数学杂志  2026, Vol. 46 Issue (5): 296-310   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
阮中杰
罗翠翠
随机波动率和一般折现函数下动态投资-消费问题的开环均衡解
阮中杰1, 罗翠翠2    
1. 中国科学院大学数学科学学院, 北京市, 100043;
2. 中国科学院大学国际学院, 北京市, 100190
摘要:本文研究了一个具有随机波动率和一般折现函数的动态投资-消费问题.金融市场由两类资产构成: 一类为无风险资产; 另一类为风险资产, 其价格过程服从随机波动率模型. 投资者偏好通过指数效用函数建模, 而折现函数则由一般非指数型折现函数描述.本文利用正倒向随机微分方程(FBSDE)的方法, 获得了开环均衡策略存在的必要条件.进一步地, 在FBSDE解存在的前提下, 本文推导出了开环均衡策略的验证定理.
关键词:随机波动率    动态投资-消费问题    正倒向随机微分方程    开环均衡策略    
OPEN-LOOP EQUILIBRIUM STRATEGY FOR DYNAMIC INVESTMENT-CONSUMPTION PROBLEM UNDER STOCHASTIC VOLATILITY AND GENERAL DISCOUNT FUNCTIONS
RUAN Zhong-jie1, LUO Cui-cui2    
1. School of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China;
2. International College, University of Chinese Academy of Sciences, Beijing 100190, China
Abstract: This paper studies a dynamic investment-consumption problem with stochastic volatility and general discount functions. The financial market consists of two assets: a risk-free asset and a risky asset whose price dynamics follow a stochastic volatility model. The investor's preferences are modeled using an exponential utility function, while future utility is discounted via general non-exponential discount functions. By employing forward-backward stochastic differential equations (FBSDEs), we derive necessary conditions for the existence of an open-loop equilibrium strategy. Furthermore, under the assumption that the FBSDE admits a solution, we establish a verification theorem for the open-loop equilibrium strategy.
Keywords: Stochastic volatility     Dynamic investment-consumption problem     forward-backward stochastic differential equations (FBSDEs)     Open-loop equilibrium strategy    
1 引言

折现效用模型(Discounted Utility, DU)是现代经济学的奠基性理论之一. Ramsey[1]在1928年以及Samuelson[2]在1937年都对该理论做出了开创性贡献. 其模型核心假设为常数折现率, 并采用指数折现函数以实现跨期效用的可比性. 然而, 实证研究[3, 4]表明, 决策者往往在未来不同时刻具有不一样的偏好, 即短期决策更具紧迫感而长期决策更为耐心, 这使得双曲型折现模型更能刻画实际行为, 同时也引出了折现率时变所导致的时间不一致性问题.

具体而言, 时间不一致性意味着当期制定的最优策略在未来时刻将不再最优, 从而导致基于贝尔曼原理的随机动态规划方法失效. Storze[5]对时间不一致性这个问题进行过研究, 并提出了时间一致的策略. 该策略将未来不同时刻的决策者视为不同的玩家, 不同玩家之间进行博弈, 其中后续决策者具有先行优势, 由此可以得到纳什均衡下的均衡策略. 该策略是时间一致的, 满足任意时刻的局部最优, 但不再是全局最优. Ekeland和Pirvu[6]首次在连续情形给出了均衡控制的定义, 并对Merton问题[8]进行求解. Bjork[9]在此定义下给出了一般情形下的时间不一致随机控制问题的广义HJB方程. 上述控制属于闭环控制均衡策略, 然而闭环控制策略的唯一性问题仍未得到圆满解决. Hu等[10, 11]提出开环控制的定义并首次证明了开环均衡策略的唯一性. 随后, Haung等[12]研究均值-方差的闭环控制解, 通过对应的时间一致问题解的策略的唯一性, 证明了对应均值方差均衡策略的唯一性. Zhang[13]在开环控制的定义下找到了对数收益率建模的均值方差问题的开环均衡策略.

动态资产配置问题是金融学中一个重要的课题. 其中Merton[7, 8]提出的动态投资-消费问题得到了研究者们广泛的研究. Hu等[14]利用倒向随机微分方程寻找上鞅, 并证明该上鞅在最优策略下是鞅, 由此研究了带约束的效用最优化问题. Cheridito等[15]利用上述Hu等[14]提出的方法研究指数效用, 对数效用和幂函数效用下的动态投资-消费问题. 然而上述研究仍然假定指数折现函数. Ekeland[6]指出, 在更一般的非指数折现情形下, 经典优化策略并不具备时间一致性. 他利用BSDE方法, 在闭环控制定义下推导出三类非指数折现函数对应的均衡策略. Zhao等[16]利用Yong[17]提出的N人微分博弈再取极限的方法研究了对数效用和非指数折现函数下的动态投资-消费问题. Hamaguchi[18]利用Horst等[19]文中的方法和BSDE研究一般折现函数和一般效用函数下的时间一致投资策略.

实际金融市场的动态通常并非由确定性参数所驱动. 为了更真实地刻画市场特征, 学界普遍采用随机波动率模型进行建模. Zariphopoulou等[20]和Kraft[21]在随机波动率下研究时间一致的期望效用理论问题. Han[22]在粗糙波动率模型下研究效用最大化问题. 本文的主要创新在于将非指数折现下的时间一致控制问题推广至随机波动率模型. 该模型能更真实地捕捉波动率聚集、杠杆效应等市场特征, 但风险溢价的无界性为分析带来本质困难. 为此, 本文提出风险溢价指数有界的关键假设, 拓展了Hamaguchi[18] 的理论框架, 在更一般的市场设定下系统研究指数与非指数折现对应的开环均衡策略存在条件与显式表达, 并给出严格的验证定理. 本文的结构如下. 第二节提出金融市场中的资产优化问题;第三节给出开环均衡策略存在的必要条件. 第四节验证了第三节给出的策略属于开环均衡解. 第五节总结全文.

2 问题建模

令$ T>0 $是有限的时间长度. $ (\Omega,\mathcal{F},P) $是附带域流的完备概率空间. 令$ (W_{1,t},W_{2,t}) $是概率空间上的相关系数$ \mathrm{Cov}[dW_{1,t},dW_{2,t}]=\rho\, dt $的两个布朗运动. 记$ {\bf 1}_A $是关于集合A的示性函数, Leb是$ \mathbb{R} $上的Lebesgue测度. 令$ n,p\in \mathbb{N} $, $ 0\leq s<t\leq T $, 定义

$ \begin{align*} L^p_{\mathcal{F}_t}(\Omega;\mathbb{R}):=&\left\{X:\Omega \to \mathbb{R}\, |\, X\text{是} \mathcal{F}_t\text{可测的随机变量, 满足}E|X|^p<+\infty \right\} ; \notag \\ L^{\infty}_{\mathcal{F}_t}(\Omega;\mathbb{R}^p):=&\bigg\{X:\Omega \to \mathbb{R}^p \,|\, X\text{是} \mathcal{F}_t\text{可测的随机变量, 满足} \exists\, C>0, s.t. P(|X|>C)=0\bigg\}; \notag \\ L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}^n):=&\bigg\{X:[s,t]\times \Omega\to\mathbb{R}^n \,|\,X(\cdot) \text{可测, 是}\mathbb{F}\text{-适应, 路径连续的随机过程, }; \notag \\ &\ \text{满足} E\left( \sup\limits_{r\in[s,t]} |X(r)|^p \right)<+\infty \bigg\}; \notag \\ L^p_{\mathbb{F}}(\Omega;L^q(s,t,\mathbb{R}^n)) =: &\bigg\{X:[s,t]\times \Omega \to\mathbb{R}^n \,|\,X(\cdot)\text{可测, }\mathbb{F}\text{-适应的随机过程, 满足} \notag \\ &\ E\left(\int^t_s|X_r|^qdr\right)^{\frac{p}{q}} <+\infty\bigg\}. \notag \end{align*} $

考虑由两个资产组成的金融市场. 一个是无风险资产, 其价格过程$ B(t) $服从如下过程:

$ \begin{equation} \begin{split} \begin{cases} dB_t=r_tB_tdt \\ B(0)=b_0>0, \end{cases} \end{split} \end{equation}$ (2.1)

其中$ r_t>0 $是确定性且有界的无风险利率, 初始值$ b_0 $是已知常数. 另一个是风险资产, 其价格过程$ S_t $服从下列随机波动率模型:

$\begin{equation} \begin{split} \begin{cases} dS_t=\mu(t,S_t,V_t)dt +\sigma(t.,S_t,V_t)dW_1(t) \\ S_0=s_0>0 \\ dV_T=a(t,V_t)dt+b(t,V_t)dW_2(t) \\ V_0=v_0>0, \end{cases} \end{split} \end{equation} $ (2.2)

在$ t $时刻, 投资者的财富是$ x_t $, 投资于股票的财富金额是$ \pi_t $. 投资者的消费金额为$ c_t $. 由此, 投资者的财富过程$ X(t)=X^{(c,\pi,t,x_t)} $满足

$\begin{equation} \begin{split} dX_t&=(X_t-\pi_t)\frac{dB_t}{dt}+\pi_t\frac{dS_t}{S_t}+(e_t-c_t)dt \\ &=(r_tX_t+e_t-c_t)dt+\theta_tu_t dt+u_t dW_1(t), \end{split} \end{equation} $ (2.3)

其中$ \theta_t=\left(\mu\left(t, S_t, V_t\right)-r_t S_t\right) / \sigma\left(t, S_t, V_t\right) $是风险溢价过程, 且$ u_t=\sigma\left(t, S_t, V_t\right) S_t^{-1} \pi_t $.

