A minimax framework for choosing an optimal intervention in a network game when the network is only partially known: the planner's intervention is pitted against an adversarial 'Nature' that reconfigures the network within an uncertainty set. Convex duality yields the planner's unique robust intervention and shows the worst-case network is rank-1, concentrating all risk along the intervention direction; the framework also measures the cost of robustness under global versus local uncertainty and the role of higher-order uncertainty.
In network games a planner intervenes by adjusting agents' incentives to steer aggregate behavior, but the network is usually only partially known. This paper formulates robust intervention as a zero…