The use of Bayes and causal modelling in decision making, uncertainty and risk

The most sophisticated commonly used methods of risk assessment (used especially in the financial sector) involve building statistical models from historical data. Yet such approaches are inadequate when risks are rare or novel because there is insufficient relevant data. Less sophisticated commonly used methods of risk assessment, such as risk registers, make better use of expert judgement but fail to provide adequate quantification of risk. Neither the data-driven nor the risk register approaches are able to model dependencies between different risk factors. Causal probabilistic models (called Bayesian networks) that are based on Bayesian inference provide a potential solution to all of these problems. Such models can capture the complex interdependencies between risk factors and can effectively combine data with expert judgement. The resulting models provide rigorous risk quantification as well as genuine decision support for risk management.