$ \mathbb{U} $是效用函数组成的集合, 若$ U:\mathbb{R}\to \mathbb{R}\in \mathbb{U} $是三次可导, 严格递增, 严格凸, 且满足Inada条件:$ \lim_{x\to 0}U^\prime (x)=0 $, $ \lim_{x\to}U^\prime (x)=+\infty $, 则$ U\in\mathbb{U} $. 本文取指数效用函数, 对消费的效用函数取$ U_1=-e^{-\gamma_1 x} $, 对终期财富的效用函数取$ U_2=-e^{-\gamma_2 x} $. 简单计算可知,

$ \begin{equation*} \frac{U'_i(x)}{U''_i(x)}=-\frac{1}{\gamma_i},\ \frac{U^{(3)}_i(x)(U'_i(x))^2}{(U''_i(x))^3}=-\frac{1}{\gamma_i},\ \frac{U^{(3)}_i(x)}{U''_i(x)}=-\gamma_i, \quad i=1,2, \end{equation*} $

以及

$ \begin{equation*} (U'_1)^{-1}(\lambda_2(s,T)U'_2(x))=\frac{\gamma_2}{\gamma_1}x-\frac{1}{\gamma_1}\log\left(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T)\right). \end{equation*} $

$ \Lambda $是折现函数构成的集合. 令$ \Delta [0,T] :=\left\{ (t,s)|0\leq t \leq s\leq T \right\} $. 若$ \lambda:\Delta[0,T]\to \mathbb{R}_+ $连续, 严格正, 且满足$ \lambda(t,t)=1 $, $ \,\forall t\in[0,T] $, 则$ \lambda\in\Lambda $. 一些举例如下

● 指数折现:$ \lambda(t,s)=e^{-\delta(s-t)} $, 其中$ \delta\ge 0 $是常系数;

● 指数折现的凸组合:$ \lambda(t,s)=\alpha e^{-\delta(s-t)}=(1-\alpha)e^{-\gamma(s-t)} $, 其中$ \alpha\in(0,1) $, $ \delta,\gamma>0 $是常系数且$ \delta>0 $;

● 准-指数折现:$ \lambda(t,s)=(1+\alpha(s-t))e^{-\delta (s-t)} $ $ \alpha,\delta>0 $;

● 双曲指数折现:$ \lambda(t,s)=\frac{1}{q+\delta(s-t)} $, $ \delta>0 $相对时间依赖的指数折现:$ \lambda(t,s)=e^{-\delta(t)(s-t)}, $ $ \delta:[0,T]\to \mathbb{R}^+ $是非负连续函数.

给定$ \lambda_1,\lambda_2\in\Lambda $, 对任意初始条件$ (t,x_t)\in [0,T)\times \mathbb{R} $, 为了使得财富过程满足可积性条件, 本文对消费-投资策略$ (c,u) $作出如下假设:

$ ({\bf H0}) $ $ \,c(\cdot),u(\cdot) $都是取值于$ \mathbb{R} $的可料过程, 且满足

$\begin{equation} \begin{split} \int^T_t (|e_s|+|c_s|)ds<+\infty \,\,a.s.,\ \ E\left[\int^T_t |u_s|^2 ds\right] <+\infty ,\ \ \\ E\left[\int^T_t \lambda_1(t,s)U_1(c_s)ds +\lambda_2(t,T)U_2(X_T^{(c,u)}+E)\right]<+\infty. \end{split} \end{equation}$ (2.4)

投资者寻找最优策略使得最大化如下目标泛函

$\begin{equation} \begin{split} R(c,\pi,;t,x_t):= E_t\left[ \int^T_t\lambda_1(t,s)U_1(c_s)ds+\lambda_2(t,T)U_2(X_T^{(c,u)}+E) \right], \end{split} \end{equation}$ (2.5)

代入$ U_i(x)=-\exp(-\gamma_i x),\quad i=1,2 $可得

$\begin{equation} \begin{split} R(c,\pi,;t,x_t):=- E_t\left[ \int^T_t\lambda_1(t,s)e^{-\gamma_1 c_s}ds+\lambda_2(t,T)e^{-\gamma_2 (X_T^{(c,u)}+E)} \right]. \end{split} \end{equation} $ (2.6)

给定控制$ (c^*,u^*) $, 任意$ t\in[0,T] $, $ \epsilon\ge 0 $, 定义如下形式的开环控制:

$\begin{equation} \begin{split} \begin{cases} c^{t,\epsilon,\kappa}_s:=c^*_s+\kappa{\bf 1}_{s\in[t,t+\epsilon]} \\ u^{t,\epsilon,\eta}:=u^*_s+\eta{\bf 1}_{s\in[t,t+\epsilon]}, \end{cases} \end{split} \end{equation} $ (2.7)

其中$ (\kappa,\eta)\in L^\infty_{\mathcal{F}_t}(\Omega;\mathbb{R}^2) $.

定义2.1  给定$ \lambda_1,\lambda_2\in \Lambda $, 给定$ u^*\in L^2_{\mathcal{F}_t}(\Omega;\mathbb{R}) $, $ X^*\in L^2_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}) $是对应的财富过程. 若对任意时刻$ t\in[0,T) $, 下式成立,

$\begin{equation} \begin{split} \limsup\limits_{\epsilon\to 0}\frac{R(c^{t,\epsilon,\kappa},u^{t,\epsilon,\eta};t,X^*_t)-R(c^{*},u^{*};t,X^*_t)}{\epsilon}\leq0 \,\, a.s., \end{split} \end{equation}$ (2.8)

则称$ (c^*.u^*) $是开环均衡解.

在验证开环均衡解的过程中, 需要处理极限问题, 在此过程中为了给定界限, 消除无关的二次项, 需要给出如下假设:

$ ({\bf H1}) $ 任意给定$ p>1 $,

$\begin{equation} E\left[ \int^T_0 (\gamma_1e^{-\gamma_1c_s})^pds+ (\gamma_2e^{-\gamma_2(X_T+E)})^p \right]<\infty. \end{equation} $ (2.9)

由Hamaguchi[18]文中引理2.6可知, 在指数效用函数下, 上述条件等价于存在$ q>1 $使得

(1) 对任意$ \delta>0 $, $ E\left[\int^T_0 M_1(c_s;\delta)^q ds\right]<\infty $;

(2) 对任意时间$ t\in[0,T) $, $ \mathcal{F}_T- $可测随机变量族

$\begin{equation} \left\{ M_2(X_T+E;|\xi_T|^q) \right\}_{\epsilon\in(0,T-t)} \end{equation}$ (2.10)

是一致可积的. 其中

$\begin{equation} M_i(x;\delta):=max_{y\in \mathbb{R},|y|\leq \delta} |\gamma_i^2e^{-\gamma_i (x+y)}| \ \ x\in\mathbb{R},\ \delta\ge 0,\ i=1,2. \end{equation}$ (2.11)

最后, 本文给无界参数$ \theta(\cdot) $下列幂次阶和指数阶有界的假设. 该假设在文献[23, 24]中都有涉及.

$ ({\bf H2}) $ 存在$ p>1 $, 使得$ \theta^p(s)\in L^2_{\mathbb{F}}(\Omega;L^2(s,t,\mathbb{R})) $; 存在足够大$ C> 0 $, 使得$ E_t\left[\exp\{ C\int^T_t\theta^2(s)ds \}\right]<+\infty $.

对任意时刻$ t\in [0,T) $, 任意$ \epsilon\in (0,T-t) $, 定义$ \xi=X^{\epsilon,\eta}-X^* $

$\begin{equation} \begin{split} d\xi^{t,\epsilon}_s =\left(r_s\xi^{t,\epsilon}_sds-\kappa{\bf 1}_{[t,t+\epsilon)}(s)+\theta_s\eta{\bf 1}_{[t,t+\epsilon)} (s)\right) ds + \eta{\bf 1}_{[t,t+\epsilon)} (s) dW_{1}(t). \end{split} \end{equation}$ (2.12)

引理2.1  令$ t\in [0,T] $, $ (\kappa,\eta)\in L^{\infty}_{\mathcal{F}_t}(\Omega;\mathbb{R}^2) $,

(1) 对任意$ \gamma\ge \frac{1}{2} $, 存在常数$ C_\gamma=C(\gamma,T,||r||_{\infty}, \left\{ E\left[ \Gamma(T)\right]^4 \right\}^{\frac{1}{4}} \sqrt{ E\left[ \Gamma^{-1}(T) \right]})>0 $, 其中$ \Gamma(T)=\exp \left(-\int_0^T \theta_s d W_{1s}-\frac{1}{2} \int_0^T \theta_s^2 d s\right) $, 使得对任意$ \epsilon\in(0,T-t) $, 下式成立

$\begin{equation} E_t\left[\sup\limits_{t\leq s \leq T} |\xi^{t,\epsilon}_s|^{2\gamma}\right] \leq (C_\gamma \epsilon(|\epsilon|^2+|\eta|^2))^{\gamma} \ \ a.s.. \end{equation} $ (2.13)

(2) 对任意$ \epsilon\in(0,T-t) $, SDE(2.12)存在唯一解$ \xi ^{t,\epsilon}\in \bigcap_{\gamma \geq 1} L_{\mathcal{F}}^\gamma(\Omega ; C([t, T] ; \mathbb{R})) $.

