数学物理学报, 2024, 44(6): 1617-1629

通货膨胀下的$\alpha$-鲁棒最优投资策略研究

陈雅婷,1, 刘海燕,1, 陈密,1,2,3,*

1福建师范大学数学与统计学院 福州 350117

2分析数学及应用教育部重点实验室 福州 350117

3统计学与人工智能福建省高校重点实验室 福州 350117

$\alpha$-Robust Optimal Investment Strategy Under Inflation

Chen Yating,1, Liu Haiyan,1, Chen Mi,1,2,3,*

1School of Mathematics and Statistics, Fujian Normal University, Fuzhou 350117

2Key Laboratory of Analytical Mathematics and Applications $($Ministry of Education$)$, Fuzhou 350117

3Fujian Provincial Key Laboratory of Statistics and Artificial Intelligence, Fuzhou 350117

通讯作者: *陈密, Email: chenmi0610@163.com

收稿日期: 2024-01-19   修回日期: 2024-05-6  

基金资助: 国家自然科学基金(11701087)
福建省自然科学基金(2023J01537)
福建省自然科学基金(2023J01538)

Received: 2024-01-19   Revised: 2024-05-6  

Fund supported: NSFC(11701087)
NSF of Fujian Province(2023J01537)
NSF of Fujian Province(2023J01538)

作者简介 About authors

陈雅婷,Email:895707632@qq.com;

刘海燕,Email:rain6397@163.com

摘要

该文主要研究通货膨胀下具有模型不确定性的最优投资问题. 假定金融市场上存在无风险资产、风险资产以及用于对冲通胀风险的通货膨胀指数债券可投资, 其中风险资产价格服从 CEV 模型, 而后利用价格指数水平折现各类资产价格呈现其真实价格, 运用 $\alpha$-maxmin 均值方差效用函数建立投资模型, 并通过求解 HJB 方程获得均衡投资策略与值函数的显式解. 最后结合数值仿真分析了参数变动下的最优投资策略变化趋势.

关键词: 通货膨胀; 指数债券; 通胀折现; $\alpha$-鲁棒均值方差准则; CEV 模型

Abstract

This paper focuses on the optimal investment problem with model uncertainty under inflation. It is assumed that there are risk-free assets, risky assets and inflation-indexed bonds used to hedge inflation risk in the financial market, in which the price of risky assets obeys the CEV model, and then the price index level is used to discount the price of each type of asset to present its true price, and the investment model is built by applying the $\alpha$-maxmin mean-variance utility function and the equilibrium investment strategy and value function are obtained by solving the HJB equation. Finally, the trend of optimal investment strategies under parameter variations is analyzed by numerical simulation.

Keywords: Inflation; Index bonds; Inflation discounting; $\alpha$-robust mean-variance criterion; CEV model

PDF (659KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

陈雅婷, 刘海燕, 陈密. 通货膨胀下的$\alpha$-鲁棒最优投资策略研究[J]. 数学物理学报, 2024, 44(6): 1617-1629

Chen Yating, Liu Haiyan, Chen Mi. $\alpha$-Robust Optimal Investment Strategy Under Inflation[J]. Acta Mathematica Scientia, 2024, 44(6): 1617-1629

1 引言

通货膨胀是指在纸币流通条件下, 因货币供给大于货币实际需求, 导致货币贬值, 而引起的一段时间内物价持续普遍上涨的现象. 该现象会直接影响经济的各个方面, 包括支出、投资和利率等. 近年来, 我国对于通货膨胀的相关研究有了一定深度, 为通货膨胀在金融保险领域的研究发展奠定了坚实基础.

在所有这些工作中, 如何将通货膨胀的影响纳入金融保险的相关模型也逐渐引起了众多学者的研究. Zhang 等[1]引入了与通货膨胀呈相同波动性的通货膨胀连接型债券对冲通胀风险, 用于研究通货膨胀下货币市场账户、股票与通胀挂钩债券中持有 DC 型养老金的最佳投资管理方式. Siu[2]考虑了金融市场含有固定利率债券、股票以及通货膨胀连接型债券, 并利用价格指数对三类资产进行折现, 研究了通胀情况下连续时间的最优投资消费问题. Kwak 和 Lim[3]同样使用价格水平指数对各类资产的价格和投资者的财富进行折现, 从而给出真实情况下人寿保险的最佳投资组合. 本文在对通胀下资产价格的处理上运用上述文献的设定方式, 将资产价格变换到真实情况下, 进而求解最优投资问题.

由于现实中无法精准估计参考模型参数, 很多文献将模型不确定性考虑在投资再保险模型中, 但这类鲁棒模型存在只考虑极端模糊厌恶态度的局限性, 与真实情况不符. 在后续的研究中, 一些学者研究了 $\alpha$-maxmin 期望效用[4,5]. 特别是 Li 等[6]在此基础上, 提出了区别于一般均值方差准则的 $\alpha$-鲁棒均值方差准则, 能考虑更加多样的模糊态度, 比传统鲁棒模型考虑范围更广也更接近现实情况. 而后续对 $\alpha$-鲁棒均值方差准则的研究相对较少, 其中包含: Yu 等[7]研究了在 $\alpha$-maxmin 均值方差准则下具有模型参数不确定性的投资组合选择问题; Zhang 和 Li[8]研究了 $\alpha$-maxmin 均值-方差准则下的最优再保险问题; Li 等[9]考虑了在均值方差框架中具有跳-扩散风险不确定性的 DC 养老金计划的 $\alpha$-鲁棒最优投资问题; Chen 等[10]运用随机延迟微分方程描述财富过程, 考虑了 $\alpha$-maxmin 均值方差准则的最优投资再保险策略; Guan 等[11]研究了随机波动率模型下保险公司和再保险公司在 $\alpha$-maxmin 均值方差准则中的 Stackelberg 博弈.

据我们所知, 在以往的研究中, 还未有研究利用 $\alpha$-maxmin 均值方差准则来分析通货膨胀下风险资产价格服从 CEV 模型的最优投资问题. 因此, 本文综合考虑通货膨胀以及多样模糊态度对投资人投资过程的影响, 研究在此背景下的最优投资策略. 本文与涉及 $\alpha$-鲁棒的相关论文之间存在两个主要差异: 第一是本文设置的金融市场除了考虑无风险资产和风险资产, 还考虑了通货膨胀指数债券, 且风险资产价格服从 CEV 模型. 第二是本文将各类资产进行通胀折现, 从而研究了通货膨胀下的 $\alpha$-鲁棒均值方差投资策略, 与一般研究关注的投资模型不同.

2 基本模型

2.1 金融市场

假设在连续时间的金融市场中有三种金融资产, 分别为: 无风险资产、风险资产以及通货膨胀指数债券.

假定无风险资产在 $t$ 时刻的价格为 $S_{0}(t)$, 且满足以下常微分方程

$\begin{equation} {\rm d}S_{0}(t)=\mu_{0}S_{0}(t){\rm d}t, \end{equation}$

其中 $\mu_{0}>0$ 为常数, 表示无风险资产的收益率.

风险资产在 $t$ 时刻的价格为 $S_{1}(t)$, 且价格过程满足常方差弹性模型 (CEV)

$\begin{equation} {\rm d}S_{1}(t)=\mu_1S_{1}(t){\rm d}t+\sigma_{1}S_{1}(t)^{\beta+1}{\rm d}W_{1}(t),\end{equation}$

其中, $\mu_{1}>\mu_{0}$ 为常数, 表示风险资产的平均收益率, $\sigma_{1}>0$ 为常数, 表示风险资产的波动率, $\sigma_{1}S_{1}(t)^{\beta+1}$ 为一个瞬时波动率, $\beta$ 表示弹性系数, $W_{1}(t)$ 为概率空间$(\mathbf{\Omega},\mathcal{F},\mathbf{P})$ 上的标准布朗运动. 若弹性系数 $\beta=0$, 则 CEV 模型退化为 GBM 模型.

