On Efficient Information Aggregation Networks

This paper considers a population of agents that are connected through a network that allows them to aggregate locally their pieces of private information about some uncertain (exogenous) parameter of interest. The agents wish to match their actions to the true value of the parameter and to the actions of the other agents. I ask how the design of (interim) efficient (minimally connected) networks depends on the level of complementarity in the agents’ actions. When the level of complementarity is either low or high, efficient networks are characterized by a high number of different neighborhoods and, as a consequence, by low levels of connectivity. For intermediate levels of complementarity in actions, efficient networks tend to feature low numbers of highly connected neighborhoods. The implications of this paper are relevant in security environments where agents are naturally interpreted as analysts who try to forecast the value of a parameter that describes a potential threat to security.