证  当$ s\in[t,t+\epsilon] $时,

$\begin{equation} \begin{split} d\xi^{t,\epsilon}_s =&\left(r_s\xi^{t,\epsilon}_s-\kappa+\theta_s\eta (s)\right) ds + \eta dW_{1}(t) \\ =&\left(r_s\xi^{t,\epsilon}_s-\kappa \right)ds + \eta d\tilde{W}_{1}(t), \end{split} \end{equation} $ (2.14)

其中由假设$ ({\bf H2}) $可知, Novikov条件满足, 故$ \tilde{W}_{1t} $是下列概率测度下的布朗运动,

$\begin{equation} \frac{dQ}{dP}= \Gamma(T)=\exp \left(-\int_0^T \theta_s d W_{1s}-\frac{1}{2} \int_0^T \theta_s^2 d s\right), \end{equation} $ (2.15)
$\begin{equation} \xi_{s}=\bigg[ \xi_t+\int^s_t-\kappa V^{-1}(v) dv+\int^s_t\eta V^{-1}(v)d\tilde{W}_{1v} \bigg]. \end{equation}$ (2.16)

其中$ V(v)=\exp\{\int^v_t r_sds\} $, $ \xi_t=0 $,

$\begin{equation} { \begin{split} & E\left[\sup\limits_{t\leq s\leq t+\epsilon} |\xi^{t,\epsilon}_s|^{2\gamma}{\bf 1}_A\right]\\ \leq& C_1E^Q \bigg\{\bigg[ \left(\int^{t+\epsilon}_t |\kappa| |V^{-1}(v)|{\bf 1}_A dv\right)^{2\gamma}+\left(\int^{t+\epsilon}_t |\eta|^2V^{-2}(v){\bf 1}_A dv\right)^{\gamma}\bigg] \left[ \exp \left(\int_0^T \theta_s d W_s+\frac{1}{2} \int_0^T \theta_s^2 d s\right) \right] \bigg\} \\ \leq& C_2\sqrt{E^Q\left[ |\epsilon|^{2\gamma}|\kappa|^{2\gamma}{\bf 1}_A+|\epsilon|^{\gamma}|\eta|^{2\gamma}{\bf 1}_A \right]^2} \sqrt{E^Q\left[ \exp\left(\int^T_0 2\theta_sdW_{1s}+\int^T_0\theta^2_sds \right) \right]} \\ \leq& C_3\sqrt{E\left[ \Gamma(T) \left(|\epsilon|^{2\gamma}|\kappa|^{2\gamma}{\bf 1}_A+|\epsilon|^{\gamma}|\eta|^{2\gamma}{\bf 1}_A\right) \right]^2} \sqrt{E\left[\Gamma(T) \exp\left(\int^T_0 2\theta_sdW_{1s}+\int^T_0\theta^2_sds \right) \right]} \\ \leq& C_3 \left\{ E\left[ \left(|\epsilon|^{2\gamma}|\kappa|^{2\gamma}{\bf 1}_A+|\epsilon|^{\gamma}|\eta|^{2\gamma}{\bf 1}_A\right) \right]^4 \right\}^{\frac{1}{4}} \left\{ E\left[ \Gamma(T)\right]^4 \right\}^{\frac{1}{4}} \sqrt{ E\left[ \Gamma^{-1}(T) \right] } \\ \leq& C_{\gamma}\sqrt{E\left[ |\epsilon|\left(|\kappa|^{2\gamma}+\eta|^{2\gamma}\right){\bf 1}_A \right]^2}. \end{split}} \end{equation}$ (2.17)

同理, 由假设$ ({\bf H2}) $可以知道

$ \begin{equation} \left\{ E\left[ \Gamma(T)\right]^4 \right\}^{\frac{1}{4}} \sqrt{ E\left[ \Gamma^{-1}(T) \right] }<\infty. \end{equation} $ (2.18)

由于$ A\in \mathcal{F}_t $是任意选取的, 所以可得结论(1)在$ s\in[t,t+\epsilon] $这一段上成立. 当$ s\in(t+\epsilon,T] $时,

$\begin{equation} \begin{split} d\xi^{t,\epsilon}_s =r_s\xi^{t,\epsilon}_s ds. \end{split} \end{equation}$ (2.19)

容易解得$ \xi_s=\xi_{t+\epsilon}e^{\int^s_{t+\epsilon}r_vdv} $. 由$ r_s $一致有界且可得, 在该段区间内, 结论(1)也成立. 由结论(1), 立得结论(2).

3 均衡解的必要条件

本小节给出动态投资-消费问题开环均衡策略存在的必要条件.

定理3.1  固定初始财富$ x\in \mathbb{R} $, 若$ (c^*,\pi^*) $是开环均衡解, 则存在随机过程$ (Y(\cdot),Z(\cdot)) $, 其中$ Z(\cdot)=(Z_1(\cdot),Z_2(\cdot)) $, 满足

(1) $ Y(\cdot) $是连续适应过程使得$ U_2^{'}(X^*+Y)\in L_{\mathcal{F}}^p(\Omega,C(0,T);\mathbb{R}) $, $ Z(\cdot) $是取值于$ \mathbb{R}^2 $的可料过程, 使得$ \int^T_0|Z_s|^2ds<\infty\ \ a.s. $;

(2) $ (Y(\cdot),Z(\cdot)) $满足下列倒向随机微分方程

$\begin{equation} \begin{cases} dY_s=Z_{1s} dW_{1s}+Z_{2s} dW_{2s}+f^*(s,Y_s,Z_{1s},Z_{2s})\,ds,\ s\in[0,T]\\ Y_T=0, \end{cases} \end{equation}$ (3.1)

生成元$ f^*: \Omega\times[0,T]\times\mathbb{R}\times\mathbb{R}^d\to\mathbb{R} $定义

$ \begin{equation*} \begin{split} &f^*(s,y,z_1,z_2)\\:=&-\frac{1}{2}\frac{U^{(3)}_2}{U''_2}(X^*_s+y)\left[(z_1+u^*)^2+2\rho z_2(z_1+u^*)+z_2^2\right] -r_s\frac{U'_2}{U''_2}(X^*_s+y)-r_sX^*_s-u^*_s \theta_s -e_s+c^*_s, \end{split} \end{equation*} $

其中$ s\in[0,T] $.

(3) 开环均衡策略可以如下表示

$\begin{equation} \begin{cases} c_s^*=(U^{'}_1)^{-1}\left(\lambda_2(s,T)U_2^{'}(X^*_s+Y_s)\right) \\ u^*_s=-\frac{U'_2}{U''_2}(X^*_s+Y_s)\theta_s-Z_{1s}-\rho Z_{2s}. \end{cases} \end{equation} $ (3.2)

本文模仿Hamaguchi[18]的方法给出定理3.1的证明. 首先, 本文先给出如下四个引理.

引理3.2  对任意开环均衡解$ (c^*,\pi^*) $, 存在满足定理3.1中条件(1)(2)的$ (Y(\cdot),Z(\cdot)) $使得对任意$ \epsilon\in(0,T-t) $, 下式成立, a.s.

$\begin{equation} { \begin{split} & \frac{R\left(c^{t, \epsilon}, u^{t, \epsilon} ; t, X_t^*\right)-R\left(c^*, u^* ; t, X_t^*\right)}{\epsilon} \\ =&\left(\frac{1}{\epsilon} E_t\left[\int_t^{t+\epsilon} \left(\lambda_1(t, s) U_1^{\prime}\left(c_s^*\right) -\lambda_2(t,T)U^\prime_2(X^*_s+Y_s) \right)d s\right]\right) \kappa \\ &+\lambda_2(t,T) {E}_t\bigg[\int^{t+\epsilon}_t\bigl(U^\prime_2(X^*_s+Y_s)\theta_s +U''_2(X^*_s+Y_s)(u^*_s+Z_{1s}) +\rho U''_2(X^*_s+Y_s)Z_{2s}\bigr)\,ds\bigg]\eta \\ & -\frac{1}{ \epsilon} E_t\left[\int_t^{t+\epsilon} \lambda_1(t, s) \int_0^1\left|z U_1^{\prime \prime}\left(c_s^{t,\epsilon}- z \kappa\right)\right| d z d s|\kappa|^2+\lambda_2(t, T) \int_0^1\left| z U_2^{\prime \prime}\left(X_T^{\epsilon}+E-z\xi_T^{t, \epsilon}\right)\right| d z\left|\xi_T^{t, \epsilon}\right|^2\right]. \end{split}} \end{equation}$ (3.3)

