Online services such as social networks or grid computing have gained much attention in recent years. The participants of these services are autonomous and decide on their own how they behave in interactions with other participants. Therefore the success of an interaction depends on the cooperativeness of the participants. In this work we develop a new model which allows the analysis of the strategic behavior of the participants. Each participant formulates so called trust policies, which specify whether he wants to interact with other participants. This approach allows for two research directions: (1) Questions from an economic point of view: What trust policies do participants formulate? How successful are they in their interactions? We designed and implemented behavioral experiments to answer these questions. Further experiments are currently planned. (2) The evaluation of trust policies is based on the processing of data on the past behavior of the participants. We investigate approaches for the efficient evaluation of trust policies for large systems. The focus is on optimizing the calculation of centrality measures, i.e., graph-based methods for determining the reputation of participants.