Targetting interventions in networks

Individuals interact strategically with their network neighbors. A planner can shape incentives in pursuit of an aggregate goal, such as maximizing welfare or minimizing volatility. We analyze a variety of targeting problems by identifying how a given profile of incentive changes is amplified or attenuated by the strategic spillovers in the network. The optimal policies are simplest when the budget for intervention is large. If actions are strategic complements, the optimal intervention changes all agents' incentives in the same direction and does so in proportion to their eigenvector centralities. In games of strategic substitutes, the optimal intervention is very different: it moves neighbors' incentives in opposite directions, dividing local communities into positively and negatively targeted agents, with few links across these two categories. To derive these results and characterize optimal interventions more generally, we introduce a method of decomposing any potential intervention into principal components determined by the network. A particular ordering of principal components describes the planner's priorities across a range of network intervention problems.