在很多情况下, 通货膨胀的典型指标可以使用 CPI (消费价格指数) 来表示$^{\left[1-3\right]}$, 而消费价格指数可以视为一个价格水平过程, 它可以体现投资者所在地区的通货膨胀水平. 这里假设价格水平 $P(t)$ 的变化规律为

$\begin{equation} \frac{{\rm d}P(t)}{P(t)}=\mu_{L}{\rm d}t+\sigma_{L}{\rm d}W_{L}(t),P(0)=1,\end{equation}$

其中 $\mu_{L}$ 为常数, 可以作为传统经济学的通胀率, 近似于真实利率与名义利率的差值; $\sigma_{L}$ 为常数, 表示的是通货膨胀的波动率; $W_{L}(t)$ 为另一个标准布朗运动, 且该布朗运动与 $W_{1}(t)$ 相互独立.

为了对冲通货膨胀风险, 本文假设市场上存在一种通货膨胀指数债券, 该债券主要是投资者用于对冲通货膨胀风险的投资产品, 其在 $t$ 时刻的价格过程用 $S_2(t)$ 表示, 且服从以下方程

$\begin{equation}\frac{{\rm d}S_2(t)}{S_2(t)}=(\mu_{0}-\mu_{L}){\rm d}t+\frac{{\rm d}P(t)}{P(t)}=\mu_{0}{\rm d}t+\sigma_{L}{\rm d}W_{L}(t),\end{equation}$

其中 $\mu_{0}$--$\mu_{L}$ 可以理解为真实利率.

投资者可以通过投资上文描述的无风险资产、风险资产以及用于对冲通胀风险的通货膨胀指数债券来实现财富的积累与增值, 以获得更高的财富效用.

2.2 金融市场的通胀调整

由于通货膨胀会对各类资产价格产生一定影响, 本文需先对各类资产价格模型运用价格指数进行调整, 得到其经过调整后不受通货膨胀影响的模型[12].

经过调整后的无风险资产价格为: $\tilde{S}_{0}(t)=\frac{S_{0}(t)}{P(t)}$, 其服从

$\begin{equation}\frac{{\rm d}\tilde{S}_{0}(t)}{\tilde{S}_{0}(t)}=(\mu_{0}-\mu_{L}+\sigma_L^2){\rm d}t-\sigma_{L}{\rm d}W_{L}(t),\end{equation}$

其中 $\mu_{0}-\mu_{L}+\sigma_L^2$ 为通过通货膨胀调整后的无风险资产的收益率.

经过调整的风险资产价格为: $\tilde{S}_{1}(t)=\frac{S_{1}(t)}{P(t)}$, 其服从

$\begin{equation} \frac{{\rm d}\tilde{S}_{1}(t)}{\tilde{S}_{1}(t)}=(\mu_{1}-\mu_{L}+\sigma_L^2){\rm d}t+\sigma_{1}S_1^\beta {\rm d}W_{1}(t)-\sigma_{L}{\rm d}W_{L}(t),\end{equation}$

其中 $\mu_{1}-\mu_{L}+\sigma_L^2$ 表示通过通货膨胀调整后的风险资产收益率.

经过调整的通货膨胀指数债券价格为: $\tilde{S_2}(t)=\frac{S_2(t)}{P(t)}$, 其服从

$\begin{equation}\frac{{\rm d}\tilde{S_2}(t)}{\tilde{S_2}(t)}=(\mu_{0}-\mu_{L}){\rm d}t.\end{equation}$

可以观察到经过通胀折现调整后的三类资产中, 通货膨胀指数债券价格形式变为寻常的无风险资产价格形式, 即不受波动率影响; 而无风险资产与风险资产价格都变为带漂移的几何布朗运动形式.

此外, 假设在 $t$ 时刻有 $\pi (t)$ 的财富用于投资风险资产, $\pi _{B}(t)$ 财富用于投资无风险资产, 剩余的 $X(t)-\pi (t)-\pi _{B}(t)$ 用于投资通货膨胀指数债券, 且允许投资中存在卖空行为, 即 $\pi (t)\in \mathbb{R},\pi _{B}(t)\in \mathbb{R}$. 因此, 投资者经过通货膨胀系数调整后财富过程的动态模型如下

$\begin{equation}\begin{aligned} {\rm d}X(t)&=\pi (t)\frac{{\rm d}\tilde{S}_{1}(t)}{\tilde{S}_{1}(t)}+\pi _{B}(t)\frac{{\rm d}\tilde{S}_{0}(t)}{\tilde{S}_{0}(t)}+\big(X(t)-\pi (t)-\pi _{B}(t)\big)\frac{{\rm d}\tilde{S_2}(t)}{\tilde{S_2}(t)}\\&=\Big[\pi (t)\big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big)+\pi _{B}(t)\sigma_{L}^2+X(t)\big(\mu_{0}-\mu_{L}\big)\Big]{\rm d}t\\& +\pi (t)\sigma_{1}S_{1}^{\beta}{\rm d}W_{1}(t)-\sigma_{L}\big(\pi (t)+\pi _{B}(t)\big){\rm d}W_{L}(t), \end{aligned}\end{equation}$

其初值为 $X(0)=x_0.$

3 基于 $\alpha$-鲁棒的投资模型

上述模型框架是建立在确定的概率测度 $\mathbf{P}$ 下进行的, 但鉴于现实环境中, 概率测度 $\mathbf{P}$ 下的参考模型参数难以精准估计, 投资者无法确定参考模型是否为真实模型. 这里设置一组可替代测度 $\mathbf{Q}$, 且有 $\mathbf{Q}\sim\mathbf{P}$, 即可替代测度 $\mathbf{Q}$ 与参考概率测度 $\mathbf{P}$ 总是等价的, 则可用概率测度集 $Q=\{\mathbf{Q}|\mathbf{Q}\sim\mathbf{P}\}$ 来刻画一系列备选模型. 而后为了引入模糊性的概念, 这里给出以下定义

定义 3.1 过程 $\left\lbrace\phi(t)=(\phi_1(t),\phi_2(t))\mid t\in[T]\right\rbrace$, 满足对任意的 $t\in[T]$

(1) $\phi_1(t),\phi_2(t)$$\mathcal{F}_t$ 可测的$;$

(2) $\phi_1(t),\phi_2(t)$ 满足$E\left[\exp\big\{\frac{1}{2}\int_{t}^{T}(\phi_1^2(s)+\phi_2^2(s)){\rm d}s\big\}\right]<\infty$, 称该条件为 Novikov 条件, 该条件保证后面的 $\Lambda^{\phi}(t)$ 是参考测度 $\mathbf{P}$ 下的鞅.

$\Phi$ 表示所有随机过程 $\{\phi(t)\}$ 构成的空间, 并且对任意的概率测度 $\mathbf{Q}\in Q$, 都存在过程 $\phi(t)\in \Phi$, 使得 Radon-Nikodym 导数过程 $\frac{{\rm d}\mathbf{Q}}{{\rm d}\mathbf{P}}\mid_{\mathcal{F}_t}:=\Lambda^{\phi}(t)$ 满足

$\begin{equation} \Lambda^{\phi}(t)=\exp\left\lbrace -\int_{0}^{t}\phi_1(s){\rm d}W_1(s)-\frac{1}{2}\int_{0}^{t}\phi_{1}^{2}(s){\rm d}s-\int_{0}^{t}\phi_2(s){\rm d}W_L(s)-\frac{1}{2}\int_{0}^{t}\phi_{2}^{2}(s){\rm d}s\right\rbrace. \end{equation}$

根据 Girsanov 定理, 有

$\begin{equation}\begin{aligned} {\rm d}W_1^{\phi}(t)&={\rm d}W_1(t)+\phi_1(t){\rm d}t,\\ {\rm d}W_L^{\phi}(t)&={\rm d}W_L(t)+\phi_2(t){\rm d}t, \end{aligned}\end{equation}$

其中 $W_1^{\phi}(t), W_L^{\phi}(t)$$\mathbf{Q}^{\phi}$ 测度下的布朗运动.将 (3.2) 式带入 (2.8) 式则可得到在替代测度 $\mathbf{Q}^{\phi}$ 下投资者的财富过程

$\begin{equation}\begin{aligned} {\rm d}X^{\phi}(t)&=\Big[\pi (t)\big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big)+\pi _{B}(t)\sigma_{L}^2+X(t)\big(\mu_{0}-\mu_{L}\big)-\pi (t)\sigma_{1}S_{1}^{\beta}\phi_1(t)\\ & +\sigma_{L}\big(\pi (t)+\pi _{B}(t)\big)\phi_2(t)\Big]{\rm d}t+\pi (t)\sigma_{1}S_{1}^{\beta}{\rm d}W_1^{\phi}(t)-\sigma_{L}\big(\pi (t)+\pi _{B}(t)\big){\rm d}W_L^{\phi}(t), \end{aligned}\end{equation}$

其初值为 $X^{\phi}(0)=x$.

