数学杂志  2020, Vol. 40 Issue (6): 707-716   PDF    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
陈威
李志民
张雪峰
基于超前倒向随机微分方程的动态风险度量
陈威, 李志民, 张雪峰    
安徽工程大学数理学院, 安徽 芜湖 241000
摘要:本文研究了基于超前倒向随机微分方程的时间相容的过程的动态凸(一致性)风险度量的问题.利用对超前倒向随机微分方程生成元的适当假设,建立超前倒向随机微分方程生成元与过程的动态凸(一致性)风险度量的对应模型,证明了超前倒向随机微分方程的解可以定义时间相容的过程的风险度量.得到了基于超前倒向随机微分方程的风险度量,推广了基于倒向随机微分方程的动态风险度量.由于超前倒向随机微分方程生成元中包含当前时刻和未来时刻的解,因此本文的结论对风险的预测更加可靠.
关键词超前倒向随机微分方程    随机过程    风险度量    时间相容性    
DYNAMIC RISK MEASUREMENT VIA ANTICIPATED BACKWARD STOCHASTIC DIFFERENTIAL EQUATIONS
CHEN Wei, LI Zhi-min, ZHANG Xue-feng    
School of Mathematics and Physics, Anhui Polytechnic University, Wuhu 241000, China
Abstract: This paper studies the problem of time-consistent dynamic convex (coherent) risk measures for processes via anticipated backward stochastic differential equations. Using appropriate assumptions on the generator of the anticipated backward stochastic differential equation, the corresponding model of the anticipated backward stochastic differential equation generator and the dynamic convex (coherent) risk measure of the process is established, which proves that the solution of the anticipated backward stochastic differential equation can be defined for the risk measurement of time-consistent processes, the risk measurement based on the anticipated backward stochastic differential equation is obtained, and the dynamic risk measurement based on the backward stochastic differential equation is extended. Because the anticipated backward stochastic differential equations generator contains the current and future solutions, the conclusion of this paper is more reliable for risk prediction.
Keywords: anticipated backward stochastic differential equations     stochastic processes     risk measures     time-consistency    
1 引言

金融系统中传统的风险度量方法注重在一定概率空间上由随机变量描述不确定收益, 在这种情况下, 付款通常被认为是贴现的, 时间对于风险度量来说并不重要.然而, 货币的时间价值可以通过一个简单的折现过程来解决的假设在很多情况下过于局限.基于传统风险度量所存在的问题, Artzner等[1]在公理化的假设下, 讨论了一致性风险度量的方法. F{ö}llmer和Schied[2]对一致性风险度量的概念进行延伸, 引入了凸风险度量的概念.

Frittelli和Rosazza Gianin[3]在此基础上讨论了一组定义凸风险度量的公理, 其中将货币风险度量与传统的效用函数区别开的一个主要公理是现金不变性.出于监管目的, 现金不变性给出了风险度量最小资本要求的解释, 但它也假定未来的偿付和目前的资本储备是用相同的数字表示的, 因此现金不变性不允许人们处理贴现不确定性的问题.为了弥补这一缺点, El Karoui和Ravanelli[4]用现金次可加性取代了现金不变性.在此基础上, Cheridito等[5]考虑在整个时间间隔内度量价值未来演化的风险来解决贴现不确定性, 并提出了过程的一致性和凸风险度量.随机过程的框架允许人们放松现金不变性的公理且不丢失风险度量作为最低资本要求的解释.同时, Riedel[6]、Bayraktar[7]、Cheng和Riedel[8]对货币时间价值的不确定性做出了解释. Acciaio等[9]引入的过程的风险度量为处理模型的模糊性提供了一个自然地框架, 并在离散时间框架下引入了过程凸风险度量的鲁棒表示. Jiang[10]在对生成元的适当假设下, 建立基于$ g $-期望的相关过程的风险度量.

为解决模型的模糊性以及货币时间的不确定性, Penner和R{é}veillac[11]提供了一种连续时间的风险度量方法, 并建立了过程的风险度量和倒向随机微分方程之间的联系. Peng和Yang[12]在倒向随机微分方程(简称:BSDEs)和随机时滞微分方程(简称:SDDEs)的基础上引入了一类新的随机微分方程---超前倒向随机微分方程(简称:超前BSDEs), 考虑其生成元包含当前和未来时刻的解, 证明方程解的存在唯一性, 并对解的比较定理进行研究. Xu和Chen[13]对超前BSDEs生成元满足随机Lipschitz条件的情况进行研究, 得到了超前BSDEs解的存在唯一性和比较定理.在对有限或无限时间间隔内的BSDEs生成元的适当假设下, Ji等[14]建立了基于BSDEs的过程的动态风险度量模型, 证明了BSDEs生成元与过程的动态风险度量的对应关系.

文献[14]中BSDEs的生成元只包含当前$ t $时刻的解, 即通过BSDEs定义的风险度量只能度量当前时刻的风险, 并不能对未来时刻的风险做预测, 这限制了理论的进一步的应用.为此本文考虑生成元中包含将来时刻的解, 在此模型下通过对超前BSDEs生成元进行适当假设, 建立基于超前BSDEs的过程的动态风险度量的模型, 研究超前BSDEs生成元与过程的动态凸(一致性)风险度量的一一对应关系.全文结构如下:第2节回顾过程的动态风险度量的有关符号和概念以及时间相容性的概念; 第3节考虑一种类型的超前BSDEs并对超前BSDEs生成元做适当假设, 建立了基于超前BSDEs的过程的动态风险度量的模型; 第4节证明了超前BSDEs生成元与基于超前BSDEs的时间相容的过程的动态凸(一致性)风险度量的对应关系; 第5节是对具体例子进行分析并对文章做一个整体性的总结.

