Subjective intervention in formal models

Success in forecasting using mathematical/statistical models requires that the models be open to intervention by the user. In practice, a model is only one component of a forecasting system, which also includes the users/forecasters as intergral components. Interaction between the user and the model is necessary to adequately cater for events and changes that go beyond the existing form of the model. In this paper the authors consider Bayesian forecasting models open to interventions, of essentially any form, to incorporate subjective information made available to the user. They discuss principles of intervention and derive theoretical results that provide the means to formally incorporate feedforward interventions into Bayesian models. Two example time series are considered to illustrate why and when such interventions may be necessary to sustain predictive performance.