证  固定$ t\in[0,T] $, $ (\kappa,\eta)\in L^{\infty}_{\mathcal{F}_t}(\Omega;\mathbb{R}^2) $, 以及$ \epsilon\in(0,T-t) $, 定义$ X^{t,\epsilon}:=X^{c^{t,\epsilon} , \pi^{t,\epsilon}, t,X^*_t} $, $ \xi^{t,\epsilon}=X^{t,\epsilon}-X^* $是SDE(2.12)的解. 注意$ U_1^{''} $和$ U_2^{''} $是负的,

$\begin{equation} { \begin{split} \begin{aligned} & \frac{R\left(c^{t, \epsilon}, u^{t, \epsilon} ; t, X_t^*\right)-R\left(c^*, u^* ; t, X_t^*\right)}{\epsilon} \\ = &\frac{1}{\epsilon} E_t\left[\int_t^T \lambda_1(t, s)\left(U_1\left(c_s^{t, \epsilon}\right)-U_1\left(c_s^*\right)\right) d s+\lambda_2(t, T)\left(U_2\left(X_T^{t, \epsilon}+E\right)-U_2\left(X_T^*+E\right)\right)\right] \\ =&\left(\frac{1}{\epsilon} E_t\left[\int_t^{t+\epsilon} \lambda_1(t, s) U_1^{\prime}\left(c_s^*\right) d s\right]\right) \kappa+\lambda_2(t, T) \frac{1}{\epsilon} E_t\left[U_2^{\prime}\left(X_T^*+E\right) \xi_T^{t, \epsilon}\right] \\ & -\frac{1}{ \epsilon} E_t\left[\int_t^{t+\epsilon} \lambda_1(t, s) \int_0^1\left|z U_1^{\prime \prime}\left(c_s^{t,\epsilon}- z \kappa\right)\right| d z d s|\kappa|^2+\lambda_2(t, T) \int_0^1\left| z U_2^{\prime \prime}\left(X_T^{\epsilon}+E-z\xi_T^{t, \epsilon}\right)\right| d z\left|\xi_T^{t, \epsilon}\right|^2\right]. \end{aligned} \end{split}} \end{equation} $ (3.4)

接下来, 讨论条件期望$ E_t\left[U^{'}_2(X^*_T+E)\xi^{t,\epsilon}_T\right] $的展开. 由假设$ ({\bf H1}) $可知$ U'_2(X^*_T+E)\in L^p_{\mathcal{F}_T}(\Omega;\mathbb{R}) $. 令$ (\alpha,\beta)\in L^p_{\mathcal{F}}(\Omega;C([0,T];\mathbb{R}))\times L^p_{\mathcal{F}}(\Omega;L^2(0,T;\mathbb{R}^2)) $, $ \beta=(\beta_1,\beta_2) $是下列倒向随机微分方程的唯一解

$ \begin{equation} \begin{split} \begin{cases} d\alpha_s=-r_s\alpha_s ds +\beta_{1s}dW_{1s}+\beta_{2s}dW_{2s} \\ \alpha_T=U^\prime (X^*_T+E). \end{cases} \end{split} \end{equation} $ (3.5)

可以解得$ \alpha_t=E_t\left[e^{-\int^T_t r_s ds }U^\prime _2(X^*_T)\right] $. 模仿Horst等[19]文中, 定义过程$ Y:=(U^\prime_2)^{-1}(\alpha)-X^* $, 则Y是连续适应过程, 使得$ U^{'}_2(X^*+Y)=\alpha\in L^p_{\mathcal{F}}(\Omega;C([0,T];\mathbb{R})) $, 且$ Y_T=E $. 由Itô公式可得

$\begin{equation} { \begin{split} d\left(X_s^*+Y_s\right) =&d\left(U_2^{\prime}\right)^{-1}\left(\alpha_s\right) \\ =&\frac{1}{U_2^{\prime \prime}\left(\left(U_2^{\prime}\right)^{-1}\left(\alpha_s\right)\right)} d \alpha_s-\frac{1}{2} \frac{U_2^{(3)}\left(\left(U_2^{\prime}\right)^{-1}\left(\alpha_s\right)\right)}{\left(U_2^{\prime \prime}\left(\left(U_2^{\prime}\right)^{-1}\left(\alpha_s\right)\right)\right)^3} d\langle\alpha, \alpha\rangle_s \\ =&\frac{1}{U_2^{\prime \prime}\left(X_s^*+Y_s\right)} \left(\beta_{1s}d W_{1s}+\beta_{2s}d W_{2s} \right) \\ & +\left(-\frac{1}{2} \frac{U_2^{(3)}\left(X_s^*+Y_s\right)}{\left(U_2^{\prime \prime}\left(X_s^*+Y_s\right)\right)^3}\left(\beta_{1s}^2+2\rho \beta_{1s}\beta_{2s}+\beta_{2s}^2\right)-r_s \frac{U_2^{\prime}}{U_2^{\prime \prime}}\left(X_s^*+Y_s\right)\right) d s, \end{split}} \end{equation} $ (3.6)

由此可得,

$\begin{equation} { \begin{split} d Y_s= & \left(\frac{1}{U_2^{\prime \prime}\left(X_s^*+Y_s\right)} \beta_{1s}-u_s^*\right) d W_{1s} +\left(\frac{1}{U_2^{\prime \prime}\left(X_s^*+Y_s\right)}\beta_{2s}\right)d W_{2s} \\ & +\bigg(-\frac{1}{2} \frac{U_2^{(3)}\left(X_s^*+Y_s\right)}{\left(U_2^{\prime \prime}\left(X_s^*+Y_s\right)\right)^3}\left(\beta_{1s}^2+2\rho \beta_{1s}\beta_{2s}+\beta_{2s}^2\right)-r_s \frac{U_2^{\prime}}{U_2^{\prime \prime}}\left(X_s^*+Y_s\right) -r_s X_s^*- \theta_s u_s^*-e_s+c_s^*\bigg) d s . \end{split}} \end{equation} $ (3.7)

定义$ Z_1:=\frac{1}{U^{\prime\prime}_2(X^*+Y)}\beta_1-u^* $, $ Z_2:=\frac{1}{U^{\prime\prime}_2(X^*+Y)}\beta_2 $. 由此可知, $ Z=(Z_1,Z_2) $是取值于$ \mathbb{R}^2 $的可料过程, 满足$ \int^T_0 |Z_s|^2 ds<\infty $ a.s. 故$ (Y,Z) $满足BSDE(2.20). 故$ (Y,Z) $满足定理3.1中的(1)和(2).最后我们证明等式(3.3)成立. 对$ (\alpha_s\xi^{t,\epsilon}_s)_{s\in[t+\epsilon,T]} $使用Itô引理

$\begin{equation} \begin{split} \alpha_T\xi^{t,\epsilon}_T=\alpha_{t+\epsilon}\xi^{t,\epsilon}_{t+\epsilon}+\int^T_{t+\epsilon}\xi^{t,\epsilon}_s\left(\beta_{1s} dW_{1s}+\beta_{2s} dW_{2s}\right). \end{split} \end{equation}$ (3.8)

由$ \xi^{t,\epsilon}\in\bigcap_{\gamma\geq1}L^\gamma_{\mathbb{F}}(\Omega;C([t,T];\mathbb{R})) $, $ (\beta_1,\beta_2)\in L^p_{\mathbb{F}}(\Omega;L^2(0,T;\mathbb{R}^2)) $, $ p>1 $, 以及Hôlder不等式可以得到$ E\Bigl[\Bigl(\int^T_t|\xi^{t,\epsilon}_s\beta_{1s}|^2\,ds\Bigr)^{1/2}\Bigr]<\infty $, 从而可知$ dW_{1s} $项是鞅. 同理, $ dW_{2s} $项也是鞅. 由此, 两边取条件期望可得

$\begin{equation} \begin{split} E_t[U'_2(X^*_T+E)\xi^{t,\epsilon}_T]&=E_t[\alpha_T\xi^{t,\epsilon}_T]=E_t\bigl[E_{t+\epsilon}[\alpha_T\xi^{t,\epsilon}_T]\bigr] \\ & =E_t[\alpha_{t+\epsilon}\xi^{t,\epsilon}_{t+\epsilon}]=E_t[U'_2(X^*_{t+\epsilon}+Y_{t+\epsilon})\xi^{t,\epsilon}_{t+\epsilon}]. \end{split} \end{equation}$ (3.9)

由Itô引理可得

$\begin{equation} { \begin{split} &U'_2(X^*_{t+\epsilon}+Y_{t+\epsilon})\xi^{t,\epsilon}_{t+\epsilon}\\ =&-\left(\int^{t+\epsilon}_tU^\prime_2(X^*_s+Y_s)ds\right)\kappa+\bigg(\int^{t+\epsilon}_t\bigl(U^\prime_2(X^*_s+Y_s)\theta_s+U''_2(X^*_s+Y_s)(u^*_s+Z_{1s}) +\rho U''_2(X^*_s+Y_s)Z_{2s}\bigr)\,ds\bigg)\eta\\ &+\int^{t+\epsilon}_tU^\prime_2(X^*_s+Y_s)\eta dW_{1s}+\int^{t+\epsilon}_t\xi^{t,\epsilon}_sU''_2(X^*_s+Y_s)(u^*_s+Z_{1s}) dW_{1s} +\int^{t+\epsilon}_t\xi^{t,\epsilon}_sU''_2(X^*_s+Y_s)Z_{2s} dW_{2s}. \end{split}} \end{equation} $ (3.10)