接下来根据 $\alpha$-maxmin 均值方差效用准则[6], 给出投资模型对应的 $\alpha$-鲁棒均值方差回报函数

$\begin{equation}\begin{aligned} J_{\alpha}^{u}(t,x,s_1):&=\alpha\inf\limits_{\phi\in\Phi}\underline{J}^{u,\phi}(t,x,s_1)+\hat{\alpha}\sup\limits_{\phi\in\Phi}\overline{J}^{u,\phi}(t,x,s_1)\\&=\alpha\underline{J}^{u,\underline{\phi}}(t,x,s_1)+\hat{\alpha}\overline{J}^{u,\overline{\phi}}(t,x,s_1), \end{aligned}\end{equation}$

其中 $\alpha\in[0,1],\hat{\alpha}=1-\alpha$,

$\begin{equation}\begin{aligned} \underline{J}^{u,\phi}(t,x,s_1)=E_{t,x,s_1}^{\phi}[X^{u}(T)]-\frac{\gamma}{2}{\rm Var}_{t,x,s_1}^{\phi}[X^{u}(T)]+\int_{t}^{T}h^{\psi}(\phi(s)){\rm d}s, \end{aligned}\end{equation}$
$\begin{equation}\begin{aligned} \overline{J}^{u,\phi}(t,x,s_1)=E_{t,x,s_1}^{\phi}[X^{u}(T)]-\frac{\gamma}{2}{\rm Var}_{t,x,s_1}^{\phi}[X^{u}(T)]-\int_{t}^{T}h^{\psi}(\phi(s)){\rm d}s, \end{aligned}\end{equation}$

其中 $\gamma>0$ 为投资者的风险厌恶系数, $E_{t,x,s_1}\left[\cdot\right]$${\rm Var}_{t,x,s_1}\left[\cdot\right]$ 分别为给定 $\{X(t)=x,S_1=s_1\}$ 下的条件期望与条件方差. $h^{\psi}$ 为惩罚函数

$\begin{equation}\begin{aligned} h^{\psi}(\phi(t))=\frac{\phi_{1}^{2}(t)}{2\psi_1(t)}+\frac{\phi_{2}^{2}(t)}{2\psi_2(t)}, \end{aligned}\end{equation}$

其中 $\psi_1(t),\psi_2(t)$ 分别表示股票价格过程的模糊水平以及通货膨胀波动率过程的模糊水平. 这里假设: $\psi_1(t)=\psi_1,\psi_2(t)=\psi_2$, 其中 $\psi_1,\psi_2$ 为非负常数. 当 $\psi_1(t),\psi_2(t)$ 数值越大, 则惩罚函数值越小, 表示模糊厌恶型投资者的模糊厌恶水平越高. 特别地, 当 $\psi_i\downarrow 0(i=1,2)$ 时, 而且有 $h^{\psi}\uparrow\infty$, 则 $\inf\limits_{\phi\in\Phi}\underline{J}^{u,\phi}(t,x,s_1)$$\sup\limits_{\phi\in\Phi}\overline{J}^{u,\phi}(t,x,s_1)$ 都能够在 $\phi(t)=(0,0)$ 取得, 这时的 (3.4) 式就会退化为经典的均值方差回报函数. 反之, 当 $\psi_i\uparrow \infty(i=1,2)$ 时, 对应的 $h^{\psi}\downarrow0$, 则表示投资者完全不相信参考测度 $\mathbf{P}$. 此外, $\alpha$ 值越高, 则表示模糊厌恶水平越高. 特别地, $\alpha=0$ 表示投资者为模糊偏好型, $\alpha=\frac{1}{2}$ 表示投资者为模糊中立型, $\alpha=1$ 表示投资者为模糊厌恶型.

本文的主要目标是研究时间一致的均值方差策略下的 $\alpha$-鲁棒最优投资问题, 即

$\begin{equation}\begin{aligned} V(t,x,s_1):=\sup\limits_{u\in \mathcal {U}}J_{\alpha}^{u}(t,x,s_1). \end{aligned}\end{equation}$

下面将给出可容许策略与均衡策略的定义.

定义 3.2 (可容许策略) 对任意的 $t\in\left[T\right]$, 若投资策略 $u(t)=(\pi (t),\pi _{B}(t))$ 满足下述条件, 则称该策略为可容许策略

(1) 对于任意的 $t\in[T]$, $\pi (t),\pi _B(t)$$\mathcal{F}_t$ 循序可测的$;$

(2) 对于任意的 $(t,x,s_1)\in[T]\times \mathbb{R}^2$, 有$E_{t,x,s_1}^{\underline{\phi}}\left[\int_{t}^{T}\big(\pi ^2(s)+\pi _{B}^2(s)\big){\rm d}s\right]<+\infty$, 而且有

$E_{t,x,s_1}^{\overline{\phi}}\left[\int_{t}^{T}\big(\pi ^2(s)+\pi _{B}^2(s){\rm d}s\big)\right]<+\infty;$

(3) 对于任意的 $(t,x,s_1)\in[T]\times \mathbb{R}^2$, 随机微分方程(3.3) 存在唯一解 $X^{\phi}(t).$

$\mathcal{U}=\{u(t)|u(t)=(\pi (t),\pi _{B}(t)),t\in[T]\},$

为所有可容许策略构成的集合.

定义 3.3 (均衡策略) 对于一个可容许策略

$u^{*}(t)=\left\lbrace \pi ^{*}(t),\pi _{B}^{*}(t)\right\rbrace_{t\in[T]}\in \mathcal{U},$

定义如下策略

$\begin{equation}\begin{aligned} u_{\varepsilon}(s)=\begin{cases}(\tilde{\pi }(s),\tilde{\pi }_B(s)),\quad &t\le s <t+\varepsilon,\\u^{*}(s),\quad &t+\varepsilon\le s\le T,\end{cases} \end{aligned}\end{equation}$

其中 $\tilde{\pi }\in \mathbb{R},\tilde{\pi }_B\in \mathbb{R}$$\varepsilon > 0$. 若对任意 $\big(\pi ^{*}(t),\pi _{B}^{*}(t)\big)\in\mathbb{R}^2$, 且对任意的 $(t,x,s_1)\in[T]\times \mathbb{R}^2$, 有

$\begin{equation}\begin{aligned} \lim_{\varepsilon\downarrow0}\inf\frac{J_{\alpha}^{u^{*}}(t,x,s_1)-J_{\alpha}^{u_{\varepsilon}}(t,x,s_1)}{\varepsilon}\ge0, \end{aligned}\end{equation}$

那么 $u^{*}$ 就为问题 (3.8) 的均衡投资策略, $V(t,x,s_1)=J_{\alpha}^{u^{*}}(t,x,s_1)$ 为相应的均衡值函数.

4 主要结果

为方便起见, 令 $C^{1,2,2}([T]\times \mathbb{R}^2)$ 为所有连续函数 $\Psi(t,x,s_1)$ 构成的空间, 其中 $\Psi(t,\cdot,\cdot)$$[T]$ 上一阶连续可微, $\Psi(\cdot,x,\cdot)$$\mathbb{R}$ 上二阶连续可微, $\Psi(\cdot,\cdot,s_1)$$\mathbb{R}$ 上二阶连续可微. 对于任意的 $\Psi(t,x,s_1)\in C^{1,2,2}([T]\times \mathbb{R}^2)$, 定义如下微分算子

$\begin{aligned} \mathcal{A}^{u,\phi}\Psi(t,x,s_1)=\ &\Psi_t(t,x,s_1)+\Big[\pi \big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big) +\pi _{B}\sigma_{L}^2 \\ &+x\big(\mu_{0}-\mu_{L}\big) -\pi \sigma_{1}s_{1}^{\beta}\phi_1 +\sigma_{L}\big(\pi +\pi _{B}\big)\phi_2\Big]\Psi_x(t,x,s_1) \\ &+\frac{1}{2} \Big[\pi ^2\sigma_{1}^2s_{1}^{2\beta}+\sigma_{L}^2(\pi +\pi _{B})^2\Big]\Psi_{xx}(t,x,s_1)+ \mu_1s_1\Psi_{s_1}(t,x,s_1) \\ &+\frac{1}{2}\sigma_1^2s_1^{2\beta+2}\Psi_{s_1s_1}(t,x,s_1) +\pi \sigma_1^2s_1^{2\beta+1}\Psi_{xs_1}(t,x,s_1). \end{aligned}$

以下验证定理的证明与 Li 等[6]相似, 因此我们在此省略.