2 动态风险度量的定义及符号

$ T\in[0, \infty] $, $ (\Omega, \mathcal{F}, \mathbb{P}) $$ T $上的完备概率空间, $ (W_{t})_{(t\geq0)} $是定义在该空间上的$ d $维标准维纳过程, $ (\mathcal{F}_{t} )_{(t\geq0)} $是由$ (W_{t})_{(t\geq0)} $生成的$ \sigma $域流.对任意的$ z\in{R^{n}} $, 记$ \|z\| $为欧式范数, 记$ \mathcal{W}(R^{n}) $是所有$ R^{n} $值, $ \mathcal{F}_{t} $循序可测过程构成的空间.

定义过程的空间如下

$ \begin{eqnarray*} && \ L^2 (\Omega, \mathcal{F}_t, \mathbb{P}) = \{\xi;\xi\mbox{是}R\mbox{值}, \mathcal{F}_t\mbox{可测的}, \mbox{并且}E_\mathbb{P}[\xi^2 ]<\infty\}, \\ && \ L^\infty (\Omega, \mathcal{F}_t, \mathbb{P}) = \{\xi;\xi\mbox{是}R\mbox{值}, \mathcal{F}_t\mbox{可测并且有界的}\}, \\ && \ S^2 = \{\delta\in \mathcal{W}(R^{n});\delta\mbox{是右连左极的}, \mbox{并且}E_\mathbb{P} [\sup\limits_{(t\in[0, T])}|\delta|^2 ]<\infty\}. \end{eqnarray*} $

考虑动态框架下的风险度量, $ x\in\mathbb{R}^{\infty} $表示为一个模拟某种金融资产演化的价值过程, 过程$ m1_{[t, T]} $表示时间$ {t}\leq{T} $时一次支付$ m $单位的现金.对$ \forall{0}\leq{t}\leq{s}\leq{T} $, $ x\in\mathbb{R}^{\infty}, r\in[0, T] $, 定义映射$ \pi_{(t, s)}:{\mathbb{R}^{\infty}}\rightarrow{\mathbb{R}^{\infty}} $如下

$ \begin{equation*} \pi_{(t, s)}{(X)}_{r}: = {1_{[t, T]}{(r)}{X}_{({r}\wedge{s})}}, \quad \mathbb{R}_{(t, s)}^{\infty}: = \pi_{(t, s)}{(\mathbb{R}^{\infty})}, \quad \mathbb{R}_{t}^{\infty}: = \pi_{(t, T)}{(\mathbb{R}^{\infty})}. \end{equation*} $

为进一步明确风险度量, 我们引入文献[11]中有关过程的动态一致性风险度量的符号.

定义2.1   对所有的$X_1, X_2\in{\mathbb{R}_t^\infty}$, 当$ t \in [0, T]$时的映射$\rho_t:\mathbb{R}_t^{\infty}\rightarrow{L^\infty}{(\Omega, \mathcal{F}_t, \mathbb{P})}$满足下列性质时, 映射$ \rho_t $被称为过程的条件一致性风险度量:

(1) (条件现金不变性)对$ \forall{m}\in{L^\infty}(\Omega, \mathcal{F}_t, \mathbb{P}) $, 有$\rho_t(X+m1_{[t, T]})=\rho_t(X)-m $.

(2) (条件正齐次性)对$ \forall t \in [0, T]$, 有$ \rho_t(X_1+X_2)\leq\rho_t(X_1)+\rho_t(X_2) $.

(3) (单调性)如果$ X_1\leq{X_2} $, 则$ \rho_t(X_1)\geq\rho_t(X_2) $.

(4) (正则性)$ \rho_t(0) = 0 $.

对任意的$ t \in [0, T] $, 如果映射$ \rho_t:\mathbb{R}_t^{\infty}\rightarrow{L^\infty}{(\Omega, \mathcal{F}_t, \mathbb{P})} $是过程的条件一致性风险度量, 那么序列$ (\rho_t)_{t \in [0, T]} $称为过程的动态一致性风险度量.

考虑到一致性风险度量存在的局限性, 为了更广泛的研究金融市场的风险度量问题, 我们接下来引入过程的动态凸风险度量的概念.

定义2.2   对所有的$ X_1, X_2\in{\mathbb{R}_t^\infty} $, 当$ t \in [0, T] $时的映射$ \rho_t:\mathbb{R}_t^{\infty}\rightarrow{L^\infty}{(\Omega, \mathcal{F}_t, \mathbb{P})} $满足下列性质时, 映射$ \rho_t $被称为过程的条件凸风险度量:

(1) (条件现金不变性)对$ \forall{m}\in{L^\infty}(\Omega, \mathcal{F}_t, \mathbb{P}) $, 有$ \rho_t(X+m1_{[t, T]}) = \rho_t(X)-m $.

(2) (条件凸性)对所有的$ \lambda\in{L^\infty}(\Omega, \mathcal{F}_t, \mathbb{P}) $, 有$ \rho_t(\lambda{X_1}+(1-\lambda)X_2)\leq\lambda\rho_t(X_1)+(1-\lambda)\rho_t(X_1) $.

(3) (单调性)如果$ X_1\leq{X_2} $, 则$ \rho_t(X_1)\geq\rho_t(X_2) $.

(4) (正则性)$ \rho_t(0) = 0 $.

对任意的$ t \in [0, T] $, 如果映射$ \rho_t:\mathbb{R}_t^{\infty}\rightarrow{L^\infty}{(\Omega, \mathcal{F}_t, \mathbb{P})} $是过程的条件凸风险度量, 那么序列$ (\rho_t)_{t \in [0, T]} $称为过程的动态凸风险度量.

