In this paper, a bi-level energy management framework based on Conditional Value at Risk (CVaR) and game theory is presented in the context of different ownership of multiple microgrid systems (MMGS) and microgrid aggregators (MAs). The energy interaction between MMGS and MAs can be regarded as a master−slave game, where microgrid aggregators as the leaders set the differentiated tariff for each MG to maximize its benefits, and MMGS as the follower responds to the tariff decision specified by the leader through peer-to-peer (P2P) energy sharing. The P2P energy sharing of MMGS can be regarded as a co-operative game, employing asymmetric Nash bargaining theory to allocate the co-operative surplus. The Conditional Value at Risk model was used to characterize the expected losses by microgrid aggregators due to the uncertainties of renewable energy resources. The Karush−Kuhn−Tucker conditions, Big-M method, and strong duality theory were employed to transform the bi-level nonlinear model of energy management into a single-level mixed integer linear programming model. The simulation results show that when MGs adopt the P2P energy-sharing operation mode, the total operating cost of MMGS can be reduced by 7.82%. The simulation results show that the proposed co-operative optimization framework can make the multiple microgrid systems obtain extra benefits and improve the risk resistance of microgrid aggregators.