Abstract:This paper studies linear-quadratic-Gaussian (LQG) games of stochastic large population system with jump-diffusion processes. The most distinguishing feature, compared with the well-studied mean-field LQG games, is that the game system follows linear SDEs driven by random jumps. The individual agents of large population system are coupled both in their state dynamics and in their individual cost functionals. Each agent in large population system has negligible impact on others, but their collective behaviors will impose some significant impact on all agents. The so-called NCE methodology is introduced to deal with the dimensionality difficulty. This methodology derives a set of decentralized strategies, which is an epsilon-Nash equilibrium for a finite N population system where . A numerical example is provided to illustrate the consistency of the mean field estimation and the impact of the population's collective behaviors. As applications, a pricing problem is studied and the decentralized suboptimal price strategy is obtained.

Ε-Nash Mean-Field Games for Linear-Quadratic Systems with Random Jumps and Applications.