由

$\begin{equation} \begin{split} &U'_2(X^*+Y)=\alpha\in L^p_{\mathbb{F}}(\Omega;C([0,T];\mathbb{R})),\ U''_2(X^*+Y)(u^*+Z_1)=\beta_1\in L^p_{\mathbb{F}}(\Omega;L^2(0,T;\mathbb{R})),\\ & U''_2(X^*+Y)Z_2=\beta_2\in L^p_{\mathbb{F}}(\Omega;L^2(0,T;\mathbb{R})),\ \eta\in L^\infty_{\mathcal{F}_t}(\Omega;\mathbb{R}),\ \xi^{t,\epsilon}\in\bigcap\limits_{\gamma\geq1}L^\gamma_{\mathbb{F}}(\Omega;C([t,T];\mathbb{R})). \end{split} \end{equation} $ (3.11)

可得$ dW_1 $和$ dW_2 $项都是鞅. 故可得

$\begin{equation} { \begin{split} {E}_t[U'_2(X^*_T+E)\xi^{t,\epsilon}_T]=& -{E}_t\left[\int^{t+\epsilon}_tU^\prime_2(X^*_s+Y_s)ds\right]\kappa +{E}_t\bigg[\int^{t+\epsilon}_t\bigl(U^\prime_2(X^*_s+Y_s)\theta_s \\ & +U''_2(X^*_s+Y_s)(u^*_s+Z_{1s}) +\rho U''_2(X^*_s+Y_s)Z_{2s}\bigr)\,ds\bigg]\eta. \end{split}} \end{equation}$ (3.12)

最后, 由公式(3.4)以及公式(3.12)可得公式(3.3).

为了处理引理3.2右侧的极限问题, 本文引入文献[28]中的引理.

引理3.3  假设$ f(\cdot)\in L^2_{\mathbb{F}}(\Omega;L^2(s,t,\mathbb{R})) $, 对几乎处处$ t\in[0,T] $, 存在$ \{\epsilon_n\}\to 0,\,n\to \infty $, 使得

$\begin{equation} \lim\limits_{\epsilon_n\to 0}\frac{1}{\epsilon_n}\int^{t+\epsilon_n}_tE_t\left[f(s)-f(t)\right]=0,\quad a.s.. \end{equation}$ (3.13)

下面处理引理3.2右侧一次项的极限问题.

引理3.4  考虑引理3.2中的$ (Y(\cdot),Z(\cdot)) $, 存在可测集$ E_1\subset[0,T) $, $ \mathrm{Leb}([0,T]\setminus E_1)=0 $使得对任意$ t\in E_1 $, 存在序列$ \{\epsilon^t_n\}_{n\in\mathbb{N}}\subset(0,T-t) $满足$ \lim_{n\to\infty}\epsilon^t_n=0 $, 且

$\begin{equation} { \begin{split} \lim\limits_{n\to\infty}\frac{1}{\epsilon^t_n}E_t\left[\int^{t+\epsilon^t_n}_t\bigl(\lambda_1(t,s)U'_1(c^*_s)-\lambda_2(t,T)U'_2(X^*_s+Y_s)\bigr)\,ds\right] =U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t), \end{split}} \end{equation}$ (3.14)

以及

$\begin{equation} { \begin{split} &\lim\limits_{n\to\infty}\frac{1}{\epsilon^t_n}{E}_t\bigg[\int^{t+\epsilon}_t\bigl(U^\prime_2(X^*_s+Y_s)\theta_s +U''_2(X^*_s+Y_s)(u^*_s+Z_{1s}) +\rho U''_2(X^*_s+Y_s)Z_{2s}\bigr)\,ds\bigg]\\ =&U'_2(X^*_t+Y_t)\theta_t+U''_2(X^*_t+Y_t)(u^*_t+Z_{1t})+\rho U''_2(X^*_t+Y_t)Z_{2t}. \end{split}} \end{equation} $ (3.15)

证  由Young不等式, 存在$ p,q>1,\frac{1}{p}+\frac{1}{q}=1 $, 使得

$\begin{equation} { \begin{split} E_t\left[\int^T_t\left|U_2^\prime(X^*_s+Y_s)\theta(s)\right|^2ds\right] \leq& E_t\left[\int^T_t\left(\frac{U^\prime_2(X^*_s+Y_s)^p}{p}+\frac{\theta(s)^q}{q}\right)^2ds\right] \\ \leq &\left\{CE_t\left[\int^T_t |U_2^\prime(X^*_s+Y_s)\left|^{2p}\right|ds\right] +E_t\left[\int^T_t\left| \theta(s)\right|^{2q}ds\right]\right\}. \end{split}} \end{equation}$ (3.16)

由$ U^\prime_2(X^*_s+Y_s)\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}) $, 假设$ ({\bf H2}) $和引理3.3可知

$\begin{equation} { \begin{split} \lim\limits_{n\to\infty}\frac{1}{\epsilon^t_n}E_t\left[\int^{t+\epsilon^t_n}_t\bigl(U'_2(X^*_s+Y_s)\theta_s\bigr)\,ds\right] =U'_2(X^*_t+Y_t)\theta_t. \end{split}} \end{equation}$ (3.17)

应用Hamaguchi[18]中的方法即可得到引理3.4结论.

引理3.5  令$ E_1 $, $ \{\epsilon^t_n\}_{n\in\mathbb{N}} $是引理3.4中的集合与序列, 任意$ t\in E_1 $固定任意$ \delta>0 $, 存在可测集$ E_2\subset E_1 $, $ \mathrm{Leb}([0,T]\setminus E_2)=0 $, 使得对任意$ t\in E_2 $, $ |\kappa|\leq\delta $, 存在$ \{\epsilon^t_n\}_{n\in\mathbb{N}} $的子列(仍然记为$ \{\epsilon^t_n\}_{n\in\mathbb{N}} $)满足, a.s.

$ \begin{align*} &\limsup\limits_{n\to\infty} \frac{1}{ \epsilon} E_t\bigg[\int_t^{t+\epsilon} \lambda_1(t, s) \int_0^1\left|z U_1^{\prime \prime}\left(c_s^{t,\epsilon}- z \kappa\right)\right| d z d s|\kappa|^2 +\lambda_2(t, T) \int_0^1\left| z U_2^{\prime \prime}\left(X_T^{\epsilon}+E-z\xi_T^{t, \epsilon}\right)\right| d z\left|\xi_T^{t, \epsilon}\right|^2\bigg]\\ \leq & C\bigl(M_1(c^*_t;\delta)+E_t\bigl[|U''_2(X^*_T+E)|^q\bigr]^{1/q}\bigr)(|\kappa|^2+|\eta|^2). \end{align*} $

其中$ C>0 $是常数, $ q $是满足假设(H1)成立时的等价条件2.10中的参数.

证  下面模仿Hamaguchi[18]给出引理3.5的证明

$\begin{equation} { \begin{split} &\frac{1}{ \epsilon} E_t\left[\int_t^{t+\epsilon} \lambda_1(t, s) \int_0^1\left|z U_1^{\prime \prime}\left(c_s^{t,\epsilon}- z \kappa\right)\right| d z d s|\kappa|^2+\lambda_2(t, T) \int_0^1\left| z U_2^{\prime \prime}\left(X_T^{\epsilon}+E-z\xi_T^{t, \epsilon}\right)\right| d z\left|\xi_T^{t, \epsilon}\right|^2\right] \\ \leq & ||\lambda_1||_{\infty}\left(\frac{1}{\epsilon}E_t\left[\int^{t+\epsilon}_tM_1(c^*_s;\delta)ds\right]\right)|\kappa|^2 +||\lambda_2||_{\infty}\left[M_2(X^*_T+E;|\xi^{t,\epsilon}_T|)|\xi^{t,\epsilon}_T|^2\right]. \end{split}} \end{equation} $ (3.18)

由$ M_1(c^*;\delta)\in L^p_{\mathbb{F}}(\Omega;L^p(s,t,\mathbb{R}^n)) $和引理3.3, 可知存在$ E_2\subseteq E_1 $, 且$ Leb([0,T]\setminus E_2)=0 $, 对任意$ t\in E_2 $,

$\begin{equation} \frac{1}{\epsilon^t_n}E_t\left[\int^{t+\epsilon^t_n}_tM_1(c^*_s;\delta)ds\right]=M_1(c^*_t;\delta)\,\, a.s. \end{equation}$ (3.19)

固定$ t\in E_2 $, 记$ \xi_n:=|\xi_T^{t,\epsilon^t_n,\kappa,\eta}| $. 由引理2.1中结论(1)

$\begin{equation} E_T[\xi^{2\gamma}]\leq \left(C_\gamma\epsilon^t_n(|\kappa|^2+|\eta|^2)\right)^\gamma \,\, a.s. \end{equation}$ (3.20)