定理 4.1 (验证定理) 假设存在函数 $V(t,x,s_1),\underline{g}(t,x,s_1),\overline{g}(t,x,s_1)\in C^{1,2,2}([T]\times \mathbb{R}^2)$, 对任意的 $(t,x,s_1)\in[T]\times \mathbb{R}^2$, 有

(1)

$\begin{equation}\begin{aligned} 0=\,&\sup\limits_{u\in\mathcal{U}}\Big\{ \alpha \inf\limits_{\phi \in \Phi} \Big[ \mathcal{A}^{u,\phi}V(t,x,s_1)-\frac{\gamma}{2}\mathcal{A}^{u,\phi}\underline{g}^2(t,x,s_1) +\gamma \underline{g}(t,x,s_1)\mathcal{A}^{u,\phi}\underline{g}(t,x,s_1) +h^{\psi}(\phi(t)) \Big]\\ &+\hat{\alpha}\sup\limits_{\phi \in \Phi}\Big[\mathcal{A}^{u,\phi}V(t,x,s_1) -\frac{\gamma}{2}\mathcal{A}^{u,\phi}\overline{g}^2(t,x,s_1) +\gamma \overline{g}(t,x,s_1)\mathcal{A}^{u,\phi}\overline{g}(t,x,s_1)-h^{\psi}(\phi(t))\Big]\Big\}, \end{aligned}\end{equation}$
$\begin{equation} \begin{cases}V(T,x,s_1)=x,\\ \mathcal{A}^{u^{*},\underline{\phi}^{*}}\underline{g}(t,x,s_1)=\mathcal{A}^{u^{*},\overline{\phi}^{*}}\overline{g}(t,x,s_1)=0,\\ \underline{g}(T,x,s_1)=\overline{g}(T,x,s_1)=x.\end{cases} \end{equation}$

(2) 函数 $\pi ^{*}(t),\pi _{B}^{*}(t),\underline{\phi}^{*}(t),\overline{\phi}^{*}(t),\mathcal{A}^{u^{*},\underline{\phi}^{*}}\underline{g}(t,x,s_1),\mathcal{A}^{u^{*},\overline{\phi}^{*}}\overline{g}(t,x,s_1),\mathcal{A}^{u^{*}\underline{\phi}^{*}}\underline{g}^2(t,x,s_1)$ 以及 $\mathcal{A}^{u^{*},\overline{\phi}^{*}}\overline{g}^2(t,x,s_1)$ 都为关于 $t$ 的确定性函数并独立于 $x$.

(3) $\underline{\phi}^{*}=\underline{\phi}^{u^{*}},\overline{\phi}^{*}=\overline{\phi}^{u^{*}}$.

那么 $u^{*}$$\alpha$-鲁棒投资问题 (3.8) 的均衡策略, 且 $V(t,x,s_1)=J_{\alpha}^{u^{*}}(t,x,s_1)$ 为相应均衡值函数.

定理 4.2 问题 (3.8) 的均衡策略及值函数如下

(1) 均衡投资策略为

$\begin{aligned} &\pi ^{*}(t)=\frac{-\mu_1+\mu_0-2\beta\sigma_1^2k_4(t)}{{\rm e}^{(\mu_0-\mu_L)(T-t)}\sigma_1^2s_1^{2\beta}\big(-\gamma+(1-2\alpha)\psi_1\big)}, \end{aligned}$
$\begin{aligned} &\pi _B^{*}(t)\!=\!-\frac{1}{{\rm e}^{(\mu_0-\mu_L)(T-t)}(-\gamma+(1-2\alpha)\psi_2)}\!+\!\frac{\mu_1-\mu_0+2\beta\sigma_1^2k_4(t)}{{\rm e}^{(\mu_0-\mu_L)(T-t)}\sigma_1^2s_1^{2\beta}\big(-\gamma+(1-2\alpha)\psi_1\big)}. \end{aligned}$

(2) 相应的均衡值函数为

$\begin{equation} V(t,x,s_1)={\rm e}^{(\mu_0-\mu_L)(T-t)}x+\frac{B(t)}{\gamma}s_1^{-2\beta}+\frac{D(t)}{\gamma}, \end{equation}$

其中, $k_4(t),B(t),D(t)$ 见 (4.40)-(4.42) 式所示.

(4.2)式可写成

$\begin{equation}\begin{aligned} 0&=\sup\limits_{u\in\mathcal{U}}\Bigg\{ V_t+\Big[\pi \big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big)+\pi _{B}\sigma_{L}^2+x\big(\mu_{0}-\mu_{L}\big)\Big]V_x\Bigg.\\ & \Bigg.+\frac{1}{2}\Big[\pi ^2\sigma_{1}^2s_{1}^{2\beta}+\sigma_{L}^2(\pi +\pi _{B})^2\Big](V_{xx}-\alpha\gamma\underline{g}_{x}^{2}-\hat{\alpha}\gamma\overline{g}_{x}^{2})+\mu_1s_1V_{s_1}\Bigg.\\ & \Bigg.+\frac{1}{2}\sigma_1^2s_1^{2\beta+2}(V_{s_1s_1}-\alpha\gamma\underline{g}_{s_1}^{2}-\hat{\alpha}\gamma\overline{g}_{s_1}^{2})+\pi \sigma_1^2s_1^{2\beta+1}(V_{xs_1}-\alpha\gamma\underline{g}_{x}\underline{g}_{s_1}-\hat{\alpha}\gamma\overline{g}_{x}\overline{g}_{s_1})\\ & +\alpha \inf\limits_{\phi \in \Phi}\left\lbrace\left[-\pi \sigma_{1}s_{1}^{\beta}\phi_1+\sigma_{L}\big(\pi +\pi _{B}\big)\phi_2\right]V_x+\left[\frac{\phi_{1}^{2}(t)}{2\psi_1(t)}+\frac{\phi_{2}^{2}(t)}{2\psi_2(t)}\right]\right\rbrace\Bigg.\\ & \Bigg.+\hat{\alpha} \sup\limits_{\phi \in \Phi}\left\lbrace\left[-\pi \sigma_{1}s_{1}^{\beta}\phi_1+\sigma_{L}\big(\pi +\pi _{B}\big)\phi_2\right]V_x-\left[\frac{\phi_{1}^{2}(t)}{2\psi_1(t)}+\frac{\phi_{2}^{2}(t)}{2\psi_2(t)}\right]\right\rbrace \Bigg\}. \end{aligned}\end{equation}$

对 (4.7) 式应用一阶最优性条件有

$\begin{equation}\begin{aligned} \begin{cases} \underline{\phi}_{1}^{*}(t)=\psi_1 \pi \sigma_1 s_1^{\beta}V_x,\\ \underline{\phi}_{2}^{*}(t)=-\psi_2 \sigma_L(\pi +\pi _B)V_x, \end{cases} \end{aligned}\end{equation}$

$\begin{equation}\begin{aligned} \begin{cases} \overline{\phi}_{1}^{*}(t)=-\psi_1 \pi \sigma_1 s_1^{\beta}V_x,\\ \overline{\phi}_{2}^{*}(t)=\psi_2 \sigma_L(\pi +\pi _B)V_x. \end{cases} \end{aligned}\end{equation}$

