Little is known about the economic mechanisms leading to the high level of clustering in behavior commonly observed in the data. We present a model where agents can interact according to three distinct mechanisms, and we derive testable implications which allow us to distinguish between the proposed mechanisms. In our application we study students’ performance and we find that a mutual insurance mechanism is consistent with the data. Such a result bears important policy implications for all those situations in which social interactions are important, from teamwork to class formation in education and co-authorship in academic research.