\ \sigma_{V,S}=\frac{\partial V}{\partial S}\frac{S}{V}\sigma\\ \sigma_{V,\sigma}=\frac{\partial V}{\partial\sigma}\frac{q(S,\sigma,t)}{V}\\ \sigma_V=\sqrt{\sigma_{V,S}^2+\sigma_{V,\sigma}^2+2\rho\sigma_{V,S}\sigma_{V,\sigma}}
μV=V1(∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ+∂S∂VμS+∂σ∂Vp(S,σ,t))σV,S=∂S∂VVSσσV,σ=∂σ∂VVq(S,σ,t)σV=σV,S2+σV,σ2+2ρσV,SσV,σ
可以将
σ
V
,
S
,
σ
V
,
σ
\sigma_{V,S},\sigma_{V,\sigma}
σV,S,σV,σ 看作期权
V
V
V 的偏微分波动率,而
V
V
V 的总波动率表示为
σ
V
\sigma_V
σV
之前提到期权的估值方程式:
∂
V
∂
t
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
=
r
V
−
∂
V
∂
S
r
S
−
∂
V
∂
σ
ϕ
(
S
,
σ
,
t
)
\frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho=rV-\frac{\partial V}{\partial S}rS-\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t)
∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ=rV−∂S∂VrS−∂σ∂Vϕ(S,σ,t)
等式两边同时增加两项:
∂
V
∂
t
+
[
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
]
+
1
2
∂
2
V
∂
S
2
σ
2
S
2
+
1
2
∂
2
V
∂
σ
2
q
2
+
∂
2
V
∂
S
∂
σ
σ
q
S
ρ
=
r
V
−
∂
V
∂
S
r
S
−
∂
V
∂
σ
ϕ
(
S
,
σ
,
t
)
+
[
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
]
μ
V
V
=
r
V
−
∂
V
∂
S
r
S
−
∂
V
∂
σ
ϕ
(
S
,
σ
,
t
)
+
[
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
]
μ
V
−
r
=
1
V
(
∂
V
∂
S
μ
S
+
∂
V
∂
σ
p
(
S
,
σ
,
t
)
−
∂
V
∂
S
r
S
−
∂
V
∂
σ
ϕ
(
S
,
σ
,
t
)
)
=
∂
V
∂
S
S
V
(
μ
−
r
)
+
∂
V
∂
σ
1
V
(
p
(
S
,
σ
,
t
)
−
ϕ
(
S
,
σ
,
t
)
)
μ
V
−
r
=
σ
V
,
S
μ
−
r
σ
+
σ
V
,
σ
p
(
S
,
σ
,
t
)
−
ϕ
(
S
,
σ
,
t
)
q
(
S
,
σ
,
t
)
\frac{\partial V}{\partial t}+[\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t)]+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho\\=rV-\frac{\partial V}{\partial S}rS-\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t)+[\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t)]\\ \mu_VV=rV-\frac{\partial V}{\partial S}rS-\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t)+[\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t)]\\ \mu_V-r=\frac{1}{V}(\frac{\partial V}{\partial S}\mu S+\frac{\partial V}{\partial\sigma}p(S,\sigma,t)-\frac{\partial V}{\partial S}rS-\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t))\\=\frac{\partial V}{\partial S}\frac{S}{V}(\mu-r)+\frac{\partial V}{\partial\sigma}\frac{1}{V}(p(S,\sigma,t)-\phi(S,\sigma,t))\\ \mu_V-r=\sigma_{V,S}\frac{\mu-r}{\sigma}+\sigma_{V,\sigma}\frac{p(S,\sigma,t)-\phi(S,\sigma,t)}{q(S,\sigma,t)}
∂t∂V+[∂S∂VμS+∂σ∂Vp(S,σ,t)]+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ=rV−∂S∂VrS−∂σ∂Vϕ(S,σ,t)+[∂S∂VμS+∂σ∂Vp(S,σ,t)]μVV=rV−∂S∂VrS−∂σ∂Vϕ(S,σ,t)+[∂S∂VμS+∂σ∂Vp(S,σ,t)]μV−r=V1(∂S∂VμS+∂σ∂Vp(S,σ,t)−∂S∂VrS−∂σ∂Vϕ(S,σ,t))=∂S∂VVS(μ−r)+∂σ∂VV1(p(S,σ,t)−ϕ(S,σ,t))μV−r=σV,Sσμ−r+σV,σq(S,σ,t)p(S,σ,t)−ϕ(S,σ,t)
上式左侧部分表示期权的预期超额回报,要得到夏普比率,就需要同时除以期权的波动率,于是:
μ
V
−
r
σ
V
=
σ
V
,
S
σ
V
μ
−
r
σ
+
σ
V
,
σ
σ
V
p
(
S
,
σ
,
t
)
