Abstract: Virtual Power Plants (VPPs) possess flexible and rapid regulation capabilities, making them a valuable resource for frequency regulation, particularly in the context of declining thermal power capacity and scarce frequency regulation resources. However, challenges such as uncertain renewable energy output and strategic behavior among multiple agents not only discourage VPPs from participating in frequency regulation but also pose significant risks to the power system. To address these issues, an optimal trading strategy based on a two-stage stochastic game is proposed to enable VPPs to participate in joint energy and frequency regulation market. First, the non-cooperative game theory is introduced to analyze the relationship between price fluctuations and competitive interactions among multiple VPPs, and a two-stage stochastic game model is established. To mitigate the security risks caused by overbidding, the conditional value-at-risk (CVaR) is incorporated into the model. For the solution methodology, the original problem is simplified by decoupling multiple VPPs, and the sample average approximation (SAA) combined with scenario reduction is applied to approximate the complex stochastic scenarios. Moreover, a distributed best-response algorithm is designed to obtain the optimal strategy. Case studies demonstrate that the proposed method not only enhances the profitability of VPPs but also significantly reduces the risk of overbidding, thereby contributing to the stability of power systems.