Distances between dynamical models for clustering time series

Abstract In this paper we consider the clustering of time series arising from the class of scalar linear stochastic models. The properties and performance of several so-called model-free and model-based distances for these time series are compared on both artificial and real data sets. In particular, the inappropriateness of model-free distances to distinguish between time series of this class is shown, as well as several important differences between the model-based distances themselves.