Estimating Systematic Risk in Real-World Networks.

Social, technical and business connections can all give rise to security risks. These risks can be substantial when individual compromises occur in combinations, and difficult to predict when some connections are not easily observed. A significant and relevant challenge is to predict these risks using only locally-derivable information. We illustrate by example that this challenge can be met if some general topological features of the connection network are known. By simulating an attack propagation on two large real-world networks, we identify structural regularities in the resulting loss distributions, from which we can relate various measures of a network's risks to its topology. While deriving these formulae requires knowing or approximating the connective structure of the network, applying them requires only locally-derivable information. On the theoretical side, we show that our risk-estimating methodology gives good approximations on randomly-generated scale-free networks with parameters approximating those in our study. Since many real-world networks are formed through preferential attachment mechanisms that yield similar scale-free topologies, we expect this methodology to have a wider range of applications to risk management whenever a large number of connections is involved.