−
ϕ
(
S
,
σ
,
t
)
q
(
S
,
σ
,
t
)
\frac{\mu_V-r}{\sigma_V}=\frac{\sigma_{V,S}}{\sigma_V}\frac{\mu-r}{\sigma}+\frac{\sigma_{V,\sigma}}{\sigma_V}\frac{p(S,\sigma,t)-\phi(S,\sigma,t)}{q(S,\sigma,t)}
σVμV−r=σVσV,Sσμ−r+σVσV,σq(S,σ,t)p(S,σ,t)−ϕ(S,σ,t)
上式表明,在不存在无风险套利机会的条件下,随机波动率的期权定价表达式说明,期权的夏普比率是由两部分构成的:股票的夏普比率和波动率的夏普比率,各自的权重分别等于它们对于期权整体波动率的相对贡献比例
随机波动率模型的特征解
∂ V ∂ t + 1 2 ∂ 2 V ∂ S 2 σ 2 S 2 + 1 2 ∂ 2 V ∂ σ 2 q 2 + ∂ 2 V ∂ S ∂ σ σ q S ρ = r V − ∂ V ∂ S r S − ∂ V ∂ σ ϕ ( S , σ , t ) \frac{\partial V}{\partial t}+\frac{1}{2}\frac{\partial^2V}{\partial S^2}\sigma^2S^2+\frac{1}{2}\frac{\partial^2V}{\partial\sigma^2}q^2+\frac{\partial^2V}{\partial S\partial\sigma}\sigma qS\rho=rV-\frac{\partial V}{\partial S}rS-\frac{\partial V}{\partial\sigma}\phi(S,\sigma,t) ∂t∂V+21∂S2∂2Vσ2S2+21∂σ2∂2Vq2+∂S∂σ∂2VσqSρ=rV−∂S∂VrS−∂σ∂Vϕ(S,σ,t)
上式的解是在股票价格服从随机波动率条件下,期权预期损益的风险中性现值:
V
=
e
−
r
(
T
−
t
)
∑
所有路径
p
(
p
a
t
h
)
×
损益
∣
路径
V=e^{-r(T-t)}\sum_{所有路径}p(path)\times损益|_{路径}
V=e−r(T−t)所有路径∑p(path)×损益∣路径
V
V
V 代表任意标准欧式期权,
p
(
p
a
t
h
)
p(path)
p(path) 表示每条股票价格变动路径的风险中性概率
Hull 和 White(1987)曾提出,要定义股价变动路径,可以用到期日股票价格
S
T
S_T
ST 以及沿该路径的平均方差两个参数。我们假设某股票价格变动路径上的平均方差可以用如下等式表示
σ
T
2
ˉ
=
1
T
∫
0
T
σ
t
2
d
t
\bar{\sigma_T^2}=\frac{1}{T}\int_0^T\sigma_t^2dt
σT2ˉ=T1∫0Tσt2dt
我们将用
σ
T
ˉ
\bar{\sigma_T}
σTˉ 表示到期日为
T
T
T,股票变动路径的波动率,实际上就是路径上平均方差的开方值
于是风险中性现值就可以表示为两个加总项之和,即所有到期日股票价格以及所有路径波动率,也就是:
V
=
e
−
r
(
T
−
t
)
∑
所有
σ
ˉ
T
∑
给定
σ
T
ˉ
,
S
T
的路径
p
(
σ
T
ˉ
,
S
T
)
×
损益
∣
路径
V=e^{-r(T-t)}\sum_{所有\bar\sigma_T}\sum_{给定\bar{\sigma_T},S_T的路径}p(\bar{\sigma_T},S_T)\times损益|_{路径}
V=e−r(T−t)所有σˉT∑给定σTˉ,ST的路径∑p(σTˉ,ST)×损益∣路径
其中
p
(
σ
T
ˉ
,
S
T
)
p(\bar{\sigma_T},S_T)
p(σTˉ,ST) 表示某特定到期日股票价格及对应的路径波动率的概率。如果股票价格跟波动率之间没有相关性(
ρ
=
0
\rho=0
ρ=0),那么式上式中的概率就可以分解为两个独立的概率
f
f
f 和
g
g
g,于是:
p
(
σ
T
ˉ
,
S
T
)
=
f
(
σ
ˉ
T
)
×
g
(
S
T
)
p(\bar{\sigma_T},S_T)=f(\bar\sigma_T)\times g(S_T)
p(σTˉ,ST)=f(σˉT)×g(ST)
于是:
V
=
e
−
r
(
T
−
t
)
∑
所有
σ
ˉ
T
f
(
σ
ˉ
T
)
∑
给定
σ
T
ˉ
,
S
T
的路径
g
(
S
T
)
×
损益
∣
路径
V=e^{-r(T-t)}\sum_{所有\bar\sigma_T}f(\bar\sigma_T)\sum_{给定\bar{\sigma_T},S_T的路径}g(S_T)\times损益|_{路径}
V=e−r(T−t)所有σˉT∑f(σˉT)给定σTˉ,ST的路径∑g(ST)×损益∣路径
这里还可以进行进一步简化处理。在之前的双布朗运动模型中,(股票价格跟波动率之间的)相关系数也等于 0,对于给定的路径波动率,到期日所有股票价格对应的预期期权损益的现值,就等于给定波动率下的 BSM 模型期权价值
V
B
S
M
V_{BSM}
VBSM,于是:
V
B
S
M
(
S
,
t
,
K
,
T
,
r
,
σ
ˉ
T
)
=
e
−
r
(
T
−
t
)
∑
给定
σ
T
ˉ
,
S
T
的路径
g
(
S
T
)
×
损益
∣
路径
V
=
∑
所有
σ
ˉ
T
f
(
σ
ˉ
T
)
×
V
B
S
M
(
S
,
t
,
K
,
T
,
r
,
σ
ˉ
T
)
V_{BSM}(S,t,K,T,r,\bar\sigma_T)=e^{-r(T-t)}\sum_{给定\bar{\sigma_T},S_T的路径}g(S_T)\times损益|_{路径}\\ V=\sum_{所有\bar\sigma_T}f(\bar\sigma_T)\times V_{BSM}(S,t,K,T,r,\bar\sigma_T)
VBSM(S,t,K,T,r,σˉT)=e−r(T−t)给定σTˉ,ST的路径∑g(ST)×损益∣路径V=所有σˉT∑f(σˉT)×VBSM(S,t,K,T,r,σˉT)
因此,当相关系数等于0的时候,对于一个标准欧式期权而言,随机波动率条件下的解,就等于不同波动率路径下的 BSM 解的加权求和数。这一直观上令人满意的结论也被称作混合定理,最早是由 Hull和White(1987) 提出的