GDRR Working Group VII: Adversarial Risk Analysis

Group Leaders:
David Banks (Duke University)
David Rios Insua (ICMAT-CSIC and Royal Academy of Sciences)


As a common framework to deal with these type of problems arising under both focus areas, adversarial risk analysis (ARA) is a natural approach. The ARA treats these games as decision analysis problems, but uses game theoretic reasoning only to estimate the probability that the opponent will select various actions. This framework mitigates strong implausible common knowledge assumptions criticized by many. Generally speaking, ARA views two-person games through coupled influence diagrams (or decision trees), one for each opponent, often with some shared nodes. Instead of finding a joint equilibrium solution, ARA supports one of the opponents, against the other and employs a subjective expected utility model by treating the other opponent’s decisions as random actions. The critical ingredient in ARA, which distinguishes it from the conventional use of probabilistic risk analysis is that an explicit model is built for the strategic decision-making of the opponent. Effective use of ARA in general settings, such as the two focus areas above, requires new methodological and computational developments. These include (a) a general ARA framework that allows for multiple participants with interactions over time; (b) methods to predict actions of adversaries using robustness concepts and advanced structured expert judgement and (c) efficient algorithms for ARA computations.

Questions: email

Program on Games, Decisions, Risk and Reliability (GDRR)