In this paper, we apply sensitivity analysis techniques as defined in Saltelli et al. (1993) to study the effects of various sources of uncertainty when a model-based signal extraction decomposition of time series is performed. The paper illustrates the benefit that can be obtained from considering sensitivity analysis tools in time series modelling. These are mainly the evaluation of the accuracy of the series structure definition as defined by a model. It complements thus other tools for model selection, like Bayesian Information Criterion, and provides an information on the parsimony of the model. Finally, it allows to evaluate the effects of model uncertainty.
展开▼