容易验证一个过程的条件一致性风险度量同样也是一个过程的条件凸风险度量.对$ X\in{\mathbb{R}^\infty} $, 记$ \rho_t(X) = \rho_t{(\pi_{(t, T)}(X))} $, 然后引入时间相容的概念.

定义2.3   对任意的$ s\in[t, T] $, $ t \in [0, T] $, $ X\in{\mathbb{R}^\infty} $, 若过程的动态风险度量$ (\rho_t)_{t \in [0, T]} $满足:$ \rho_t(X) = \rho_t(X1_{[t, s)}-\rho_s(X)1_{[s, T]}) $, 则称过程的动态风险度量$ (\rho_t)_{t \in [0, T]} $是时间相容的.

3 超前BSDEs的基本假设和主要结论

考虑下面一种类型的超前BSDE:

$\begin{equation} \left\{ \begin{array}{lr} Y_t=\xi_T+\int_{t}^{T}g(s, Y_s, Z_s, Y_{s+\alpha(s)}, Z_{s+\beta(s)})ds-\int_{t}^{T}Z_sdW_s, & t\in{[0, T]}\\ Y_t=\xi_t, & t\in{[T, T+K]}\\ Z_t=\eta_t, & t\in{[T, T+K]} \end{array} \right. \end{equation} $ (3.1)

其中$ \alpha{(\cdot)}:[0, T]\rightarrow{R^+} $$ \beta{(\cdot)}:[0, T]\rightarrow R^+ $是满足下面条件的连续函数:

(a) 存在某一常数$ K\geq0 $, 使得对任何$ t\in[0, T] $,

$ \begin{equation*} \ t+\alpha(t)\leq{T+K}, \quad t+\beta(t)\leq{T+K}; \end{equation*} $

(b) 存在某一常数$ C\geq0 $, 使得对任何$ t\in[0, T] $及非负可积函数$ f(\cdot) $,

$ \begin{equation*} \ \int_{t}^{T}f(s+\alpha{(s)})ds\leq c \int_{t}^{T+K}f(s)ds, \quad \int_{t}^{T}f(s+\beta{(s)})ds\leq c \int_{t}^{T+K}f(s)ds. \end{equation*} $

上述超前BSDE的生成元$ g $:$ \Omega\times{R}^{m}\times{R}^{m\times{d}}\times{{L_\mathcal{F}^2}{(t, T+K;{R}^{m})}}\times{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}}\rightarrow{L_2{(\mathcal{F}_t, {R}^{m})}} $是一个使得$ (g(t, y, z, \xi, \eta))_{t \in [0, T]} $是循序可测的函数, 并且生成元$ g $满足下面假设条件

(1) $ g $$ (y, z, \xi, \eta) $上是随机Lipschitz连续的, 即存在四个正的$ \mathcal{F}_{t} $循序可测随机过程:$ \varphi_1(t) $, $ \varphi_2(t) $, $ \phi_1(t) $, $ \phi_2(t) $, 对任意的$ t \in [0,T],{y_{1,2}} \in {R^m},{z_{1,2}} \in {R^{m \times d}},{\xi ^{1,2}} \in L_{\cal F}^2(t,T + K;{R^m}), $${\eta ^{1,2}} \in L_{\cal F}^2(t,T + K;{R^{m \times d}}),{\gamma _{1,2}} \in [T,T + K] $, 有

$ \begin{eqnarray*} && \ |g(t, y_1, z_1, \xi_{\gamma_1}^1, \eta_{\gamma_2}^1)-g(t, y_2, z_2, \xi_{\gamma_1}^2, \eta_{\gamma_2}^2)| \\ \ &\leq&{\varphi_1(t)|y_1-y_2|+\phi_1(t)|z_1-z_2|+E^{\mathcal{F}_t}[\varphi_2(t)|\xi_{\gamma_1}^1-\xi_{\gamma_1}^2|+\phi_2(t)|\eta_{\gamma_2}^1-\eta_{\gamma_2}^2|]} \end{eqnarray*} $

其中, $ \int_{0}^{T}[\varphi_1(t)+\phi_1^2(t)+\varphi_2(t)+\phi_2^2(t)]dt<\infty $.

(2) $ E_\mathbb{P}[ \int_{0}^{T}|g(t, 0, 0, 0, 0)|^2dt]<\infty $.

(3) $ g $$ (y, z, \xi, \eta) $上是非增的.

(4) $ g(t, 0, 0, 0, 0) = 0 $.

(5) $ g $$ (y, z, \xi, \eta) $上是凸的, 即对任意的$ t \in [0, T] $, $ y_{1, 2}\in{R}^{m} $, $ z_{1, 2}\in{R}^{m\times{d}} $, $ \xi^{1, 2}\in{L_\mathcal{F}^2}{(t, T+} $ $ K;{R}^{m}) $, $ \eta^{1, 2}\in{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}} $, $ \gamma_{1, 2}\in[T, T+K] $, $ \lambda\in[0, 1] $, 有:

$ \begin{eqnarray*} && g(t, \lambda{y_1}+(1-\lambda){y_2}, \lambda{z_1}+(1-\lambda){z_2}, \lambda{\xi_{\gamma_1}^1}+(1-\lambda){\xi_{\gamma_1}^2}, \lambda{\eta_{\gamma_2}^1}+(1-\lambda){\eta_{\gamma_2}^2}) \\ & \leq& {\lambda{g(t, y_1, z_1, \xi_{\gamma_1}^1, \eta_{\gamma_2}^1)}+(1-\lambda){g(t, y_2, z_2, \xi_{\gamma_1}^2, \eta_{\gamma_2}^2)}}. \end{eqnarray*} $