由Hölder不等式可以得到

$\begin{equation} \begin{split} \frac{1}{\epsilon^t_n}E_t\left[M_2(X_T^*+E;\xi_n)\xi_N^2\right]\leq& \frac{1}{\epsilon^t_n}E_t\left[M_2(X_T^*+E;\xi_n)^q\right]^{\frac{1}{q}}E_t\left[\xi_n^{\frac{2q}{q-1}}\right]^{\frac{q-1}{q}} \\ \leq& CE_t\left[M_2(X_T^*+E;\xi_n)^q\right]^{\frac{1}{q}}(|\kappa|^2+|\eta|^2). \end{split} \end{equation} $ (3.21)

选取$ \{\epsilon^t_n\}_{n\in \mathbb{R}} $的子列, 使得子列趋于$ 0 $. 由可知$ E[\xi_n]\to 0 $, 从而存在子列$ \xi_m\to 0 \,\, a.s. $. 选取对应的序列作为$ \{\epsilon^t_n\}_{n\in \mathbb{R}} $的子列, 仍然用原来记号, 在此子列下可知$ \lim_{n\to \infty}\xi_n=0\,\,a.s. $. 在指数效用函数下, 显然$ M_2(x,\delta) $是连续函数, 所以

$\begin{equation} \lim\limits_{n\to\infty}M_2(X^*_T+E;\xi_n)=M_2(X^*_T+E;0)=|U^{\prime\prime}(X^*_T+E)|\,\,a.s. \end{equation}$ (3.22)

由$ \{M_2(X^*_T+E;\xi_n)^q\}_{n\in \mathbb{R}} $是一致可积的, 可得

$\begin{equation} \lim\limits_{n\to \infty}E_t\left[M_2(X_T^*+E;\xi_n)^q\right]=E_t\left[|U_2^{''}(X_T^*+E)|^q\right]. \end{equation}$ (3.23)

定理3.1的证明  固定常数$ \delta>0 $, 在$ E_2 $中任意取$ t $, 取序列$ \{\delta_m\}_{\{m\in\mathbb{R}\}} $使得$ 0<\delta_m\leq\delta $, $ m\in\mathbb{N} $, $ \lim_{m\to\infty}\delta_m=0 $. 对任意$ m\in\mathbb{R} $, 定义

$\begin{equation} \begin{cases} \kappa_m=\delta_m{\bf 1}_{\{U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)>0\}}\\ \eta_m=0. \end{cases} \end{equation}$ (3.24)

记$ \{\epsilon^t_{n,m}\}_{n\in\mathbb{N}} $是对应$ t\in E_2 $的序列, 令$ c^{n,m}=c^{t,\epsilon^t_{n,m},\kappa_m} $, $ \pi^{n,m}=\pi^{t,\epsilon^t_{n,m},\eta_m} $, $ \,n,\,m\in\mathbb{N} $. 由开环控制解的定义及引理3.2, 3.4, 3.5可以得到, 对任意$ m\in\mathbb{N} $,

$\begin{equation} \begin{split} 0 \ge&\limsup\limits_{n\to\infty}\frac{R(c^{n,m},\pi^{n,m};t,X^*_t)-R(c^*,\pi^*;t,X^*_t)}{\epsilon^t_{n,m}} \\ \ge&\bigl(U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)\bigr){\bf 1}_{\{U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)>0\}}\delta_m \\ &-C\bigl(M_1(c^*_t;\delta)+E_t\bigl[|U''_2(X^*_T+E)|^q\bigr]^{1/q}\bigr)\delta^2_m \text{ a.s..} \end{split} \end{equation} $ (3.25)

不等式两边除以$ \delta_m $并且令$ m\to\infty $, 可以得到

$\begin{equation} \bigl(U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)\bigr){\bf 1}_{\{U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)>0\}}\leq0\ \text{ a.s.}, \end{equation}$ (3.26)

由此, 可得

$\begin{equation} U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)\leq0\,\, a.s.. \end{equation} $ (3.27)

定义$ (\kappa_m,\eta_m) $如下:

$\begin{equation} \begin{cases} \kappa_m=-\delta_m{\bf 1}_{\{U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)<0\}}\\ \eta_m=0, \end{cases} \end{equation} $ (3.28)

其中$ m\in\mathbb{N} $. 由同样的推导可知

$\begin{equation} \bigl(U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)\bigr){\bf 1}_{\{U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)<0\}}\ge0\,\,a.s. . \end{equation}$ (3.29)

由此, 可以得到$ U'_1(c^*_t)-\lambda_2(t,T)U'_2(X^*_t+Y_t)\ge 0\,\, a.s.. $从而得到$ c^* $的表达式. 同理, 取

$\begin{equation} \begin{cases} \kappa_m=0\\ \eta_m=\delta_m{\bf 1}_{\{U'_2(X^*_t+Y_t)\theta_t+U''_2(X^*_t+Y_t)(u^*_t+Z_{1t})+\rho U''_2(X^*_t+Y_t)Z_{2t}>0\}}, \end{cases} \end{equation} $ (3.30)

以及

$\begin{equation} \begin{cases} \kappa_m=0\\ \eta_m=\delta_m{\bf 1}_{\{U'_2(X^*_t+Y_t)\theta_t+U''_2(X^*_t+Y_t)(u^*_t+Z_{1t})+\rho U''_2(X^*_t+Y_t)Z_{2t}<0\}}. \end{cases} \end{equation} $ (3.31)

可得$ u^* $的表达式.

4 验证定理

上一节证明了存在开环均衡策略使得FBSDE存在适定解. 这一节验证若FBSDE存在解, 则开环均衡解存在.

定理4.1  固定$ x\in\mathbb{R} $, 假设存在$ (X(\cdot),Y(\cdot),Z(\cdot)) $, 其中$ Z(\cdot)=(Z_1(\cdot),Z_2(\cdot)) $满足

(1) $ X(\cdot) $和$ Y(\cdot) $是$ \mathbb{R} $取值的连续适应过程, $ Z(\cdot) $是取值于$ \mathbb{R}^2 $的可料过程满足$ \int^T_0 |Z_s|^2<\infty $ a.s.;

(2) $ U^{'}_2(X+Y)\in L^p_{\mathbb{F}}(\Omega;C([0,T];\mathbb{R})) $, 且

$\begin{equation} E\left[\int^T_0\bigl|U_1\bigl((U'_1)^{-1}(\lambda_2(s,T)U'_2(X_s+Y_s))\bigr)\bigr|\,ds+|U_2(X_T+E)|\right]<\infty; \end{equation} $ (4.1)

(3) $ (X(\cdot),Y(\cdot),Z(\cdot)) $满足

$\begin{equation} \begin{cases} X_t=x+\int^t_0\bigg(\frac{1}{\gamma_2}\theta_s-Z_{1s}-\rho Z_{2s}\bigg) dW_{1s}\\ \quad +\int^t_0\bigg(rX_s-\frac{\gamma_2}{\gamma_1}(X_s+Y_s)-\theta_s (Z_{1s}+\rho Z_{2s})+\frac{1}{\gamma_2}|\theta_s|^2+e_s+\frac{1}{\gamma_1}\ln(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T))\bigg)\,ds\\ Y_t=E-\int^T_tZ_{1s} dW_{1s}-\int^T_tZ_{2s} dW_{2s}\\ \quad -\int^T_t\bigg(-rX_s+\frac{\gamma_2}{\gamma_1}(X_s+Y_s)+\theta_s Z_{1s}+\rho \theta_s Z_{2s}+\frac{\gamma_2(1-\rho^2)}{2}|Z_{2s}|^2\\ \quad -\frac{1}{2\gamma_2}|\theta_s|^2-e_s-\frac{1}{\gamma_1}\ln(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T))+\frac{r}{\gamma_2}\bigg)\,ds,\ \ t\in[0,T]. \end{cases} \end{equation}$ (4.2)

定义$ (c^*,\pi^*) $如下

$\begin{equation} \begin{split} \begin{cases} c^*_s=(U'_1)^{-1}(\lambda_2(s,T)U'_2(X_s+Y_s)) =\frac{\gamma_2}{\gamma_1}(X_s+Y_s)-\frac{1}{\gamma_1}\ln \left(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T) \right)\\ u^*_s=-\frac{U'_2}{U''_2}(X_s+Y_s)\theta_s-Z_{1s}-\rho Z_{2s}=\frac{1}{\gamma_2}\theta_s-Z_{1s}-\rho Z_{2s} \end{cases} \ s\in[0,T]. \end{split} \end{equation}$ (4.3)

则$ (c^*,u^*) $是开环均衡解.

