A Leader-Follower Stochastic Linear Quadratic Differential Game (Q4785671)

A linear stochastic differential game of the ``leader-follower'' type with so-called open-loop strategies driven by a standard (1D) Wiener process is considered. This is expressed by a linear finite-dimensional SDE with two controls. This kind of game means a simultaneous minimization of two cost functions using the two controls: informally speaking, the first player (a follower) minimizes their quadratic cost function for any given second strategy, the first control being a functional of the second strategy, possibly anticipating; once the first strategy has been chosen, the second player (a leader) is to minimize also their quadratic cost function. Explicitly, each of the cost functions depends only on their corresponding control. Coefficients of the dynamics can be random, and the system has a controlled diffusion part; negative weighting matrices in the cost functions may be allowed. It is shown that certain cases can be solved `conditionally'. To this end, a system of (nonlinear) forward-backward SDEs is introduced, and a solvability of the original problem is reduced to the solvability for this FBSDE system in the class of adapted solutions. State-feedback representations for both strategies are given. While the general solvability of the problem remains open, two examples are provided. For a simpler case with non-random (constant) coefficients of the dynamics, the existence of an open-loop solution with a state-feedback representation of controls is established under certain non-degeneracy type assumptions.