(6) $ g $$ (y, z, \xi, \eta) $上是次可加的, 即对任意的$ t \in [0, T] $, $ y_{1, 2}\in{R}^{m} $, $ z_{1, 2}\in{R}^{m\times{d}} $, $ \xi^{1, 2}\in{L_\mathcal{F}^2} $ $ {(t, T+K;{R}^{m})} $, $ \eta^{1, 2}\in{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}} $, $ \gamma_{1, 2}\in[T, T+K] $, 有:

$ \begin{equation*} g(t, {y_1}+{y_2}, {z_1}+{z_2}, {\xi_{\gamma_1}^1}+{\xi_{\gamma_1}^2}, {\eta_{\gamma_2}^1}+{\eta_{\gamma_2}^2}) \leq{g(t, y_1, z_1, \xi_{\gamma_1}^1, \eta_{\gamma_2}^1)}+{g(t, y_2, z_2, \xi_{\gamma_1}^2, \eta_{\gamma_2}^2)}. \end{equation*} $

(7) $ g $$ (y, z, \xi, \eta) $上是正齐次的, 即对任意的$ \theta\geq0 $, $ y_{1, 2}\in{R}^{m} $, $ z_{1, 2}\in{R}^{m\times{d}} $, $ \xi^{1, 2}\in{L_\mathcal{F}^2}{(t, T+K;} $ $ {R}^{m}), \eta^{1, 2}\in{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}} $, $ \gamma_{1, 2}\in[T, T+K] $, 有:

$ \begin{equation*} \ g(t, \theta{y}, \theta{z}, \theta{\xi}, \theta{\eta}) = \theta{g(t, y, z, \xi, \eta)}. \end{equation*} $

若生成元g满足上述假设条件(1), (2), 对任意终端条件$ \xi_T\in{{S_\mathcal{F}^2}{(T, T+K;{R}^{m})}} $$ \eta_T\in{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}} $, 由文献[13]知存在唯一适应对$ (Y_t, Z_t)_{t\in[0, T]}\in{{S_\mathcal{F}^2}{(0, T+K;{R}^{m\times{d}})}\times{L_\mathcal{F}^2}} $ $ {(0, T+K;{R}^{m\times{d}})} $满足超前BSDE$ (3.1) $.

对任意固定的$ X\in{S^{2}} $, 考虑下面形式的超前BSDE:

$ \begin{equation} \left\{ \begin{array}{lr} Y_t=-\xi_T+\int_{t}^{T}g(s, Y_s+X_s, Z_s, Y_{s+\alpha(s)}, Z_{s+\beta(s)})ds-\int_{t}^{T}Z_sdW_s, &t\in{[0, T]};\\ Y_t=\xi_t, & t\in{[T, T+K]};\\ Z_t=\eta_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation} $ (3.2)

为了强调对于给定过程$ X\in{\mathbb{R}^{\infty}} $的依赖关系, 我们有时候也可以用超前BSDE$ (X) $来表示超前BSDE$ (3.2) $, 用$ Y_t(X) $表示超前BSDE$ (X) $在时刻$ t $的解$ Y_{(t, T)}(X) $.由超前BSDE$ (X) $解的唯一性, 我们得到$ Y_t(X) = Y_t (\pi_{(t, T)}(X)) $, 这是符合约定的$ \rho_t(X) = \rho_t{(\pi_{(t, T)}(X))} $.对于$ s\in[t, T] $, 也用$ Y_{(t, s)}(X) $表示$ [0, s] $上的超前BSDE$ (X) $在时刻$ t $的解.相应地, $ Y_{(t, s)}(X) = Y_{(t, s)}(\pi_{(t, T)}(X)) $.

现在研究基于超前BSDEs的时间一致的过程的动态风险度量与超前BSDEs生成元之间的关系, 得到下面定理.

定理3.1   设生成元$ g $满足假设条件$ (1)\sim(5) $, 超前BSDE$ (3.2) $的解$ (Y_t(X), Z_t(X))_{t\in[0, T]} $通过$ \rho_t(X) = Y_t(X), t\in[0, T], X\in{\mathbb{R}^{\infty}} $定义了时间相容的过程的动态凸风险度量.

定理3.2   设生成元$ g $满足上述假设条件$ (1), (2), (4), (6), (7) $, 超前BSDE$ (3.2) $的解$ (Y_t(X) $, $ Z_t(X))_{t\in[0, T]}$通过$ \rho_t(X) = Y_t(X), t\in[0, T], X\in{\mathbb{R}^{\infty}} $定义了时间相容的过程的动态一致性风险度量.

4 主要结论的证明

本节将给出上面两个定理的证明过程.为证上述定理, 我们首先给出下面一个引理.

引理4.1   设生成元$ g $满足上述假设条件$ (1), (2) $并且$ X\in{S^2} $, 那么存在唯一一对适应过程$ (Y_t(X), Z_t(X))_{t\in[0, T]}$是超前BSDE$ (3.2) $的解.