证  接下来证明$ (c^*,\pi^*) $是开环均衡解. 对任意$ t\in[0,T) $定义, $ \epsilon\in(0,T-t) $. 定义$ X^{t,\epsilon}:=X^{(c^{t,\epsilon},\pi^{t,\epsilon},t,X^*_t)} $, $ \xi^{t,\epsilon}:=X^{t,\epsilon}-X^* $, 则$ \xi^{t,\epsilon} $是(2.12) 中的SDE. 由$ U_1,U_2 $是凹函数, 故有

$\begin{equation} \begin{split} &\frac{R(c^{t,\epsilon},u^{t,\epsilon};t,X^*_t)-R(c^*,u^*;t,X^*_t)}{\epsilon} \\ \leq& \frac{1}{\epsilon} E_t\bigg[ \int^{t+\epsilon}_t\lambda(t,s)\gamma_1e^{-\gamma_1c^*}\kappa ds +\lambda_2(t,T)\gamma_2 e^{-\gamma_2X^*_T}\xi_T \bigg] \\ =&\frac{1}{\epsilon}{E}_t\left[\int^{t+\epsilon}_t\lambda_1(t,s)\lambda_2(s,T)\gamma_2e^{-\gamma_2(X_s+Y_s)}\kappa\,ds\,+\lambda_2(t,T)U'_2(X_T)\xi^{t,\epsilon}_T\right], \end{split} \end{equation}$ (4.4)

令$ \alpha:=U_2^\prime(X+Y) =\gamma_2e^{-\gamma_2(X+Y)} $, 由定理4.1中的假设(2)可知$ \alpha\in L^p_{\mathcal{F}}(\Omega;C([0,T];\mathbb{R})) $. 由Itô引理可知

$ \begin{equation} \begin{split} d\alpha_s =& U^{''}_2(X_s+Y_s)(dX_s+dY_s)+\frac{1}{2}U^{(3)}_2(X_s+Y_s)d\langle X+Y,X+Y \rangle_{s} \\ =& -r_sU_2^{'}(X_s+Y_s)ds +U^{''}(X_s+Y_s)(\frac{\theta_s}{\gamma_2}-\rho Z_{2s})dW_{1s}+U^{''}(X_s+Y_s)Z_{2s}dW_{2s}. \end{split} \end{equation}$ (4.5)

令$ \beta=(\beta_1,\beta_2) $, 其中$ \beta_1 =U^{''}(X_s+Y_s)(\frac{\theta_s}{\gamma_2}-\rho Z_{2s}) $, $ \beta_2=U^{''}(X_s+Y_s)Z_{2s} $, 则$ (\alpha,\beta_1,\beta_2) $, $ \int^T_0|\beta_s|^2ds<\infty $. $ (\alpha,\beta) $满足BSDE

$ \begin{equation} d\alpha_s=-r_s\alpha_s+\beta_{1s}dW_{1s}+\beta_{2s}dW_{2s}. \end{equation} $ (4.6)

与引理3.2中的证明一样们可以得到下式

$\begin{equation} \begin{split} E_t\left[U^{'}_2(X_T+Y_T)\xi_T\right] =E_t\left[\alpha_T\xi_T\right] =E_t\left[E_{t+\epsilon}\left[\alpha_T\xi_T\right]\right] =E_t\left[\alpha_{t+\epsilon}\xi_{t+\epsilon}\right]. \end{split} \end{equation} $ (4.7)

对$ (\alpha_s\xi_s^{t,\epsilon})_{s\in[t,t+\epsilon]} $, 可以得到

$\begin{equation} \begin{split} \alpha_{t+\epsilon}\xi_{t+\epsilon} =\alpha_{t}\xi_{t}+\left[\int^{t+\epsilon}_t\left(\alpha_s\theta_s+\beta_{1s}+\rho \beta_{2s}\right)ds\right]\eta -\left(\int^{t+\epsilon}_t \alpha_sds\right)\kappa \\ +\int^{t+\epsilon}_t\left(\alpha_s\eta+\xi_s\beta_{1s}\right) dW_{1s} +\int^{t+\epsilon}_t \xi_s\beta_{2s} dW_{2s}, \end{split} \end{equation} $ (4.8)

由$ \alpha_s\theta_s+\beta_{1s}+\rho \beta_{2s} =0 $以及$ dW_1 $项与$ dW_2 $项为鞅可知

$\begin{equation} \begin{split} E_t\left[U^{'}(X_T+Y_T)\xi_T\right] =&-E_t\left[\left(\int^{t+\epsilon}_t \alpha_s ds\right)\kappa \right]. \end{split} \end{equation} $ (4.9)

所以

$\begin{equation} \begin{split} \frac{R(c^{t,\epsilon},u^{t,\epsilon};t,X^*_t)-R(c^*,u^*;t,X^*_t)}{\epsilon} \leq & \frac{1}{\epsilon}E_t\left[\int^{t+\epsilon}_t\left(\lambda_1(t,s)\lambda_2(s,T)-\lambda_2(t,T)\right)\alpha_s \kappa\right] \\ \leq&\max\limits_{t\leq s\leq t+\epsilon}|\lambda_1(t,s)\lambda_2(s,T)-\lambda_2(t,T)|\ E_t\left[\sup\limits_{t\leq s\leq T}|\alpha_s|\right]\,|\kappa|\ \text{a.s.} \end{split} \end{equation}$ (4.10)

由于$ \lambda_1 $和$ \lambda_2 $是连续的, 注意$ \lambda_1(t,t)=1 $, 故上述第三个等式在$ \epsilon\downarrow0 $时趋于0. 由此, 可得$ \limsup_{\epsilon\downarrow0}\frac{R(c^{t,\epsilon},\pi^{t,\epsilon};t,X^*_t)-R(c^*,\pi^*;t,X^*_t)}{\epsilon}\leq0\ \text{a.s.}, $故$ (c^*,\pi^*) $是开环均衡解.

推论4.2  若假设($ {\bf H2} $)成立, 且进一步, 假设对任意C, 都有$ E_t\left[\exp\{ C\int^T_t\theta^2(s)ds \}\right]<+\infty $, 则FBSDE(4.2)存在唯一解$ (Y(\cdot),Z_1(\cdot),Z_2(\cdot))\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R})\times L^2_{\mathbb{F}}(\Omega;L^p(s,t,\mathbb{R}^2)) $, 且$ X(\cdot)\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}) $

证  变换$ (X,Y,Z) $如下

$\begin{equation} \tilde{X}_t:=h(t)X_t,\ \tilde{Y}_t:=Y_t+(1-h(t))X_t\ \text{and}\ \tilde{Z}_{1t}:=h(t)Z_{1t}+(1-h(t))\left(\frac{1}{\gamma_2}\theta_t-\rho Z_{2t}\right), \tilde{Z}_{2t}:=Z_{2t}. \end{equation} $ (4.11)

其中$ h(t) $是非负的确定性函数, 如下定义

$ \begin{equation*} h(t):= \begin{cases} \frac{1}{1+\frac{\gamma_2}{\gamma_1}(T-t)}\ \text{if}\ r=0,\\ \frac{r}{\frac{\gamma_2}{\gamma_1}-(\frac{\gamma_2}{\gamma_1}-r)\exp(-r(T-t))}\ \text{if}\ r>0. \end{cases} \end{equation*} $

注意$ h $微分$ \dot{h}(t)=h(t)\left(\frac{\gamma_2}{\gamma_1}h(t)-r\right),\ t\in[0,T],\ h(T)=1. $通过Itô引理, 可知FBSDE(4.2)可变换如下

$\begin{equation} \begin{cases} \tilde{X}_t=h(0)x+\int^t_0\bigl(\frac{1}{\gamma_2}\theta_s-\tilde{Z}_{1s}-\rho \tilde{Z}_{2t}\bigr) dW_{1s} \\ \quad +\int^t_0\bigl(-\frac{\gamma_2}{\gamma_1}h(s)\tilde{Y}_s-\theta_s \tilde{Z}_{1s}-\rho \theta_s \tilde{Z}_{2s}+\frac{1}{\gamma_2}|\theta_s|^2+h(s)e_s+\frac{h(s)}{\gamma_1}\ln(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T))\bigr)\,ds\\ \tilde{Y}_t=E-\int^T_t\tilde{Z}_{1s} dW_{1s}-\int^T_t\tilde{Z}_{2s} dW_{2s}\\ \quad -\int^T_t\bigl(\frac{\gamma_2}{\gamma_1}h(s)\tilde{Y}_s+\theta_s \tilde{Z}_{1s}+\rho \theta \tilde{Z}_{2 s}+\frac{\gamma_2 (1-\rho^2)}{2}|\tilde{Z}_{2s}|^2\\ \quad -\frac{1}{2\gamma_2}|\theta_s|^2-h(s)e_s-\frac{h(s)}{\gamma_1}\ln(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T))+\frac{r}{\gamma_2}\bigr)\,ds, \quad t\in[0,T]. \end{cases} \end{equation}$ (4.12)

对$ W_{2t} $作如下分解, 其中$ W^\perp_{t} $与$ W_{1t} $独立.

