672 / 2024-05-07 10:07:12

Discounted Non-zero-sum Stochastic Game Between Risk-averse Players: Existence of Equilibrium and Algorithm

stochastic game,Markov chain,risk,deep learning

This paper presents a model for n-risk-averse-player finite state/action stochastic games. We consider non-zero sum discounted stochastic games with uncertainty from both supply and demand ends, where players have a risk-averse attitude towards this uncertainty. The measure of risk is conditional value-at-risk (CVaR). We prove the existence of equilibrium without complexity assumptions and propose a mathematical programming algorithm to calculate an equilibrium point. The algorithm is compared with traditional best response algorithm over multi-dimension performance.