Game analytics became a very popular and strategic tool for business intelligence in the game industry. One of the many aspects of game analytics is predictive analytics, which generates predictive models using statistics derived from game sessions to predict future events. The generation of predictive models is of great interest in the context of game analytics with numerous applications for games, such as predicting player behavior, the sequence of future events, and win probabilities. Recently, a novel approach emerged for capturing and storing data from game sessions using provenance, which encodes cause and effect relationships together with the telemetry data. In this work, we propose a stochastic approach for game analytics based on that novel game provenance information. This approach unifies all gathered provenance data from different game sessions to create a probabilistic graph that determines the sequence of possible events using the commonly known stochastic model of Markov chains and also allows for understanding the reasons to reach a specific state. We integrated our solution with an existing open-source provenance visualization tool and provided a case study using real data for validation. We could observe that it is possible to create probabilistic models using provenance graphs for short and long predictions and to understand how to reach a specific state.