Stackelberg game approach to mixed stochastic \(H_2 /H_{\infty}\) control for mean-field jump-diffusions systems (Q6142539)

The authors consider a filtered complete probability space \((\Omega , \mathcal{F},\mathbb{F},\mathbb{P})\), assuming that the filtration \(\mathbb{F} =\{F(r)\}_{0\leq r\leq T}\) satisfies the usual conditions of right continuity and \(\mathbb{P}\)-completeness, and the linear mean-field jump-diffusion system: \(dx(r)=[A(r)x(r-)+\overline{A}(r)\mathbb{E} [x(r-)]+B_{1}(r)u(r)+\overline{B}_{1}(r)\mathbb{E}[u(r)]+B_{2}(r)w(r)+ \overline{B}_{2}(r)\mathbb{E}[w(r)]dr+[C(r)x(r-)+\overline{C}(r)\mathbb{E} [x(r-)]+D_{1}(r)u(r)+\overline{D}_{1}(r)\mathbb{E}[u(r)]+D_{2}(r)w(r)+ \overline{D}_{2}(r)\mathbb{E}[w(r)]dW(r)+\int_{\mathbb{Z}}[E(\mathfrak{z} ,r)x(r-)+\overline{E}(\mathfrak{z},r)\mathbb{E}[x(r-)]+F_{1}(\mathfrak{z} ,r)u(r)+\overline{F}_{1}(\mathfrak{z},r)\mathbb{E}[u(r)]+F_{2}(\mathfrak{z} ,r)w(r)+\overline{F}_{2}(\mathfrak{z},r)\mathbb{E}[w(r)]]\widetilde{N}(d \mathfrak{z},dr)\), where \(x=x(t)\in \mathbb{R}^{n}\) is the state process, \( u(\cdot )\in \mathbb{R}^{m_{1}}\) the control input, \(w(\cdot )\in \mathbb{R} ^{m_{2}}\) the disturbance, \(z(\cdot )=\left( \begin{array}{c} Q(r)x(r) \\ R_{1}(r)u(r) \\ R_{2}(r)w(r) \end{array} \right) \in \mathbb{R}^{q}\) the output, \(A(\cdot ),\overline{A}(\cdot ),B_{i}(\cdot ),\overline{B}_{i}(\cdot ),C(\cdot ),\overline{C}(\cdot ),D_{i}(\cdot ),\overline{D}_{i}(\cdot ),E(\cdot ),\overline{E}(\cdot ),F_{i}(\)\textperiodcentered \(,\cdot ),\overline{F}_{i}(\cdot ,\cdot )\), \( i=1,2\), given deterministic matrix-valued functions with appropriate dimensions, \(\mathbb{E}\) the mathematical expectation, \(\{\widetilde{N}(d \mathfrak{z},dr)\}_{t\leq r\leq T}\) a Poisson random martingale measure, and \(W(\cdot )=\{W(r)\}_{t\leq r\leq T}\) a 1D standard Brownian motion. The authors decompose the mixed stochastic \(H_{2}/H_{\infty }\) control problem for the above system, first analyzing the \(H_{\infty }\) problem, considered as the follower problem. \(\mathbb{K}\) being a Hilbert space, they define the space \(\mathcal{S}_{\mathcal{F}}^{2}(t,T;\mathbb{K})\) as the space of all \( \mathbb{K}\)-valued \(\mathcal{F}_{r}\)-adapted càdlàg processes \( g(\cdot )\) satisfying \(\mathbb{E}[sup_{t\leq r\leq T}\left\Vert g(r)\right\Vert _{\mathbb{K}}^{2}]<\infty \), the space \(\mathcal{L}_{ \mathcal{F}}^{2}(t,T;\mathbb{K})\) of all \(\mathbb{K}\)-valued \(\mathcal{F}_{r} \)-adapted stochastic processes \(g(\cdot )\) satisfying \(\mathbb{E}[\left\Vert \int_{t}^{T}g(r)\right\Vert _{\mathbb{K}}^{2}dr]<\infty \), the space \( \mathcal{L}_{\mathcal{F},p}^{2}(t,T;\mathbb{K})\) of all \(\mathbb{K}\)-valued \( \mathcal{F}_{r}\)-predictable stochastic processes \(g(\cdot )\) satisfying \( \mathbb{E}[\left\Vert \int_{t}^{T}g(r)dr\right\Vert _{\mathbb{K} }^{2}]<\infty \), for \(g:[t,T]\times \Omega \rightarrow \mathbb{K}\), and \( \mathcal{U}_{i}=\mathcal{L}_{\mathcal{F},p}^{2}(t,T;\mathbb{R}^{m_{i}})\). They derive the coupled system of two Riccati differential equations and the mean-field backward stochastic differential equations using the four-step scheme. Assuming the existence of solutions to the coupled Riccati differential equations, they prove the existence of a solution to the mean-field backward stochastic differential equations in \(\mathcal{S}_{ \mathcal{F}}^{2}(t,T;\mathbb{R}^{n})\times \mathcal{L}_{\mathcal{F} ,p}^{2}(t,T;\mathbb{R}^{n})\times \mathcal{L}_{\mathcal{F}}^{\nu ,2}([t,T]\times \mathbb{Z};\mathbb{R}^{n})\), they derive an explicit feedback representation of the optimal control associated with the follower problem and an equivalent formulation of the cost function. The authors then analyze the \(H_{2}\) problem as the leader problem. They build the mean-field forward-backward stochastic differential equations. They prove that \(u^{\ast }(\cdot )\in \mathcal{U}_{1}\) is the optimal control for the \(H_{2}\) criterion if and only if the admissible control \(u^{\ast }(\cdot )\) satisfies some stationarity condition. Applying the convex variational method, they establish the mean-field stochastic maximum principle. Considering the two situations where the Poisson process \(N\) has a unit jump, or the control of follower does not appear in the part of the jump diffusion, they obtain the explicit feedback representation of the optimal control for the leader problem. They finally obtain a state-feedback representation for the open-loop Stackelberg equilibrium point. The paper ends with an example of a Principal-Agent problem involving moral hazard, where the profit process \(\mathcal{P}(r)\) is the solution to: \(d\mathcal{P} (r)=[\alpha (r)\mathcal{P}(r)+\beta _{1}(r)c(r)+\beta _{2}(r)a(r)]dr+[\lambda (r)\mathcal{P}(r)+\vartheta (r)c(r)+\varrho (r)a(r)]dW(r)+\int_{\mathbb{Z}}\mu (\mathfrak{z},r)\mathcal{P}(r)\widetilde{N }(d\mathfrak{z},dr)\), \(x(0)=0\), \(z(r)=[\frac{\rho }{2}Var[\mathcal{P} (r)]^{1/2}\), where \(z(\cdot )\) represents the standard deviation of the investment risk-return, \(\rho >0\) a constant weight, \(c(\cdot )\) the Principal-Agent contract designed by P for A, \(a(\cdot )\) A's investment strategy and behavioral strategy, \(\alpha (\cdot )\) the expected return of the portfolio, \(\beta _{1}(\cdot )\) P's selection parameter, \(\beta _{2}(\cdot )\) A's selection parameter, \(\lambda (\cdot )\) the portfolio volatility, \(\vartheta (\cdot )\) the contract's responsiveness to random changes over time, and \(\varrho (\cdot )\) A's strategy parameter. The authors here establish the corresponding optimal contract and optimal strategy.