In this paper I investigate mutual insurance arrangements restricted on a social network. My approach solves for Pareto-optimal sharing rules in a situation where exchanges are limited within a given social network. I provide a formal description of the sharing rule between any pair of linked households as a function of their network position. I test the theory on a unique data set of indigenous villages in the Bolivian Amazon, during the years 2004 to 2009. I find that the observed exchanges across families match the network-based sharing rule, and that the theory can account for the deviation from full insurance observed in the data. I argue that this framework provides a reinterpretation of the standard risk sharing results, predicting household heterogeneity in response to income shocks. I show that this network-based variation in consumption behavior is borne out in the data, and that it can be interpreted economically in terms of consumption volatility.