将 (4.8) 式和 (4.9) 式代入 (4.7) 式中, 有

$\begin{aligned} 0&=\sup\limits_{u\in\mathcal{U}}\Bigg\{ V_t+\Big[\pi \big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big)+\pi _{B}\sigma_{L}^2+x\big(\mu_{0}-\mu_{L}\big)\Big]V_x\Bigg. \\ & \Bigg.+\frac{1}{2}\Big[\pi ^2\sigma_{1}^2s_{1}^{2\beta}+\sigma_{L}^2(\pi +\pi _{B})^2\Big](V_{xx}-\alpha\gamma\underline{g}_{x}^{2}-\hat{\alpha}\gamma\overline{g}_{x}^{2})+\mu_1s_1V_{s_1}\Bigg. \\ & \Bigg.+\frac{1}{2}\sigma_1^2s_1^{2\beta+2}(V_{s_1s_1}-\alpha\gamma\underline{g}_{s_1}^{2}-\hat{\alpha}\gamma\overline{g}_{s_1}^{2})+\pi \sigma_1^2s_1^{2\beta+1}(V_{xs_1}-\alpha\gamma\underline{g}_{x}\underline{g}_{s_1}-\hat{\alpha}\gamma\overline{g}_{x}\overline{g}_{s_1})\Bigg. \\ & \Bigg.+\frac{1-2\alpha}{2} \left[\pi ^2\sigma_{1}^2s_{1}^{2\beta}\psi_1+\sigma_{L}^2(\pi +\pi _{B})^2\psi_2\right] V_x^2 \Bigg\}. \end{aligned}$

受边界条件 (4.3) 式的启发, 假定 (4.10) 式的解形式如下

$\begin{equation}\begin{aligned} \begin{cases} V(t,x,s_1)=A(t)x+\frac{B(t)}{\gamma}s_1^{-2\beta}+\frac{D(t)}{\gamma},\\[3mm] \underline{g}(t,x,s_1)=\underline{a}(t)x+\frac{\underline{b}(t)}{\gamma}s_1^{-2\beta}+\frac{\underline{d}(t)}{\gamma},\\[3mm] \overline{g}(t,x,s_1)=\overline{a}(t)x+\frac{\overline{b}(t)}{\gamma}s_1^{-2\beta}+\frac{\overline{d}(t)}{\gamma}, \end{cases} \end{aligned}\end{equation}$

其中, $A(t),B(t),D(t),\underline{a}(t),\underline{b}(t),\underline{d}(t),\overline{a}(t),\overline{b}(t),\overline{d}(t)$$t$ 的函数且边界条件为

$\begin{equation}\begin{aligned} &A(T)=\underline{a}(T)=\overline{a}(T)=1,\\ B(T)=D(T)&=\underline{b}(T)=\underline{d}(T)=\overline{b}(T)=\overline{d}(T)=0, \end{aligned}\end{equation}$

则有

$\begin{equation}\begin{aligned} &V_t=A_tx+\frac{B_t}{\gamma}s_1^{-2\beta}+\frac{D_t}{\gamma}, V_x=A, V_{s_1}=-2\beta\frac{B}{\gamma}s_1^{-2\beta-1},\\&V_{xx}=V_{xs_1}=0, V_{s_1s_1}=2\beta(2\beta+1)\frac{B}{\gamma}s_1^{-2\beta-2},\\ &\underline{g}_t=\underline{a}_tx+\frac{\underline{b}_t}{\gamma}s_1^{-2\beta}+\frac{\underline{d}_t}{\gamma}, \underline{g}_x=\underline{a}, \underline{g}_{s_1}=-2\beta\frac{\underline{b}}{\gamma}s_1^{-2\beta-1},\\&\underline{g}_{xx}=\underline{g}_{xs_1}=0, \underline{g}_{s_1s_1}=2\beta(2\beta+1)\frac{\underline{b}}{\gamma}s_1^{-2\beta-2},\\ &\overline{g}_t=\overline{a}_tx+\frac{\overline{b}_t}{\gamma}s_1^{-2\beta}+\frac{\overline{d}_t}{\gamma}, \overline{g}_x=\overline{a}, \overline{g}_{s_1}=-2\beta\frac{\overline{b}}{\gamma}s_1^{-2\beta-1},\\&\overline{g}_{xx}=\overline{g}_{xs_1}=0, \overline{g}_{s_1s_1}=2\beta(2\beta+1)\frac{\overline{b}}{\gamma}s_1^{-2\beta-2}.\\ \end{aligned}\end{equation}$

将 (4.13) 式代入 (4.10) 式中, 化简有

$\begin{equation}\begin{aligned} 0&=\sup\limits_{u\in\mathcal{U}}\Bigg\{ A_tx+\frac{B_t}{\gamma}s_1^{-2\beta}+\frac{D_t}{\gamma}+\Big[\pi \big(\mu_{1}+\sigma_{L}^2-\mu_{0}\big)+\pi _{B}\sigma_{L}^2+x\big(\mu_{0}-\mu_{L}\big)\Big]A\Bigg.\\ & \Bigg.+\frac{1}{2}\Big[\pi ^2\sigma_{1}^2s_{1}^{2\beta}+\sigma_{L}^2(\pi +\pi _{B})^2\Big](-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2})-2\mu_1\beta\frac{B}{\gamma}s_1^{-2\beta}\Bigg.\\ & \Bigg.+\sigma_1^2\beta(2\beta+1)\frac{B}{\gamma}-2\sigma_1^2\frac{\beta^2}{\gamma}s_1^{-2\beta}(\alpha\underline{b}^2+\hat{\alpha}\overline{b}^2)+2\beta\pi \sigma_1^2\Big[\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b}\Big]\Bigg.\\ & \Bigg.+\frac{1-2\alpha}{2} \left[\pi ^2\sigma_{1}^2s_{1}^{2\beta}\psi_1+\sigma_{L}^2(\pi +\pi _{B})^2\psi_2\right] A^2 \Bigg\}. \end{aligned}\end{equation}$

对 (4.14) 式的 $\pi,\pi _B$ 求一阶导, 有

$\begin{equation}\begin{aligned} \pi ^{*}&=\frac{-A(\mu_{1}-\mu_{0})-2\beta\sigma_1^2(\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b})}{\sigma_{1}^2 s_1^{2\beta}\left[-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2\psi_1 \right]}, \end{aligned}\end{equation}$
$\begin{equation}\begin{aligned} \pi _B^{*}=\frac{-A}{-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2}+\frac{A(\mu_{1}-\mu_{0})+2\beta\sigma_1^2(\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b})}{\sigma_{1}^2 s_1^{2\beta}\left[-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2\psi_1 \right]}. \end{aligned}\end{equation}$

将 (4.15)、(4.16) 式代入 (4.14) 式化简后变量分离, 有

$\begin{aligned} &A_tx+Ax(\mu_{0}-\mu_{L})=0, \end{aligned}$
$\begin{aligned} &B_t-\frac{\gamma\Big[-A(\mu_{1}-\mu_{0})-2\beta\sigma_1^2(\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b})\Big]^2}{2\sigma_{1}^2 \left[-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2\psi_1 \right]}-2\mu_1\beta B-2\sigma_1^2\beta^2(\alpha\underline{b}^2+\hat{\alpha}\overline{b}^2)=0, \end{aligned}$
$\begin{aligned} &D_t-\frac{\gamma A^2\sigma_L^2}{2\big(-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2\big)}+\sigma_1^2\beta(2\beta+1)B=0. \end{aligned}$

而后再结合 (4.1)式, 将 (4.8)、(4.9)、(4.13)、(4.15)、(4.16)式的值代入 (4.3) 式的

$\mathcal{A}^{u^{*},\underline{\phi}^{*}}\underline{g}(t,x,s_1)=0,$
$\mathcal{A}^{u^{*},\overline{\phi}^{*}}\overline{g}(t,x,s_1)=0,$

