Stochastic maximum principle for sub-diffusions and its applications

Shuaiqi Zhang,Zhen-Qing Chen
2023-11-26
Abstract:In this paper, we study optimal stochastic control problems for stochastic systems driven by non-Markov sub-diffusion $B_{L_t}$, which have the mixed features of deterministic and stochastic controls. Here $B_t$ is the standard Brownian motion on $R$, and $L_t:= \inf\{r>0: S_r>t\}, \quad t\geq 0,$ is the inverse of a subordinator $S_t$ with drift $\kappa >0$ that is independent of $B_t$. We obtain stochastic maximum principles (SMP) for these systems using both convex and spiking variational methods, depending on whether the convex domain is convex or not. To derive SMP, we first establish a martingale representation theorem for sub-diffusions $B_{L_t}$, and then use it to derive the existence and uniqueness result for the solutions of backward stochastic differential equations (BSDEs) driven by sub-diffusions, which may be of independent interest. We also derive sufficient SMPs. Application to a linear quadratic system is given to illustrate the main results of this paper.
Probability,Optimization and Control
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the optimal stochastic control problem of stochastic systems driven by non - Markovian sub - diffusion. Specifically, the author has studied the optimal control problem of stochastic differential equations (SDEs) driven by non - Markovian sub - diffusion \(BL_t\), and these sub - diffusions have the mixed characteristics of deterministic and stochastic controls. ### Main problem description of the paper: 1. **Research background**: - The traditional Brownian motion models (such as the Black - Scholes model) are used to describe stock prices, but they cannot capture the phenomenon that trading activities are significantly reduced in a bear market. - Sub - diffusion can better simulate this phenomenon because they remain unchanged in some time periods (i.e., the market is in an inactive state), and show stochastic fluctuations in other time periods. 2. **Research objectives**: - Derive the stochastic maximum principle (SMP) applicable to stochastic systems driven by sub - diffusion. This includes necessary and sufficient conditions. - Establish the martingale representation theorem of sub - diffusion and use it to derive the existence and uniqueness of solutions to backward stochastic differential equations (BSDEs) driven by sub - diffusion. - Apply these theoretical results to specific linear - quadratic systems to verify their validity and practicality. ### Formula summary: - **Sub - diffusion definition**: \[ BL_t \] is a continuous martingale, where \[ \langle BL_t \rangle = L_t \]. - **Inverse subordinator definition**: \[ L_t:=\inf\{ r > 0: S_r > t \}\], where \( S_t \) is a subordinator with drift \(\kappa > 0\). - **Martingale representation theorem**: For any square - integrable \( F'_T\)-measurable random variable \(\xi\), there exists an \(\{F'_t\}\)-progressively measurable process \( H_s \) such that \[ \xi=\mathbb{E}[\xi]+\int_0^T H_s dBL(s - a)^+\]. - **BSDE form**: \[ dY_t = h_1(t, Y_t)dt+h_2(t, Y_t, Z_t)dL(t - a)^++Z_t dBL(t - a)^+\], where \( Y(T)=\xi\). ### Conclusion: By introducing sub - diffusion and the corresponding martingale representation theorem, the author has successfully derived the stochastic maximum principle applicable to stochastic systems driven by sub - diffusion and proved the existence and uniqueness of solutions to relevant BSDEs. These results not only extend the classical stochastic optimization theory but also provide new tools and methods for dealing with more complex financial market models.