  对任意$ X\in{S^{2}} $, 定义一个新的函数$ l^X:\Omega\times{R}^{m}\times{R}^{m\times{d}}\times{{L_\mathcal{F}^2}{(t, T+K;{R}^{m})}}\times{{L_\mathcal{F}^2}{(t, T+K;{R}^{m\times{d}})}}\rightarrow{L_2{(\mathcal{F}_t, {R}^{m})}} $,

$ \begin{equation} \ l^X(t, y, z, \xi, \eta) = g(t, y+X_t, z, \xi, \eta).t\in[0, T], (y, z)\in{R\times{R^d}}. \end{equation} $ (4.1)

明显地, $ l^X(t, y, z, \xi, \eta)_{t\in[0, T]}$是循序可测的, 并且容易验证$ l^X $满足假设$ (1) $.根据文献[13], 要想证明超前BSDE$ (3.2) $有唯一解, 只要证明$ l^X $满足假设$ (2) $.事实上, 对任意$ X\in{S^{2}} $, 由假设$ (1) $可以得到

$ \begin{equation*} \begin{aligned} l^X(t, 0, 0, 0, 0) & = g(t, 0+X_t, 0, 0, 0) = g(t, 0, 0, 0, 0)+g(t, X_t, 0, 0, 0)-g(t, 0, 0, 0, 0) \\ & \leq{g(t, 0, 0, 0, 0)}+\varphi_1(t)|X_t|. \end{aligned} \end{equation*} $

因此,

$ \begin{equation*} \begin{aligned} E_\mathbb{P} [\int_{0}^{T}|l^X(t, 0, 0, 0, 0)|^2dt] & \leq2E_\mathbb{P}[\int_{0}^{T}|g(t, 0, 0, 0, 0)|^2dt]+2E_\mathbb{P}[(\int_{0}^{T}\varphi_1(t)|X_t|dt)^2] \\ & \leq2E_\mathbb{P}[\int_{0}^{T}|g(t, 0, 0, 0, 0)|^2dt]+2E_\mathbb{P}(\sup\limits_{(t\in[0, T])}|X_t|)^2(\int_{0}^{T}\varphi_1(t)dt)^2. \end{aligned} \end{equation*} $

$ \int_{0}^{T}\varphi_1(t)dt<\infty $以及$ X\in{S^{2}} $可以得到

$ \begin{equation*} \ E_\mathbb{P}[\int_{0}^{T}|l^X(t, 0, 0, 0, 0)|^2dt]<\infty. \end{equation*} $

因此, $ l^X $满足假设$ (2) $, 引理得证.

定理3.1的证明  为证明凸性, 设$ X', X''\in{\mathbb{R}^{\infty}}, (Y', Z'), (Y'', Z'') $满足下面超前BSDEs:

$ \begin{equation*} \left\{ \begin{array}{lr} \ Y'_t=-\xi'_T+\int_{t}^{T}g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})ds-\int_{t}^{T}Z'_sdW_s, &t\in{[0, T]};\\ \ Y'_t=\xi'_t, & t\in{[T, T+K]};\\ \ Z'_t=\eta'_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $
$ \begin{equation*} \left\{ \begin{array}{lr} \ Y''_t=-\xi''_T+\int_{t}^{T}g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}Z''_sdW_s, &t\in{[0, T]};\\ \ Y''_t=\xi''_t, & t\in{[T, T+K]};\\ \ Z''_t=\eta''_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $

设:

$ \begin{equation*} \ \hat{X} = \lambda{X'}+(1-\lambda)X''; \quad \ \hat{Y} = \lambda{Y'}+(1-\lambda)Y''; \quad \ \hat{Z} = \lambda{Z'}+(1-\lambda)Z''; \end{equation*} $
$ \begin{equation*} \ \hat{Y}_{t+\alpha(t)} = \lambda{Y'_{t+\alpha(t)}}+(1-\lambda)Y''_{t+\alpha(t)}; \quad \ \hat{Z}_{t+\beta(t)} = \lambda{Z'_{t+\beta(t)}}+(1-\lambda)Z''_{t+\beta(t)}. \end{equation*} $

其中$ \lambda\in{L^2 (\Omega, \mathcal{F}_t, \mathbb{P})} $$ \lambda\in[0, T] $, 对任意的$ t \in [0, T] $, 由假设$ (5) $可以得到:

$ \begin{equation*} \begin{aligned} \ g(t, \hat{Y}_t+\hat{X}_t, \hat{Z}_t, \hat{Y}_{t+\alpha(t)}, \hat{Z}_{t+\beta(t)})\leq & {\lambda}{g(t, Y'_t+X'_t, Z'_t, Y'_{t+\alpha(t)}, Z'_{t+\beta(t)})} \\ \ & +(1-\lambda){g(t, Y''_t+X''_t, Z''_t, Y''_{t+\alpha(t)}, Z''_{t+\beta(t)})}. \end{aligned} \end{equation*} $

因此, 对任意的$ t \in [0, T] $, 由超前BSDEs的比较定理得到:

$ \begin{align*} \ \hat{Y}_t & = \lambda{Y_t(X')}+(1-\lambda){Y_t(X'')} \\ \ & = -\lambda{X'_T}+\lambda\int_{t}^{T}g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})ds-\lambda\int_{t}^{T}Z'_sdW_s-(1-\lambda)X''_T \\ \ &{\phantom{-}\;}+(1-\lambda)\int_{t}^{T}g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-(1-\lambda)\int_{t}^{T}Z''_sdW_s \\ \ & = -(\lambda{X'_T}+(1-\lambda)X''_T)+\int_{t}^{T}\lambda{g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})} \\ \ &{\phantom{-}\;}+(1-\lambda)g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}\lambda{Z'_s}+(1-\lambda){Z''_s}dW_s \\ \ & = Y_t(\hat{X}). \end{align*} $

$ Y_t(\hat{X}) = Y_t(\lambda{X'}+(1-\lambda) X'') $

$ \begin{equation*} \begin{aligned} \ \lambda{Y_t(X')}+(1-\lambda){Y_t(X'')}\geq{Y_t(\lambda{X'}+(1-\lambda) X'')}. \end{aligned} \end{equation*} $

因此, 凸性得证.