化简后变量分离, 分别有

$\begin{aligned} &\underline{a}_tx+\underline{a}x\big(\mu_{0}-\mu_{L}\big)=0, \end{aligned}$
$\begin{array}{l}\underline{b}_{t}+\left(\mu_{1}-\mu_{0}\right) \frac{\gamma \underline{a}\left(-A\left(\mu_{1}-\mu_{0}\right)-2 \beta \sigma_{1}^{2}(\alpha \underline{a} \underline{b}+\hat{\alpha} \bar{a} \bar{b})\right)}{\sigma_{1}^{2}\left(-\alpha \gamma \underline{a}^{2}-\hat{\alpha} \gamma \bar{a}^{2}+(1-2 \alpha) A^{2} \psi_{1}\right)} \\-\frac{\gamma \psi_{1} A \underline{a}\left(-A\left(\mu_{1}-\mu_{0}\right)-2 \beta \sigma_{1}^{2}(\alpha \underline{a} \underline{b}+\hat{\alpha} \bar{a} \bar{b})\right)^{2}}{\sigma_{1}^{2}\left(-\alpha \gamma \underline{a}^{2}-\hat{\alpha} \gamma \bar{a}^{2}+(1-2 \alpha) A^{2} \psi_{1}\right)^{2}}-2 \mu_{1} \beta \underline{b}=0\end{array}$
$\begin{array}{l}\underline{d}_{t}-\frac{\gamma \underline{a} A \sigma_{L}^{2}}{-\alpha \gamma \underline{a}^{2}-\hat{\alpha} \gamma \bar{a}^{2}+(1-2 \alpha) A^{2} \psi_{2}} \\-\frac{\gamma \psi_{2} \sigma_{L}^{2} A^{3} \underline{a}}{\left(-\alpha \gamma \underline{a}^{2}-\hat{\alpha} \gamma \bar{a}^{2}+(1-2 \alpha) A^{2} \psi_{2}\right)^{2}}+\sigma_{1}^{2} \beta(2 \beta+1) \underline{b}=0,\end{array}$
$\begin{aligned} &\underline{d}_t-\frac{\gamma\underline{a}A\sigma_L^2}{-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2}\Bigg. \\&\Bigg.-\frac{\gamma\psi_2\sigma_L^2A^3\underline{a}}{\big(-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2\big)^2}+\sigma_1^2\beta(2\beta+1)\underline{b}=0, \end{aligned}$(4.22) $\begin{aligned} &\overline{a}_tx+\overline{a}x\big(\mu_{0}-\mu_{L}\big)=0, \end{aligned}$
$\begin{aligned} &\overline{b}_t+(\mu_1-\mu_0)\frac{\gamma\overline{a}\big(-A(\mu_{1}-\mu_{0})-2\beta\sigma_1^2(\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b})\big)}{\sigma_{1}^2 \big(-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2\psi_1 \big)}\Bigg. \\&\Bigg.+\frac{\gamma\psi_1A\overline{a}\big(-A(\mu_{1}-\mu_{0})-2\beta\sigma_1^2(\alpha\underline{a} \underline{b}+\hat{\alpha}\overline{a}\overline{b})\big)^2}{\sigma_{1}^2 \big(-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2\psi_1 \big)^2}-2\mu_1\beta\overline{b}=0, \end{aligned}$
$\begin{aligned} &\overline{d}_t-\frac{\gamma\overline{a}A\sigma_L^2}{-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2}\Bigg. \\&\Bigg.+\frac{\gamma\psi_2\sigma_L^2A^3\overline{a}}{\big(-\alpha\gamma\underline{a}^{2}-\hat{\alpha}\gamma\overline{a}^{2}+(1-2\alpha)A^2 \psi_2\big)^2}+\sigma_1^2\beta(2\beta+1)\overline{b}=0. \end{aligned}$

运用 (4.12) 式的边界条件, 求解 (4.17)、(4.20) 式和 (4.23) 式, 得

$\begin{equation}\begin{aligned} A(t)=\underline{a}(t)=\overline{a}(t)={\rm e}^{(\mu_0-\mu_L)(T-t)}. \end{aligned}\end{equation}$

将 (4.26) 式代回 (4.18)、(4.19)、(4.21)、(4.22)、(4.24) 及 (4.25)式, 得

$B_t-\frac{\gamma\Big[\mu_{1}-\mu_{0}+2\beta\sigma_1^2(\alpha\underline{b}+\hat{\alpha}\overline{b})\Big]^2}{2\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)}-2\mu_1\beta B-2\sigma_1^2\beta^2(\alpha\underline{b}^2+\hat{\alpha}\overline{b}^2)=0$
$D_t-\frac{\gamma \sigma_L^2}{2\big(-\gamma+(1-2\alpha) \psi_2\big)}+\sigma_1^2\beta(2\beta+1)B=0,$
$\underline{b}_t-\gamma(\mu_1-\mu_0)\frac{\mu_{1}-\mu_{0}+2\beta\sigma_1^2(\alpha\underline{b}+\hat{\alpha}\overline{b})}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1 \big)}-\gamma\psi_1\frac{\big(\mu_{1}-\mu_{0}+2\beta\sigma_1^2(\alpha\underline{b}+\hat{\alpha}\overline{b})\big)^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)^2}-2\mu_1\beta\underline{b}=0,$
$\underline{d}_t-\frac{\gamma\sigma_L^2}{-\gamma+(1-2\alpha) \psi_2}-\frac{\gamma\psi_2\sigma_L^2}{\big(-\gamma+(1-2\alpha) \psi_2\big)^2}+\sigma_1^2\beta(2\beta+1)\underline{b}=0,$
$\overline{b}_t-\gamma(\mu_1-\mu_0)\frac{\mu_{1}-\mu_{0}+2\beta\sigma_1^2(\alpha\underline{b}+\hat{\alpha}\overline{b})}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1 \big)}+\gamma\psi_1\frac{\big(\mu_{1}-\mu_{0}+2\beta\sigma_1^2(\alpha\underline{b}+\hat{\alpha}\overline{b})\big)^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)^2}-2\mu_1\beta\overline{b}=0,$
$\overline{d}_t-\frac{\gamma\sigma_L^2}{-\gamma+(1-2\alpha) \psi_2}+\frac{\gamma\psi_2\sigma_L^2}{\big(-\gamma+(1-2\alpha) \psi_2\big)^2}+\sigma_1^2\beta(2\beta+1)\overline{b}=0.$

这里将 (4.29) 式乘上系数 $\alpha$ 再与 (4.31) 式乘上系数 $\hat{\alpha}$ 后的公式进行求和运算, 整理可得

$\begin{equation}\begin{aligned} 0&=\alpha\underline{b}_t+\hat{\alpha}\overline{b}_t-\Big(2\mu_1\beta+\frac{2\beta\gamma(\mu_1-\mu_0)}{ -\gamma+(1-2\alpha)\psi_1}-\frac{4\beta\gamma\psi_1(\mu_{1}-\mu_{0})(1-2\alpha)}{ \big(-\gamma+(1-2\alpha)\psi_1\big)^2}\Big)(\alpha\underline{b}+\hat{\alpha}\overline{b})\\ & +\frac{4\beta^2\sigma_1^2\gamma\psi_1(1-2\alpha)}{\big(-\gamma+(1-2\alpha)\psi_1\big)^2}(\alpha\underline{b}+\hat{\alpha}\overline{b})^2 -\frac{\gamma(\mu_1-\mu_0)^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)}+\frac{\gamma\psi_1(1-2\alpha)(\mu_{1}-\mu_{0})^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)^2}. \end{aligned}\end{equation}$

求解过程涉及解黎卡提方程, 为计算简便, 这里采用 Sun 等[13]的方法, 求解可得

$\begin{equation}\begin{aligned} \alpha\underline{b}+\hat{\alpha}\overline{b}=\frac{k_3[1-{\rm e}^{k_1(T-t)}]}{2k_1+(k_1+k_2)[{\rm e}^{k_1(T-t)}-1]}, \end{aligned}\end{equation}$