$\begin{equation} dW_{2t}=\rho dW_{1t}+\sqrt{1-\rho^2}dW_{t}^\perp. \end{equation}$ (4.13)

从而在式(2.15)定义的测度Q下有

$\begin{equation} \begin{split} \tilde{Y}_t=&E-\int^T_t(\tilde{Z}_{1s} +\rho \tilde{Z}_{2s}) d\tilde{W}_{1s}-\int^T_t\sqrt{1-\rho^2}\tilde{Z}_{2s} dW_{s}^\perp \\ &-\int^T_t\bigl(\frac{\gamma_2}{\gamma_1}h(s)\tilde{Y}_s+\frac{\gamma_2 (1-\rho^2)}{2}|\tilde{Z}_{2s}|^2 -\frac{1}{2\gamma_2}|\theta_s|^2-h(s)e_s-\frac{h(s)}{\gamma_1}\ln(\frac{\gamma_2}{\gamma_1}\lambda_2(s,T))+\frac{r}{\gamma_2}\bigr)\,ds. \end{split} \end{equation} $ (4.14)

定理中进一步加强的假设条件可知, $ (4.14) $满足文献[27]中的假设条件, 故该FBSDE存在唯一解.

$ \begin{equation} |\tilde{Y_t}|\leq \gamma_2(1-\rho^2)\log E_t^Q\left(K_1+\gamma_2(1-\rho^2)\int^T_0e^{K_2T}\theta^2(s)ds\right)<+\infty, \end{equation}$ (4.15)

其中$ K_1 $, $ K_2 $是有界常数. 故$ \tilde{Y_t}\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}) $.

$\begin{equation} { \begin{split} E\left[\left(\int^T_0|\tilde{Z}_{2s}|^2ds\right)^{\frac{p}{2}}\right]=&E^Q\left[\left(\int^T_0|\tilde{Z}_{2s}|^2ds\right)^{\frac{p}{2}}\Lambda^{-1}(T)\right] \leq \sqrt{E^Q\left[\left(\int^T_0|\tilde{Z}_{2s}|^2ds\right)^p\right]} \sqrt{E^Q\left[\Lambda^{-2}(T)\right]} \\ <&+\infty. \end{split}} \end{equation}$ (4.16)

同理可得$ E\left[\left(\int^T_0|\tilde{Z}_{2s}|^2ds\right)^{\frac{p}{2}}\right]<+\infty $, 所以$ (\tilde{Z}_{1}(\cdot),Z_{2}(\cdot))\in L^2_{\mathbb{F}}(\Omega;L^p(s,t,\mathbb{R}^2)) $.

$\begin{equation} { \begin{split} E\left[\sup\limits_{s\in[0,T]}|\tilde {X}_s|^p\right] \leq& K_1+K_2E\left[\sup\limits_{s\in[0,T]}\left(\int^s_0\theta_s^2+\tilde{Z}^2_{1s}+ \tilde{Z}^2_{2s} ds\right)^{\frac{p}{2} }\right] +K_3E\left[\left(\int^T_0\tilde{Y}_s+\theta^2_s+\tilde{Z}^2_{1s}+\tilde{Z}^2_{2s}\right)^p \right] \\ <& +\infty. \end{split}} \end{equation} $ (4.17)

由于$ (X(\cdot),Y(\cdot),Z_{1}(\cdot),Z_{2}(\cdot)) $和$ (\tilde{X}(\cdot),\tilde{Y}(\cdot),\tilde{Z}_{1}(\cdot),\tilde{Z}_{2}(\cdot)) $存在一一对应关系, 所以$ Y(\cdot),Z_1(\cdot),Z_2(\cdot))\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R})\times L^2_{\mathbb{F}}(\Omega;L^p(s,t,\mathbb{R}^2)) $, 且$ X(\cdot)\in L^p_{\mathbb{F}}(\Omega;C[s,t];\mathbb{R}) $.

5 总结

本文将文献的结论推广到了随机波动率模型的情景. 通过引入风险溢价过程指数阶有界的假设, 本文可以处理随机参数无界的情形. 借助FBSDE, 本文给出了动态投资-消费问题的开环均衡解存在的必要条件. 随后, 本文验证了该策略为开环均衡策略. 由于效用函数由指数函数建模, 由指数阶有界的假设, 可以得到FBSDE解的存在唯一性.

参考文献
[1] Ramsey F P. A mathematical theory of saving[J]. The Economic Journal, 1928, 38(152): 543–559. DOI:10.2307/2224098
[2] Samuelson P A. A note on measurement of utility[J]. The Review of Economic Studies, 1937, 4(2): 155–161. DOI:10.2307/2967612
[3] Ainslie G. Picoeconomics: The strategic interaction of successive motivational states within the person[M]. Cambridge Univ. Press, 1992.
[4] Loewenstein G., Prelec D. Anomalies in intertemporal choice: Evidence and an interpretation[J]. The Quarterly Journal of Economics, 1992, 107(2): 573–597. DOI:10.2307/2118482
[5] Strotz R H. Myopia and inconsistency in dynamic utility maximization[M]. Springer, 1973.
[6] Ekeland I, Pirvu T A. Investment and consumption without commitment[J]. Mathematics and Financial Economics, 2008, 2(1): 57–86. DOI:10.1007/s11579-008-0014-6
[7] Merton R C. Lifetime portfolio selection under uncertainty: The continuous-time case[J]. The review of Economics and Statistics, 1969, 51(3): 247–257. DOI:10.2307/1926560
[8] Merton R. Optimum consumption and portfolio rules in a continuous-time model[J]. Journal of Economic Theory, 1971, 3(4): 373–413. DOI:10.1016/0022-0531(71)90038-X
[9] Björk T, Khapko M, Murgoci A. On time-inconsistent stochastic control in continuous time[J]. Finance and Stochastics, 2017, 21(2): 331–360. DOI:10.1007/s00780-017-0327-5
[10] Hu Y, Jin H, Zhou X Y. Time-inconsistent stochastic linear-quadratic control[J]. SIAM journal on Control and Optimization, 2012, 50(3): 1548–1572. DOI:10.1137/110853960
[11] Hu Y, Jin H, Zhou X Y. Time-inconsistent stochastic linear-quadratic control: Characterization and uniqueness of equilibrium[J]. SIAM Journal on Control and Optimization, 2017, 55(2): 1261–1279. DOI:10.1137/15M1019040
[12] Huang J, Li X, Wang T. Characterizations of closed-loop equilibrium solutions for dynamic mean- variance optimization problems[J]. Systems & Control Letters, 2017, 110(1): 15–20.
[13] Zhang J, Chen P, Jin Z, et al. Open-loop equilibrium strategy for mean-variance portfolio selection: A log-return model[J]. Journal of Industrial & Management Optimization, 2021, 17(2): 765–777.
[14] Hu Y, Imkeller P, Müller M. Utility maximization in incomplete markets[J]. Annals of Applied Probability, 2005, 15(3): 1691–1712.
[15] Cheridito P, Hu Y. Optimal consumption and investment in incomplete markets with general constraints[J]. Stochastics and Dynamics, 2011, 11(2-3): 283–299.
[16] Zhao Q, Shen Y, Wei J. Consumption-investment strategies with non-exponential discounting and logarithmic utility[J]. European Journal of Operational Research, 2014, 238(3): 824–835. DOI:10.1016/j.ejor.2014.04.034
[17] Yong J. Time-inconsistentoptimal control problems and the equilibrium hjb equation[J]. Mathematical Control and Related Fields, 2012, 2(3): 271–329. DOI:10.3934/mcrf.2012.2.271
[18] Hamaguchi Y. Time-inconsistent consumption-investment problems in incomplete markets under general discount functions[J]. Siam Journal on Control and Optimization, 2021, 59(3): 2121–2146. DOI:10.1137/19M1303782
[19] Horst U, Hu Y, Imkeller P, et al. Forward-backward systems for expected utility maximization[J]. Stochastic Processes and their Applications, 2014, 124(5): 1813–1848. DOI:10.1016/j.spa.2014.01.004
[20] Zariphopoulou T. A solution approach to valuation with unhedgeable risks[J]. Finance and Stochastics, 2001, 5(1): 61–82. DOI:10.1007/PL00000040
[21] Kraft* H. Optimal portfolios and heston's stochastic volatility model: an explicit solution for power utility[J]. Quantitative Finance, 2005, 5(3): 303-313.
[22] Han B, Wong H Y. Mean-variance portfolio selection under volterra heston model[J]. Applied Mathematics & Optimization, 2021, 84(1): 683–710.
[23] Shen Y. Mean-variance portfolio selection in a complete market with unbounded random cofficients[J]. Automatica, 2015, 55(c): 165–175.
[24] Yan T, Wong H Y. Open-loop equilibrium strategy for mean-variance portfolio problem under stochastic volatility[J]. Automatica, 2019, 107(1): 211–223.
[25] He X D, Jiang Z. Mean-variance portfolio selection with dynamic targets for expected terminal wealth[J]. Mathematics of Operations Research, 2022, 47(1): 587–615. DOI:10.1287/moor.2021.1142
[26] Su X, Wang T, Wei J, et al. Non-markovian mean-variance portfolio selection problems via closedloop equilibrium strategies[J]. Applied Mathematics & Optimization, 2024, 89(1): 15.
[27] Briand Philippe, Hu Ying. Quadratic BSDEs with convex generators and unbounded terminal conditions[J]. Probability Theory and Related Fields, 2008, 141(3): 543–567.
[28] Wang Tianxiao. Necessary conditions in stochastic linear quadratic problems and their applications[J]. Journal of Mathematical Analysis and Application, 2019, 469(1): 280–297. DOI:10.1016/j.jmaa.2018.09.013