为证单调性, 对任意的$ t \in [0, T] $, 假设$ X', X''\in{\mathbb{R}^{\infty}} $, 并且$ X''\leq{X'} $.因此对$ \forall{y\in\mathbb{R}^{m}} $, $ y+X''_t\leq{y+X'_t} $, 由$ g $是非增的可得

$ \begin{equation*} \begin{aligned} \ l^{X''}(t, y, z, \xi, \eta) & = g(t, y+X''_t, z, \xi, \eta) \geq{g(t, y+X'_t, z, \xi, \eta)} = l^{X'}(t, y, z, \xi, \eta). \end{aligned} \end{equation*} $

成立a.s.$ d\mathbb{P}\times{dt} $.

根据超前BSDEs的比较定理得到: $ Y_t(X'')\geq{Y_t(X')} $.单调性得证.

为证条件现金不变性, 设$ X\in{\mathbb{R}^{\infty}} $, $ {m}\in{L^\infty}(\Omega, \mathcal{F}_t, \mathbb{P}) $, 考虑如下形式的超前BSDE:

$ \begin{equation*} \left\{ \begin{array}{lr} \ Y'_t\! = -\xi_T\!-m\!+\int_{t}^{T}g(s\!, Y'_s\!+X'_s\!+m, Z'_s\!, Y'_{s+\alpha(s)}\!, Z'_{s+\beta(s)}\!)ds-\int_{t}^{T}Z'_s\!dW_s, & t\!\in{[0, T]};\\ \ Y'_t = \xi'_t, & t\in{[T, T+K]};\\ \ Z'_t = \eta'_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $

因此

$ \begin{equation*} \left\{ \begin{array}{lr} \ Y'_t\!+m\! = -\xi_T\!+\int_{t}^{T}g(s\!, (Y'_s\!+m)+X'_s\!, Z'_s\!, Y'_{s+\alpha(s)}\!, Z'_{s+\beta(s)}\!)ds-\int_{t}^{T}Z'_s\!dW_s, & t\!\in{[0, T]};\\ \ Y'_t = \xi'_t, & t\in{[T, T+K]};\\ \ Z'_t = \eta'_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $

由此可知$ (Y'+m, Z') $是超前BSDE$ (3.2) $的解.由超前BSDE$ (3.2) $解的存在性与唯一性得

$ \begin{equation*} \ Y'_t+m = Y_t(X), Y'_t = Y_t(X+m1_{[t, T]}). \end{equation*} $

因此, $ Y_t(X)-m = Y_t(X+m1_{[t, T]}) $.条件现金不变性得证.

正则性由$ g(t, 0, 0, 0, 0) = 0, $ a.s.$ d\mathbb{P}\times{dt} $和方程解的存在性与唯一性可得.

下面证明时间相容性.对任意的$ s\in[t, T] $, 记$ (Y, Z) $是超前BSDE$ (3.2) $的解.由解得存在性与唯一性知

$ \begin{align*} \ Y_{t, T}(X) & = -X_T+\int_{s}^{T}g(k, Y_k+X_k, Z_k, Y_{k+\alpha(k)}, Z_{k+\beta(k)})dk-\int_{s}^{T}Z_kdW_k \\ \ &{\phantom{-}\;}+\int_{t}^{s}g(k, Y_k+X_k, Z_k, Y_{k+\alpha(k)}, Z_{k+\beta(k)})dk-\int_{t}^{s}Z_kdW_k \\ \ & = Y_{s, T}(X)+\int_{t}^{s}g(k, Y_k+X_k, Z_k, Y_{k+\alpha(k)}, Z_{k+\beta(k)})dk-\int_{t}^{s}Z_kdW_k \\ \ & = Y_{t, s}(X1_{[t, s)}-Y_{s, T}(X)1_{[s]}). \end{align*} $

$ X'' = X1_{[t, s)}-Y_{s, T}(X)1_{[s, T]} $, $ (Y'', Z'') $是超前BSDE$ (3.2) $关于$ X = X'' $时的解, 则

$ \begin{align*} \ Y''_t & = Y_{s, T}(X)+\int_{t}^{s}g(k, Y''_k+X_k, Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk-\int_{t}^{s}Z''_kdW_k \\ \ &{\phantom{Y}\;}+\int_{s}^{T}g(k, Y''_k-Y_{k, T}(X), Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk-\int_{s}^{T}Z_kdW_k \\ \ & = Y_{s, T}(X)+\int_{t}^{s}g(k, Y'_k+X_k, Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk-\int_{t}^{s}Z''_kdW_k \\ \ &{\phantom{Y}\;}+(Y_{s, T}(X)+\int_{s}^{T}g(k, Y''_k-Y_{k, T}(X), Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk-\int_{s}^{T}Z_kdW_k) -Y_{s, T}(X). \end{align*} $

由解的存在性与唯一性可知

$ \begin{align*} \ Y_{t, T}(X) = Y_{s, T}(X)+\int_{t}^{s}g(k, Y''_k+X_k, Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk-\int_{t}^{s}Z''_kdW_k, \end{align*} $

以及

$ \begin{align*} Y_{s, T}(-Y_{s, T}(X)1_{[s, T]})& = Y_{s, T}(X)+\int_{s}^{T}g(k, Y''_k-Y_{k, T}(X), Z''_k, Y''_{k+\alpha(k)}, Z''_{k+\beta(k)})dk -\int_{t}^{s}Z''_kdW_k. \end{align*} $

由条件现金不变性和正则性知

$ \begin{align*} \ Y_{s, T}(-Y_{s, T}(X)1_{[s, T]}) = Y_{s, T}(X). \end{align*} $

因此, $ Y_t(X) = Y_t(X1_{[t, s)}-Y_{s, T}(X)1_{[s, T]}) $.时间相容性得证.