其中

$\begin{equation}\begin{aligned} k_1&=\Bigg\{\Big(2\mu_1\beta+\frac{2\beta\gamma(\mu_1-\mu_0)}{ -\gamma+(1-2\alpha)\psi_1}-\frac{4\beta\gamma\psi_1(\mu_{1}-\mu_{0})(1-2\alpha)}{ \big(-\gamma+(1-2\alpha)\psi_1\big)^2}\Big)^2\Bigg.\\ & \Bigg.+\frac{16\beta^2\sigma_1^2\gamma\psi_1(1-2\alpha)}{\big(-\gamma+(1-2\alpha)\psi_1\big)^2}\Big[\frac{\gamma(\mu_1-\mu_0)^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)}-\frac{\gamma\psi_1(1-2\alpha)(\mu_{1}-\mu_{0})^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)^2}\Big]\Bigg\}^{\frac{1}{2}},\\ k_2&=2\mu_1\beta+\frac{2\beta\gamma(\mu_1-\mu_0)}{ -\gamma+(1-2\alpha)\psi_1}-\frac{4\beta\gamma\psi_1(\mu_{1}-\mu_{0})(1-2\alpha)}{ \big(-\gamma+(1-2\alpha)\psi_1\big)^2},\\ k_3&=\frac{2\gamma(\mu_1-\mu_0)^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)}-\frac{2\gamma\psi_1(1-2\alpha)(\mu_{1}-\mu_{0})^2}{\sigma_{1}^2 \big(-\gamma+(1-2\alpha)\psi_1\big)^2}. \end{aligned}\end{equation}$

将 (4.34) 式代回 (4.27)-(4.32) 式, 求解可得

$\underline{b}(t)=\Big(\int_T^t\frac{\gamma}{\sigma_1^2}\big[(\mu_1-\mu_0)k_5(s)+\psi_1k_5^2(s)\big]{\rm e}^{2\mu_1\beta(T-s)}{\rm d}s\Big){\rm e}^{-2\mu_1\beta(T-t)},$
$\overline{b}(t)=\Big(\int_T^t\frac{\gamma}{\sigma_1^2}\big[(\mu_1-\mu_0)k_5(s)-\psi_1k_5^2(s)\big]{\rm e}^{2\mu_1\beta(T-s)}{\rm d}s\Big){\rm e}^{-2\mu_1\beta(T-t)},$
$\underline{d}(t)=\Big[-\frac{\gamma\sigma_L^2}{-\gamma+(1-2\alpha) \psi_2}-\frac{\gamma\psi_2\sigma_L^2}{\big(-\gamma+(1-2\alpha) \psi_2\big)^2}\Big](T-t)-\sigma_1^2\beta(2\beta+1)\int_T^t\underline{b}(s){\rm d}s,$
$\overline{d}(t)=\Big[-\frac{\gamma\sigma_L^2}{-\gamma+(1-2\alpha) \psi_2}+\frac{\gamma\psi_2\sigma_L^2}{\big(-\gamma+(1-2\alpha) \psi_2\big)^2}\Big](T-t)-\sigma_1^2\beta(2\beta+1)\int_T^t\overline{b}(s){\rm d}s,$
$B(t)={\rm e}^{-2\mu_1\beta(T-t)}\bigg\{\int_T^t{\rm e}^{2\mu_1\beta(T-s)}\bigg\{\frac{\gamma k_5(s)}{2\sigma_1^2}(\mu_1-\mu_0+2\beta\sigma_1^2k_4(s)) \\ +2\sigma_1^2\beta^2(\alpha\underline{b}^2(s)+\hat{\alpha}\overline{b}^2(s))\bigg\}{\rm d}s\bigg\},$
$D(t)=-\frac{\gamma\sigma_L^2}{2\big(-\gamma+(1-2\alpha) \psi_2\big)}(T-t)-\sigma_1^2\beta(2\beta+1)\int_T^tB(s){\rm d}s,$

其中

$\begin{equation}\begin{aligned} k_4(t)=\frac{k_3[1-{\rm e}^{k_1(T-t)}]}{2k_1+(k_1+k_2)[{\rm e}^{k_1(T-t)}-1]}, k_5(t)=\frac{\mu_1-\mu_0+2\beta\sigma_1^2k_4(t)}{-\gamma+(1-2\alpha)\psi_1}. \end{aligned}\end{equation}$

这里将 (4.13)、(4.26)、(4.34)-(4.42) 式结合, 代入(4.8)、(4.9)、(4.11) 式与 (4.15)、(4.16) 式可得 $\underline{\phi}^{*},\overline{\phi}^{*},\pi ^*,\pi _B^*$ 以及值函数 $V(t,x,s_1)$ 的具体表达式.

最后可以验证上述结果满足验证定理的条件, 从而定理证毕.

推论 4.1$\alpha=1$, 问题 [3.8] 退化为极端模糊厌恶的最优投资问题, 其均衡策略及值函数如下

(1) 均衡投资策略为

$\begin{equation}\begin{aligned} \pi ^{*}(t)=\frac{\mu_1-\mu_0+2\beta\sigma_1^2\tilde{k}_4(t)}{{\rm e}^{(\mu_0-\mu_L)(T-t)}\sigma_1^2s_1^{2\beta}(\gamma+\psi_1)}, \end{aligned}\end{equation}$
$\begin{equation}\begin{aligned} \pi _B^{*}(t)&=\frac{1}{{\rm e}^{(\mu_0-\mu_L)(T-t)}(\gamma+\psi_2)}-\frac{\mu_1-\mu_0+2\beta\sigma_1^2\tilde{k}_4(t)}{{\rm e}^{(\mu_0-\mu_L)(T-t)}\sigma_1^2s_1^{2\beta}(\gamma+\psi_1)}. \end{aligned}\end{equation}$

(2) 相应的均衡值函数为

$\begin{equation}\begin{aligned} V(t,x,s_1)={\rm e}^{(\mu_0-\mu_L)(T-t)}x+\frac{\tilde{B}(t)}{\gamma}s_1^{-2\beta}+\frac{\tilde{D}(t)}{\gamma}, \end{aligned}\end{equation}$

其中, 当 $\tilde{k}_1,\tilde{k}_2,\tilde{k}_3$ 为下式时,

$\begin{equation}\begin{aligned} \tilde{k}_1&=\Bigg\{\Big(2\mu_1\beta-\frac{2\beta\gamma(\mu_1-\mu_0)}{ \gamma+\psi_1}+\frac{4\beta\gamma\psi_1(\mu_{1}-\mu_{0})}{ (\gamma+\psi_1\big)^2}\Big)^2\Bigg.\\ & \Bigg.-\frac{16\beta^2\sigma_1^2\gamma\psi_1}{(\gamma+\psi_1)^2}\Big[\frac{\gamma\psi_1(\mu_{1}-\mu_{0})^2}{\sigma_{1}^2 (\gamma+\psi_1)^2}-\frac{\gamma(\mu_1-\mu_0)^2}{\sigma_{1}^2 (\gamma+\psi_1)}\Big]\Bigg\}^{\frac{1}{2}},\\ \tilde{k}_2&=2\mu_1\beta-\frac{2\beta\gamma(\mu_1-\mu_0)}{ \gamma+\psi_1}+\frac{4\beta\gamma\psi_1(\mu_{1}-\mu_{0})}{ (\gamma+\psi_1\big)^2},\\ \tilde{k}_3&=\frac{2\gamma\psi_1(\mu_{1}-\mu_{0})^2}{\sigma_{1}^2 (\gamma+\psi_1)^2}-\frac{2\gamma(\mu_1-\mu_0)^2}{\sigma_{1}^2 (\gamma+\psi_1)}, \end{aligned}\end{equation}$

对应的 $\tilde{k}_4(t),\tilde{k}_5(t),\tilde{\underline{b}}(t),\tilde{B}(t),\tilde{D}(t)$

$\tilde{k}_4(t)=\frac{\tilde{k}_3[1-{\rm e}^{\tilde{k}_1(T-t)}]}{2\tilde{k}_1+(\tilde{k}_1+\tilde{k}_2)[{\rm e}^{\tilde{k}_1(T-t)}-1]},$
$\tilde{k}_5(t)=-\frac{\mu_1-\mu_0+2\beta\sigma_1^2\tilde{k}_4(t)}{\gamma+\psi_1},$
$\tilde{\underline{b}}(t)=\Big(\int_T^t\frac{\gamma}{\sigma_1^2}\big[(\mu_1-\mu_0)\tilde{k}_5(s)+\psi_1\tilde{k}_5^2(s)\big]{\rm e}^{2\mu_1\beta(T-s)}{\rm d}s\Big){\rm e}^{-2\mu_1\beta(T-t)},$
$\tilde{B}(t)={\rm e}^{-2\mu_1\beta(T-t)}\Bigg\{\int_T^t{\rm e}^{2\mu_1\beta(T-s)}\bigg\{2\sigma_1^2\beta^2\tilde{\underline{b}} ^2(s)-\frac{\gamma(\mu_1-\mu_0+2\beta\sigma_1^2\tilde{k}_4(s))^2}{2\sigma_1^2(\gamma+\psi_1)}\bigg\}{\rm d}s\Bigg\},$
$\tilde{D}(t)=\frac{\gamma\sigma_L^2}{2(\gamma+\psi_2)}(T-t)-\sigma_1^2\beta(2\beta+1)\int_T^t\tilde{B}(s){\rm d}s.$

由上述推论可得, 若 $\mu_L=\sigma_L=0,\beta=0$, 则问题退化为不考虑通货膨胀风险的最优投资问题且股票价格服从几何布朗运动模型.

