Abstract

This paper investigates informal risk-sharing on social networks in the context of a laboratory experiment, featuring a repeated interaction among subjects receiving random endowment realizations. We vary both the network structure, and the information environment. Under Restricted Information subjects only observe their neighbors’ realized endowments, while under Updated Information they also observe in real time how their neighbors’ budgets change as they send and receive transfers to others. We find that there is significant but imperfect risk-sharing in all treatment conditions. Consistent with theoretical predictions risk-sharing is most efficient on complete networks, and in general denser networks facilitate more risk-sharing. Fixing the network, Updated Information facilitates more risk-sharing than Restricted Information, and the information environment matters more on denser networks. Risk-sharing is less efficient the more it requires indirect transfers through intermediators, partly because central agents on the network keep a fraction of the transfers sent through them, and foreseeing this, periphery agents send less transfers. This mistrust in central agents is also confirmed in a setting of center-connected stars, even though the centers of subcommunities are the only ones who can receive transactions from outside the subcommunity.