Probabilistic forecasts from probabilistic models: A case study in the oil market

Probabilistic forecasts, probabilistic models, and contingent policy recommendations are inextricably intertwined. This article describes a case study in the use of inherently probabilistic belief network models to produce probabilistic forecasts of average annual oil prices. Belief networks are flexible enough to capture both standard, data-driven economic variables, and quantified expert judgements about the politics of the oil market (particularly the production and capacity policies of key OPEC members). These variables are interrelated by a combination of algebraic formulas, conditional probabilities, and econometric relations. The resultant network is used to test the impact of a variety of different scenarios. The probabilistic forecasts generated by running Monte Carlo analyses on these scenario networks provide corporate decision-makers with useful insights and recommendations.