5 数值分析

本节主要通过曲线变化趋势简洁描述参数对最优投资策略的影响, 除特别说明外, 基本参数设置见表1.

表1   参数设置

新窗口打开| 下载CSV


图1 显示: 风险资产的投资数量随着时间 $t$ 的增加呈现先增后减的趋势,且随着股票价格过程模糊程度 $\psi_1$ 的增加而降低. 即随着时间推移, 投资者应当先增加购买风险资产, 而后再减少对风险资产购买, 才能更好的控制投资风险, 实现效用最大化. 另外投资者对风险资产价格的模糊程度越高, 则其越难把握风险资产的收益率, 为了降低风险, 应减少对风险资产的投资.

图1

图1   $t,\psi_1$$\pi ^*(t)$ 的影响


图2 显示: 无风险资产的投资数量随着时间 $t$ 及通货膨胀波动率过程模糊水平 $\psi_2$ 的增加而降低. 无风险资产在通胀折现后变为带漂移的几何布朗运动形式, 因此同样基于对投资风险的把控, 应该逐渐减少对无风险资产的购买. 另外对通货膨胀波动率过程的模糊程度越高, 也会使得无风险资产实际的收益情况难以知晓, 为了降低风险, 也应当减少购买无风险资产, 研究结果与通常认知相符.

图2

图2   $t,\psi_2$$\pi _B^*(t)$ 的影响


图3 显示: 风险资产的投资数量随着模糊厌恶型投资者模糊态度 $\alpha$ 的增加呈现先增后减的趋势, 且随着风险厌恶系数 $\gamma$ 的增加而减少. 投资者由模糊中立态度向模糊厌恶态度转变时, 对风险资产的投资数量也会随之先增后减, 投资者越厌恶模糊, 则越会减少对风险资产的投资来降低风险. 同样当风险厌恶系数越大时, 投资者则越想规避风险, 因而减少对风险资产的投资.

图3

图3   $\alpha,\gamma$$\pi ^*(t)$ 的影响


图4 显示: 无风险资产的投资数量随着模糊偏好型投资者模糊态度的增加呈现先减后增的趋势, 且随着风险厌恶系数 $\gamma$ 的增加而降低. 投资者由模糊偏好态度向模糊中立态度转变时, 对无风险资产的投资数量也会随之先减后增, 投资者越对模糊呈中立态度, 则越会增加无风险资产投资. 而当风险厌恶系数越大时, 投资者也将越想规避风险, 则其将减少对通胀折现后无风险资产的投资.

图4

图4   $\alpha,\gamma$$\pi _B^*(t)$ 的影响


参考文献

Zhang A H, Korn R, Ewald C O.

Optimal management and inflation protection for defined contribution pension plans

Blätter Der DGVFM, 2007, 28(2): 239-258

[本文引用: 1]

Siu T K.

Long-term strategic asset allocation with inflation risk and regime swiching

Quantitative Finance, 2011, 11(10): 1565-1580

[本文引用: 1]

Kwak M, Lim B H.

Optimal portfolio selection with life insurance under inflation risk

Journal of Banking & Finance, 2014, 46: 59-71

[本文引用: 1]

Li B, Luo P, Xiong D W.

Equilibrium strategies for alpha-maxmin expected utility maximization

SIAM Journal on Financial Mathematics, 2019, 10(2): 394-429

DOI:10.1137/18M1178542      [本文引用: 1]

In the existing literature of robust utility maximization with ambiguity, agents are generally assumed to be extremely ambiguity-averse as they tend to only consider expected payoffs in the worst-case scenario. However, experimental studies have shown that agents' attitude to ambiguity is not systematically negative and can even be ambiguity-seeking when they consider themselves knowledgeable or competent. To conceptually distinguish between an agent's perception of ambiguity and ambiguity aversion, the so-called alpha-maxmin expected utility (alpha-MEU) was proposed in the economics literature as a linear aggregation of the most and least favorable prior beliefs. Although the axiomatic characterization of alpha-MEU has been well studied, there has been little work on the benchmark maximization problem for alpha-MEU. The main difficulty stems from the dynamic inconsistency of two distinct extreme priors and nonconvexity (and nonconcavity) of the value function. In this paper, we study the maximization problem for alpha-MEU and solve for the equilibrium strategies of open-loop type. Under logarithmic risk preference, we obtain the explicit form of equilibrium investment strategies, which involves a two-dimensional system of fully coupled quadratic backward stochastic differential equations (BSDE). The main challenge in completing the verification theorem is to study the existence, uniqueness, and stability of this system of BSDE. For this purpose, we consider a general Markov model of the financial market, which leads to a system of quadratic Markovian BSDE. We find that the equilibrium investment strategy becomes more conservative if the agent is more ambiguity-averse or the agent perceives more ambiguity in the financial market. The equilibrium strategy is close to the classical nonrobust strategy (without ambiguity) if the agent is ambiguity-neutral. As time approaches maturity, ambiguity-seeking (ambiguity-averse) agents adopt more aggressive (conservative) equilibrium strategies. Additionally, the equilibrium strategy of an ambiguity-neutral agent converges to the classical nonrobust strategy at maturity.

石媛, 赵永霞.

违约风险下目标收益型养老金计划的 $\alpha$-鲁棒最优投资策略

数学物理学报, 2022, 42A(3): 943-960

[本文引用: 1]

Shi Y, Zhao Y X.

$\alpha$-robust optimal investment strategy for target benefit pension plans under default risk

Acta Math Sci, 2022, 42A(3): 943-960

[本文引用: 1]

Li B, Li D P, Xiong D W.

Alpha-robust mean-variance reinsurance-investment strategy

Journal of Economic Dynamics and Control, 2016, 70: 101-123

[本文引用: 3]

Yu X Y, Shen Y, Li X, Fan K.

Portfolio selection with parameter uncertainty under $\alpha$-maxmin mean-variance criterion

Operations Research Letters, 2020, 48(6): 720-724

[本文引用: 1]

Zhang L M, Li B.

Optimal reinsurance under the $\alpha$-maxmin mean-variance criterion

Insurance: Mathematics and Economics, 2021, 101: 225-239

[本文引用: 1]

Li D P, Bi J N, Hu M C.

Alpha-robust mean-variance investment strategy for DC pension plan with uncertainty about jump-diffusion risk

RAIRO-Operations Research, 2021, 55: S2983-S2997

[本文引用: 1]

Chen D S, He Y, Li Z Q.

Robust optimal reinsurance-investment for $\alpha$-maxmin mean-variance utility under Heston's SV model

The North American Journal of Economics and Finance, 2023, 67: 101921

[本文引用: 1]

Guan G H, Liang Z X, Song Y L.

A Stackelberg reinsurance-investment game under $\alpha$-maxmin mean-variance criterion and stochastic volatility

Scandinavian Actuarial Journal, 2024, 2024(1): 28-63

[本文引用: 1]

Huang Z Y, Wang H Y, Wu Z.

A kind of optimal investment problem under inflation and uncertain time horizon

Applied Mathematics and Computation, 2020, 375: 125084

[本文引用: 1]

Sun J Y, Yao H X, Kang Z L.

Robust optimal investment-reinsurance strategies for an insurer with multiple dependent risks

Insurance: Mathematics and Economics, 2019, 89: 157-170

[本文引用: 1]

/