定理3.2的证明  条件现金不变性和单调性以及时间相容性由定理1的证明可得, 下面证明次可加性和条件正齐次性.

首先证明次可加性.设$ X', X''\in{\mathbb{R}^{\infty}} $, $ (Y', Z' ), (Y'', Z'' ) $满足下面超前BSDEs

$ \begin{equation*} \left\{ \begin{array}{lr} \ Y'_t=-\xi'_T+\int_{t}^{T}g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})ds-\int_{t}^{T}Z'_sdW_s, &t\in{[0, T]};\\ \ Y'_t=\xi'_t, & t\in{[T, T+K]};\\ \ Z'_t=\eta'_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $
$ \begin{equation*} \left\{ \begin{array}{lr} \ Y''_t=-\xi''_T+\int_{t}^{T}g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}Z''_sdW_s, &t\in{[0, T]};\\ \ Y''_t=\xi''_t, & t\in{[T, T+K]};\\ \ Z''_t=\eta''_t, & t\in{[T, T+K]}. \end{array} \right. \end{equation*} $

$ \begin{equation*} \ \check{X} = {X'}+X''; \quad \ \check{Y} = {Y'}+Y''; \quad \ \check{Z} = {Z'}+Z''; \end{equation*} $
$ \begin{equation*} \ \check{Y}_{t+\alpha(t)} = \lambda{Y'_{t+\alpha(t)}}+(1-\lambda)Y''_{t+\alpha(t)}; \quad \ \check{Z}_{t+\beta(t)} = \lambda{Z'_{t+\beta(t)}}+(1-\lambda)Z''_{t+\beta(t)}. \end{equation*} $

由假设$ (6) $知, 对任意的$ t \in [0, T] $

$ \begin{equation*} \begin{aligned} \ g(t, \check{Y}_t+\check{X}_t, \check{Z}_t, \check{Y}_{t+\alpha(t)}, \check{Z}_{t+\beta(t)})\leq & {g(t, Y'_t+X'_t, Z'_t, Y'_{t+\alpha(t)}, Z'_{t+\beta(t)})} \\ \ & +{g(t, Y''_t+X''_t, Z''_t, Y''_{t+\alpha(t)}, Z''_{t+\beta(t)})}. \end{aligned} \end{equation*} $

因此, 由超前BSDEs的比较定理, 对任意的$ t \in [0, T] $

$ \begin{align*} \ \check{Y}_t & = {Y_t(X')}+{Y_t(X'')} \\ \ & = -{X'_T}+\int_{t}^{T}g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})ds-\int_{t}^{T}Z'_sdW_s \\ \ &{\phantom{-}\;}-X''_T+\int_{t}^{T}g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}Z''_sdW_s \\ \ & = -({X'_T}+X''_T)+\int_{t}^{T}{g(s, Y'_s+X'_s, Z'_s, Y'_{s+\alpha(s)}, Z'_{s+\beta(s)})} \\ \ &{\phantom{-}\;}+g(s, Y''_s+X''_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}{Z'_s}+{Z''_s}dW_s \\ \ & = Y_t(\check{X}). \end{align*} $

$ Y_t(\check{X}) = Y_t(X'+ X'') $

$ \begin{equation*} \begin{aligned} \ {Y_t(X')}+{Y_t(X'')}\geq{Y_t(X'+ X'')}. \end{aligned} \end{equation*} $

次可加性得证.

现在证明条件正齐次性, 设$ X'\in{\mathbb{R}^{\infty}} $, $ t \in [0, T] $, $ r\in L^\infty (\Omega, \mathcal{F}_t, \mathbb{P}) $, $ r\geq 0 $并且$ X'' = rX' $, 由此可知$ X''\in{\mathbb{R}^{\infty}} $.设$ (Y', Z' ), (Y'', Z'' ) $是超前BSDE$ (3.2) $的解, 对任意的$ t \in [0, T] $, 由假设$ (7) $可知

$ \begin{equation*} \ g(t, rY'_t+rX'_t, rZ'_t, rY'_{t+\alpha(t)}, rZ'_{t+\beta(t)}) = r{g(t, Y'_t+X'_t, Z'_t, Y'_{t+\alpha(t)}, Z'_{t+\beta(t)})}. \\ \end{equation*} $
$ \begin{equation*} \ g(t, rY''_t+rX''_t, rZ''_t, rY''_{t+\alpha(t)}, rZ''_{t+\beta(t)}) = r{g(t, Y''_t+X''_t, Z''_t, Y''_{t+\alpha(t)}, Z''_{t+\beta(t)})}. \\ \end{equation*} $

因此, 对任意的$ t \in [0, T] $,

$ \begin{align*} rY_t(X') & = rY'_t = -rX'_T+\int_{t}^{T}g(s, rY'_s+rX'_s, rZ'_s, rY'_{s+\alpha(s)}, rZ'_{s+\beta(s)})ds-\int_{t}^{T}rZ'_sdW_s. \\ \ Y_t(rX') & = Y''_t = -rX'_T+\int_{t}^{T}g(s, Y''_s+rX'_s, Z''_s, Y''_{s+\alpha(s)}, Z''_{s+\beta(s)})ds-\int_{t}^{T}Z''_sdW_s. \end{align*} $

根据超前BSDE$ (3.2) $解的存在性与唯一性, 对任意的$ t \in [0, T] $, $ X\in{\mathbb{R}^{\infty}} $, $ r\in L^\infty (\Omega, \mathcal{F}_t, \mathbb{P}) $, $ r\geq 0 $, 有$ rY_t(X) = Y_t(rX) $.条件正齐次性得证.

