Article dans une revue: We quantitatively evaluate a model of insurance with limited commitment where the requirement that contracts be immune to deviations by subcoalitions makes group size endogenous, as proposed by Genicot and Ray. We compare the model’s predictions to panel data from rural Indian villages. Apart from predicting a realistic degree of insurance, the model captures the evidence along two new dimensions: First, the largest coalition-proof groups are substantially smaller than typical villages. Second, with strong insurance in small groups, individual consumption responds symmetrically to income rises and falls, while alternative models predict strong counterfactual asymmetry.