The aim of this paper is to investigate how the capacity of an economic system to absorb shocks depends on the specific pattern of interconnections established among financial firms. The key trade-off at work is between the risk-sharing gains enjoyed by firms when they become more interconnected and the large-scale costs resulting from an increased risk exposure. We focus on two dimensions of the network structure: the size of the (disjoint) components into which the network is divided, and the "relative density" of connections within each component. We find that when the distribution of the shocks displays "fat" tails extreme segmentation is optimal, while minimal segmentation and high density are optimal when the distribution exhibits "thin" tails. For other, less regular distributions intermediate degrees of segmentation and sparser connections are also optimal. We also find that there is typically a conflict between efficiency and pairwise stability, due to a "size externality" that is not internalized by firms who belong to components that have reached an individually optimal size. Finally, optimality requires perfect assortativity for firms in a component.