5 例子及讨论

例5.1   设$ \varphi_1(t) $, $ \varphi_2(t) $, $ \phi_1(t) $, $ \phi_2(t) $是正的$ \mathcal{F}_{t} $循序可测随机过程, $ \int_{0}^{T}[\varphi_1(t)+\phi_1^2(t)+\varphi_2(t)+\phi_2^2(t)]dt<\infty $.考虑$ g $有如下定义:

$ \begin{equation} \ g(t, y, z, \xi, \eta) = -a\varphi_1(t)y+b\phi_1(t)|z|+e\varphi_2(t)E^{\mathcal{F}_t}(\xi)+f{\phi_2(t)}E^{\mathcal{F}_t}(\eta). \end{equation} $ (5.1)

$ a, b, e, f\in R $并且大于等于零.设$ (Y_t(X), Z_t(X))_{t\in[0, T]}$是超前BSDE$ (3.2) $的解, 定义$ \rho_t(X) = Y_t(X), t\in[0, T], X\in{\mathbb{R}^{\infty}} $.那么$ (\rho_t)_{t \in [0, T]} $是时间相容的过程的动态一致性(凸)风险度量.

例5.2   设$ a\geq 0 $并且$ a\in R $.考虑$ g $有如下定义:

$ |z|\leq 1 $时,

$ \begin{equation} \ g(t, y, z, \xi, \eta) = -ay+z^2+\xi+\eta. \end{equation} $ (5.2)

$ |z|>1 $时,

$ \begin{equation} \ g(t, y, z, \xi, \eta) = -ay+2z-1+\xi+\eta. \end{equation} $ (5.3)

$ (Y_t(X), Z_t(X))_{t\in[0, T]}$是超前BSDE$ (3.2) $的解, 定义$ \rho_t(X) = Y_t(X), t\in[0, T], X\in{\mathbb{R}^{\infty}} $.那么$ (\rho_t)_{t \in [0, T]} $仅仅是时间相容的过程的动态凸风险度量.

由上面两个例子可以知道, 时间相容的过程的动态一致性风险度量也是一个时间相容的过程的动态凸风险度量, 反之不成立.因此, 在对金融市场做风险评估时, 更多的选择凸风险度量.由于超前BSDEs生成元中不仅包含当前时刻的解的情况, 也包含未来时刻的解的情况, 因此其可以把将来不确定的目标变为当前确定的结果来做出当前的决定.通过超前BSDEs定义的时间相容的过程的风险度量不仅可以对金融市场的当前的风险做更加准确评估, 同时可以预测未来时刻存在的风险, 这在金融市场中有着十分广阔的应用前景.

参考文献
[1] Artzner P, Delbaen F, Eber J M, Heath D. Coherent measures of risk[J]. Mathematical Finance, 1999, 9(3): 203–228. DOI:10.1111/1467-9965.00068
[2] Föllmer H, Schied A. Convex measures of risk and trading constraints[J]. Finance and Stochastics, 2002, 6(4): 429–447. DOI:10.1007/s007800200072
[3] Frittelli M, Rosazza Gianin E. Putting order in risk measures[J]. Journal of Bank and Finance, 2002, 26(7): 1473–1486. DOI:10.1016/S0378-4266(02)00270-4
[4] Karoui N E, Ravanelli C. Cash subadditive risk measures and interest rate ambiguity[J]. Mathematical Finance, 2009, 19(4): 561–590. DOI:10.1111/j.1467-9965.2009.00380.x
[5] Cheridito P, Delbaen F, Kupper M. Coherent and convex monetary risk measures for bounded càdlàg processes[J]. Stochastic Process and their Application, 2004, 112(1): 1–22.
[6] Riedel F. Optimal stopping with multiple priors[J]. Econometrica, 2009, 77(3): 857–908. DOI:10.3982/ECTA7594
[7] Bayraktar E, Karatzas I, Yao Song. Optimal stopping for dynamic convex risk measures[J]. Illinois Journal of Mathematics, 2010, 54(3): 1025–1067. DOI:10.1215/ijm/1336049984
[8] Cheng Xue, Riedel F. Optimal stopping under ambiguity in continuous time[J]. Mathematics and Financial Economics, 2013, 7(1): 29–68.
[9] Acciaio B, Föllmer H, Penner I. Risk assessment for uncertain cash flows:model ambiguity, discounting ambiguity, and the role of bubbles[J]. Finance and Stochastics, 2012, 16(4): 669–709. DOI:10.1007/s00780-012-0176-1
[10] Jiang Long. Convexity, translation invariance and subadditivity for g-expectations and related risk measures[J]. The Annals of Applied Probability, 2008, 18(1): 245–258. DOI:10.1214/105051607000000294
[11] Penner I, Réveillac A. Risk measures for processes and BSDEs[J]. Finance and Stochastics, 2015, 19(1): 23–66. DOI:10.1007/s00780-014-0243-x
[12] Peng Shige, Yang Zhe. Anticipated backward stochastic differential equations[J]. The Annals of Probability, 2009, 37(3): 877–902. DOI:10.1214/08-AOP423
[13] Xu Xuecheng, Chen Min. Solutions and comparison theorems for anticipated backward stochastic differential equations with stochastic Lipschitz generators[J]. Journal of Advances in Mathematics, 2016, 12(3): 6139–6147.
[14] Ji Ronglin, Shi Xuejun, Wang Shijie, Zhou Jinming. Dynamic risk measures for processes via backward stochastic differential equations[J]. Insurance:Mathematics and Economics, 2019, 86(MAY): 43–50.